Transforming Personalized Travel Recommendations: Integrating Generative AI with Personality Models

Abstract

Over the past few years, the incorporation of generative Artificial Intelligence (AI) techniques, particularly the Retrieval-Augmented Generator (RAG) framework, has opened up revolutionary opportunities for improving personalized travel recommendation systems. The RAG framework seamlessly combines the capabilities of large-scale language models with retriever models, facilitating the generation of diverse and contextually relevant recommendations tailored to individual preferences and interests, all of which are based on natural language queries. These systems iteratively learn and adapt to user feedback, thereby continuously refining and improving recommendation quality over time. This dynamic learning process enables the system to dynamically adjust to changes in user preferences, emerging travel trends, and contextual factors, ensuring that the recommendations remain pertinent and personalized. Furthermore, we explore the incorporation of personality models like the Myers–Briggs Type Indicator (MBTI) and the Big Five (BF) personality traits into personalized travel recommendation systems. By incorporating these personality models, our research aims to enrich the understanding of user preferences and behavior, allowing for even more precise and tailored recommendations. We explore the potential synergies between personality psychology and advanced AI techniques, specifically the RAG framework with a personality model, in revolutionizing personalized travel recommendations. Additionally, we conduct an in-depth examination of the underlying principles, methodologies, and technical intricacies of these advanced AI techniques, emphasizing their ability to understand natural language queries, retrieve relevant information from vast knowledge bases, and generate contextually rich recommendations tailored to individual personalities. In our personalized travel recommendation system model, results are achieved such as user satisfaction (78%), system accuracy (82%), and the performance rate based on user personality traits (85% for extraversion and 75% for introversion).

1. Introduction

Recently, the swift progress in technology has revolutionized how we plan, experience, and savor travel. With the proliferation of online travel platforms, social media, and mobile applications, travelers are presented with an overwhelming array of options for destinations, accommodations, activities, and experiences [1]. However, amidst this abundance of choices, finding a perfect travel itinerary tailored to one’s preferences, interests, and needs can be a daunting task [2]. In recent years, the realm of travel planning has experienced a significant shift, driven by the merging of cutting-edge technologies and changing consumer expectations. In the digital age, travelers are inundated with an overwhelming abundance of information and options when selecting destinations, accommodations, activities, and experiences for their trips. Personalized travel recommendation systems have become essential tools for helping travelers navigate the intricacies of trip planning while crafting tailored travel experiences that align with their individual preferences and interests [3].

In response to this challenge, personalized travel recommendation systems have emerged as a valuable solution for assisting travelers in discovering and planning their ideal trips. These systems utilize state-of-the-art technologies like data mining, machine learning, and artificial intelligence to process extensive travel-related data, encompassing user preferences, past behavior, social interactions, and contextual factors. By harnessing the power of data-driven algorithms, personalized travel recommendation systems aim to deliver tailored travel suggestions that align closely with the individual preferences and requirements of each traveler [4]. Conventional personalized travel recommendation systems depend on a range of techniques, including collaborative filtering, content-based filtering, and matrix factorization, to assess user preferences and historical data, thereby producing tailored travel suggestions. However, these approaches often face limitations in capturing the nuanced and multifaceted nature of travelers’ preferences, leading to suboptimal recommendations and user experiences [5].

The fusion of generative Artificial Intelligence (AI) techniques, like the RAG framework, with personality models has unlocked promising new opportunities for boosting the capabilities and effectiveness of personalized travel recommendation systems. The RAG framework merges the strengths of large-scale language models with retrospective models, facilitating the creation of varied and contextually appropriate recommendations in response to natural language queries [6,7,8]. By leveraging personality models, personalized travel recommendation systems can iteratively learn and adapt to user feedback, continuously refining and improving the quality of recommendations over time [9]. This dynamic learning process allows the system’s ability to adjust to shifts in user preferences, emerging travel trends, and contextual factors, ensuring that the recommendations remain relevant and personalized [10]. In this study, we examine the potential of generative AI methods, specifically the RAG framework with a personality model, for revolutionizing personalized travel recommendation systems. We examine the underlying principles, methodologies, and technical intricacies of these advanced AI techniques, highlighting their ability to understand natural language queries, retrieve relevant information from vast knowledge bases, and generate contextually rich recommendations.

Furthermore, we discuss the implications of integrating generative AI methods with personalized travel recommendation systems, including the potential benefits of enhancing recommendation diversity, addressing data sparsity issues, and improving user engagement and satisfaction. Also, we analyze real-world applications and case studies that demonstrate the effectiveness and influence of these advanced AI techniques in revolutionizing the travel planning experience for users. Through our investigation, we seek to offer insights into the changing landscape of personalized travel recommendation systems and the transformative potential of generative AI techniques in influencing the future of travel planning and tourism. By leveraging AI-driven personalization, we foresee a future where travelers can seamlessly uncover and craft enriching travel experiences that align with their unique preferences and aspirations.

2. Related Works/Literature Review

Travel recommendation systems have become increasingly prevalent in recent years, driven by the growing availability of user-generated data and advancements in data mining and machine learning techniques. These systems aim to assist travelers in discovering personalized travel experiences based on their preferences, past behavior, and contextual factors.

In this literature review, we examine significant research contributions within the field of travel recommendation systems, with an emphasis on the different approaches and methodologies used to improve the recommendation process. Building upon this foundation, researchers have proposed various innovative approaches to enhance personalized travel recommendations. Nitu et al. [11] proposed an approach to improving tailored travel recommendations by integrating recency effects. Their study highlights the importance of considering the temporal aspects of user preferences to deliver timely and appropriate recommendations. Memon et al. [12] introduced a travel recommendation method leveraging geo-tagged photos from social media platforms. By analyzing the spatial distribution of user-generated content, their approach offers location-based recommendations tailored to tourists’ interests. Liu et al. presented a personalized travel package recommendation system that utilizes data mining techniques to extract user preferences from historical travel records [13]. Their research emphasizes the significance of understanding individual travel behavior to offer tailored itinerary suggestions. Chen et al. introduced a travel recommendation framework that derives people attributes and travel group types from community-shared photos [14]. By analyzing social context and group dynamics, their approach enhances the relevance and diversity of travel suggestions. Sun et al. [15] introduced a route-based travel recommendation system that utilizes geographically tagged images to infer travel preferences and routes. Their research highlights the potential of visual content analysis for generating personalized travel itineraries.

Moreover, as travel recommendation systems evolve, the integration of context awareness emerges as a pivotal theme. Zheng and Xie explored the use of user-generated GPS traces for learning travel recommendations [16]. By mobility patterns and historical trajectories, their approach provides personalized suggestions based on individuals’ movement behaviors. Ricci et al. suggested a case-driven travel recommendation system that suggests destinations derived from similarities between user preferences and historical cases [17]. Their research emphasizes the importance of leveraging past experiences to guide future travel decisions. Majid et al. [18] introduced a context-sensitive personalized travel recommendation system that utilizes geographically tagged social media data mining. By considering contextual factors such as time, location, and social interactions, their approach offers tailored recommendations aligned with travelers’ current situations. Cheng et al. developed a personalized travelling recommendation method derived from extracting people attributes from user-contributed photos [19]. Their research emphasizes the influence of social context and user-created content in enhancing travel suggestions. Zheng et al. suggested a context-sensitive travel recommendation method that relaxes contextual constraints to improve recommendation accuracy [20]. By dynamically adjusting the relevance of contextual factors, their method adapts to users’ evolving preferences and situations.

Looking ahead, the future of travel recommendation systems holds promise for autonomous decision-making with the integration of generative AI technologies like ChatGPT [21]. Their study explored the role of conversational agents in assisting travelers with real-time recommendations and decision support. Chen et al. [22] introduced a trip reinforcement recommendation approach based on graph-based representation learning. By modeling the relationships between travel entities, their method enhances recommendation effectiveness and diversity. In summary, these studies emphasize the wide array of methodologies and strategies utilized in travel recommendation systems, spanning from content-based and collaborative filtering strategies to context-sensitive and social media-driven approaches. Future research directions may focus on integrating multiple data sources, leveraging advanced AI techniques, and addressing emerging challenges such as privacy concerns and data sparsity issues to further enhance the efficiency and user-friendliness of travel recommendation systems.

3. Background

Recommendation systems have evolved from early collaborative filtering methods, such as user–item matrix factorization, to more sophisticated techniques incorporating machine learning algorithms and deep learning models. Collaborative filtering approaches from Resnick and Varian rely on user feedback and preferences to generate recommendations, but they face challenges including cold-start issues and data sparsity [23]. As computational power increased and data availability improved, content-based filtering methods from Pazzani and Billsus emerged, utilizing item attributes and user profiles to make personalized recommendations [24]. These methods are effective but limited at capturing complex user preferences and contextual information.

The advent of AI-driven recommendation systems marked a significant paradigm shift, leveraging advanced machine learning algorithms, natural language processing (NLP) methods, and deep learning frameworks [25,26]. Adomavicius and Tuzhilin employed these methods in combination with collaborative filtering, content-driven filtering, and combined approaches to produce highly personalized recommendations rooted in user behavior, historical data, and contextual signals [27,28,29,30]. For example, He et al. integrated neural collaborative filtering models for deep learning architectures with collaborative filtering techniques to improve recommendation accuracy and scalability [31].

3.1. Evolution of Recommendation Systems and AI-Driven Conversational Interfaces

Concurrently, AI-driven conversational interfaces have evolved from rule-based chatbots to intelligent virtual assistants capable of natural language understanding and generation. Turkle showed that early chatbots relied on predefined rules and scripted responses, limiting their conversational capabilities and adaptability to user inputs [32]. However, Brown et al. led advancements in NLP, machine learning, and deep learning to develop sophisticated conversational agents powered by large-scale language models such as ChatGPT [33]. These agents can engage in human-like conversations, understand complex queries, and provide contextually relevant responses [34].

The integration of recommendation systems with AI-driven conversational interfaces represents a synergistic approach to enhancing user experiences and engagement. By leveraging recommendation algorithms within conversational interfaces, users can receive personalized recommendations seamlessly during conversations, enriching their interactions and facilitating decision-making processes, which were described by Adomavicius and Tuzhilin [27]. For instance, an e-commerce chatbot can suggest products based on user preferences and context, leading to higher conversion rates and user satisfaction.

Overall, the evolution of recommendation systems and AI-driven conversational interfaces has been propelled by advancements in AI, machine learning, and natural language processing technologies. These advancements continue to drive innovations in personalized user experiences, intelligent decision support systems, and human–computer interactions, shaping the future of AI-driven applications in various domains according to Brown et al. [33].

Generative AI methods encompass a diverse set of techniques aimed at generating new data, content, or outputs based on learned patterns and models. The proposed recommendation system leverages the RAG framework to retrieve contextually relevant information based on user queries. The RAG framework is particularly effective in identifying and filtering high-quality content from extensive databases, thereby improving the accuracy of initial recommendations. To further refine these recommendations, personality models—specifically the MBTI and BF—are integrated within the system.

Personality traits guide the personalization process, enabling the system to align recommendations with individual user preferences more precisely. For instance, users with high Openness may receive suggestions for novel and adventurous travel experiences, while those with high Conscientiousness may receive structured and detail-oriented recommendations. By combining the RAG framework’s robust data retrieval capabilities with personality-driven personalization, the system provides tailored travel recommendations that are both accurate and highly relevant to user needs [6]. This framework has been instrumental in enhancing the accuracy and relevance of AI-generated content, particularly in applications requiring contextual understanding and information synthesis.

3.2. Personality Psychology: Understanding Individual Differences

Personality psychology is a field within psychology focused on studying the variations in personality traits among individuals and understanding the psychological mechanisms that contribute to people’s unique characteristics. Personality traits are stable patterns of thoughts, emotions, and behaviors that differentiate individuals from one another and remain relatively consistent over time [35]. These traits encompass a wide range of dimensions, including extraversion, agreeableness, conscientiousness, neuroticism, and openness to experience. The areas it explores include the following:

• Constructing a coherent personality-based understanding of an individual’s core psychological processes;

• Analyzing psychological variations in relation to personality;

• Exploring human nature and the psychological commonalities among individuals.

Personality psychology is defined as the study of how people differ due to psychological influences and how personality traits function collectively [36]. This field aims to build a comprehensive understanding of an individual’s basic psychological processes, examine psychological differences among individuals, and explore human nature and the psychological commonalities between people [35,36]. The concept of ‘personality’ refers to a dynamic and organized set of traits that distinctly shape an individual’s interactions with their environment, cognition, emotions, motivations, and behaviors. Personality also relates to the patterns of thoughts, feelings, social adjustment, and behavior that are consistently displayed over time and significantly influence one’s expectations, self-perceptions, values, and attitudes [37].

3.3. Personality Classification Systems: MBTI and BF

Personality research has been extensively explored in psychology, covering a variety of theoretical traditions that include several major theories, such as the dispositional (trait) perspective, as well as psychodynamic, humanistic, biological, behavioral, evolutionary, and social learning perspectives [34]. Many researchers and psychologists do not confine themselves to a single perspective, preferring instead to adopt an eclectic approach [36]. Research in this field is often empirically driven, employing dimensional models based on multivariate statistics like factor analysis, or it emphasizes theory development, as illustrated in psychodynamic theory [35]. Additionally, there is significant focus on applied personality testing. In psychological education and training, understanding the nature and development of personality is frequently considered a prerequisite for courses in abnormal or clinical psychology [34]. Personality type refers to the psychological classification of individuals into distinct categories. Personality types are different from personality traits, which vary in degree. Numerous personality theories exist, each containing several sub-theories [38]. The development of the MBTI and the BF personality traits is particularly significant in personality psychology [39]. The Myers–Briggs Type Indicator (MBTI) categorizes individuals into distinct psychological types [40]. For instance, personality theories often classify people as either introverts or extroverts, whereas trait theories view introversion and extroversion as part of a continuous spectrum, with many individuals situated at a midpoint.

The MBTI is a popular personality classification system rooted in Carl Jung’s theories of psychological types. The MBTI divides individuals into 16 distinct personality types, each defined by preferences across four dichotomous dimensions: extraversion versus introversion, sensing versus intuition, thinking versus feeling, and judging versus perceiving, as outlined by [41] This classification system provides insights into individuals’ cognitive preferences, decision-making styles, and communication preferences, making it valuable in various personal and professional contexts.

The BF personality traits, also referred to as the five-factor model (FFM), offer a broad framework for understanding personality [42]. The BF traits mentioned by Goldberg include openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism, each representing a broad dimension of personality variation [42,43]. This model is based on empirical research and is widely accepted in personality psychology for its robustness and predictive validity across different cultures and contexts.

Personality classification systems such as the MBTI and the BF provide valuable frameworks for assessing and categorizing individuals based on their personality traits [40]. These systems have applications in various fields, including career counseling, team building, interpersonal relationships, and organizational development. By understanding individuals’ personality profiles, practitioners can tailor interventions, communication strategies, and decision-making processes to align with individuals’ preferences and tendencies.

Based on Jung’s writings and observations, many tests, such as the Golden, PTI-Pro and JTI [40,41,42], were subsequently developed based on the Myers–Briggs model. These theories can also be viewed as ‘approaches’ to personality or psychology and are frequently referred to as models. The model represents an older theoretical approach that views introversion and extraversion as fundamental psychological orientations, connected to two pairs of psychological functions, type dynamics and development, and the cognitive functions associated with each type. The MBTI categorizes certain psychological differences into four opposing pairs or ‘binaries’, leading to 16 distinct psychological types. None of these types are seen as ‘better’ or ‘worse’, but Briggs and Myers proposed that individuals naturally ‘prefer’ a particular combination of type differences [41]. Just as it is difficult for a left-handed person to use their right hand, people tend to find it more difficult to use their contrasting psychological preferences, but with practice and development, they can enhance their proficiency (and thus become more behaviorally adaptable).

The 16 types are typically denoted by a four-letter acronym, representing the initials of each of the four preference types (with the exception of intuition, which is abbreviated as ‘N’ to avoid confusion with introversion). This can be illustrated with the following examples: ENTJ stands for extraversion (E), intuition (N), thinking (T), and judging (J); ISFP stands for introversion (I), sensing (S), feeling (F), and perceiving (P). These abbreviations are consistent across all 16 types, as shown in Table 1.

The BF personality traits represent a proposed taxonomy or grouping of personality characteristics, developed within psychological trait theory starting in the 1980s. By the 1990s, five distinct factors were identified for the U.S.-British population, typically labeled as follows [41]:

• -. openness to experience (creative/curious vs. consistent/cautious)

• -. conscientiousness (efficient/organized vs. wasteful/careless)

• -. extroversion (outgoing/energetic vs. lonely/withdrawn)

• -. compatibility (friendly/compassionate vs. critical/rational)

• -. neuroticism (sensitive/nervous vs. flexible/confident)

Factor analysis reveals semantic relationships by analyzing personality questionnaire data. In this sense, there is often a relationship between words used to refer to the same person [40]. For example, for a person characterized as conscientious, the term “always prepared” is more commonly used than “disorganized”. These connections highlight five broad dimensions that serve as a common language for describing human personality, temperament, and mood [41,42,43]. The five factors are easily remembered using the acronyms ‘OCEAN’ or ‘CANO’. Each global factor encompasses a set of interrelated, more specific primary traits. For instance, extraversion is generally linked to traits like sociability, assertiveness, thrill-seeking, warmth, activity, and positive emotions. These traits are not absolute but exist along a continuum. This understanding of personality can now be applied to the development of language models, as the next section explores.

3.4. Introduction to Language Models: Understanding Natural Language Queries and Generating Responses

This study seeks to close the gap between personality psychology and language models. By creating a mapping between the MBTI system and the BF traits, we hope to equip language models with a deeper understanding of human personality. This, in turn, could allow these models to tailor their responses to better reflect the user’s individual characteristics.

So, first we create a mapping from the MBTI to the BF and implement protectionist interfaces in the language model. We then extend this methodology to a time-sensitive design where conducting a fixed effect analysis is straightforward [40]. When several individuals are observed across multiple time points, these time points are accumulated over ten-year intervals to obtain a discriminative observation result in the popular use of LLMs such as ChatGPT [6,7].

Introduction to language models such as the ChatGPT encompasses a sophisticated understanding of natural language processing (NLP) techniques and the pivotal role these models play in comprehending and responding to natural language queries.

ChatGPT is an advanced language model developed by OpenAI, built on the generative pretrained transformer (GPT) architecture. It utilizes deep learning techniques and large-scale transformer models to capture complex patterns and structures in natural language data, as demonstrated by Radford et al. [44]. ChatGPT’s architecture enables it to create coherent and contextually appropriate responses to diverse natural language inputs, making it highly versatile for various NLP tasks.

Language models such as ChatGPT play a crucial role in understanding natural language queries by leveraging advanced NLP capabilities that utilize techniques such as attention mechanisms, transformer architectures, and pre-training on large datasets to capture semantic relationships, syntactic structures, and contextual nuances in language [45]. As a result, they can interpret user queries, extract relevant information, and generate meaningful responses that align with the query’s intent.

The ability of language models such as ChatGPT to understand natural language queries is further enhanced by fine-tuning particular tasks or areas. Fine-tuning involves training the model on task-specific data or providing domain-specific prompts [46], allowing the model to specialize in understanding and generating content relevant to a particular context, as shown by Raffel et al. This fine-tuning process improves the model’s accuracy and relevance in handling natural language queries within specific domains or applications.

In addition to understanding individual queries, language models such as the ChatGPT contribute to building conversational agents and AI-driven interfaces capable of engaging in human-like interactions. Brown et al. investigated how these models facilitate dialog generation, question answering, sentiment analysis, and content recommendation tasks, enhancing the user experience and enabling more natural and effective communication with AI systems [43].

In conclusion, language models such as the ChatGPT model play a critical role in understanding natural language queries by leveraging advanced NLP techniques, fine-tuning specific tasks or domains, and contributing to the development of conversational AI systems.

4. Methods

To evaluate the proposed system, we need to consider framework modeling that includes components such as a travel data retrieval mechanism, personality driven prompt augmentation, and travel recommendation generation with a knowledge base as well as data collection and preprocessing tasks, as shown in Figure 1.

4.1. Travel Data Retrieval Mechanism

The travel data retrieval mechanism is a crucial part of the RAG architecture designed to provide personalized travel recommendations. It encompasses the search process and the travel data document base, which work together to fetch relevant information for the user. When a user submits a query, it is first received by the search component. The search component is responsible for interpreting the query and determining the most relevant data sources to consult as query processing.

The search component sends a vectorized request to the document base related to travel data from the web. The request includes the specific parameters of the user’s query with personalized preferences and other relevant details. The travel data document base includes a web crawler module that processes the request and retrieves documents that match the query parameters. The document base consists of a vast collection of travel-related information sourced from the web (www). It aggregates data from various online sources to ensure comprehensive coverage of travel destinations, budget, travel guides/tips, activities, accommodation options, local attractions, user reviews, travel magazines/articles (travel trends) and other travel-related aspects. The document base is regularly updated to include the latest travel information so that users receive current and relevant recommendations. Documents in the base are categorized and indexed to facilitate efficient retrieval. The retrieved documents from the travel data document base are sent back to the search component. The search component then forwards improved queries including this retrieved information from documents for further processing with personality-driven prompt augmentation logic.

By integrating the travel data retrieval mechanism involving the search logic component and the travel data document base, the RAG architecture effectively retrieves and processes relevant travel information. This mechanism ensures that users receive tailored travel recommendations that cater to their unique preferences and requirements, enhancing their overall travel planning experience.

In this study, we have refined the methodology section to detail our experimental design. This includes clarifying the objectives, participant selection criteria, and the structure of the experimental and control groups. For example, participants were selected based on a diverse range of demographic and psychographic traits to assess the impact of the recommendation system across various personality types. The experiment was structured with a control group using a traditional system and an experimental group utilizing our personality-driven model, allowing us to measure comparative effectiveness.

The data collection process was expanded to provide specific information on the sources, methods, and criteria used. Data were gathered from multiple sources, including travel booking sites, social media, and user-generated reviews. We applied rigorous inclusion criteria to ensure data relevance and used APIs and web scraping tools to automate collection. A detailed data filtering protocol was implemented to exclude outdated or irrelevant data, ensuring high-quality inputs for the recommendation model.

Data validation steps were added to strengthen the robustness of the methodology. We incorporated data accuracy checks through cross-referencing with verified sources and utilized preprocessing steps such as deduplication and normalization. Furthermore, external datasets were used to benchmark and validate the recommendation data, ensuring the reliability and accuracy of the information processed by our system.

We provided a clearer description of the specific algorithms and models used in the study, including the RAG framework and its configuration. This section now explains the integration of personality models based on MBTI and BF traits, detailing how the system tailors recommendations based on individual characteristics. Additionally, the benchmarking process for these models was included, with key performance metrics like accuracy, user satisfaction, and recommendation relevance highlighted to validate the effectiveness of our model.

4.2. Personality Model and Prompt Augmentation

The second of these steps starts with the collection of a diverse dataset of travel-related queries and corresponding recommendations from existing travel recommendation systems, online forums, and social media platforms. Each query was associated with relevant contextual information such as user preferences, location, and travel time. Before query augmentation, we preprocessed the retrieved data to remove noise, standardize formats, and anonymize personal information.

These algorithms can be developed with existing numerical methods in the literature or machine learning-based computations. All of these methods can even be hybrid methods that can be combinations of these methods [37]. It may be quite appropriate to model them as linear or nonlinear systems with numerical methods. In addition, probabilistic (Bayesian or Monte Carlo) or fuzzy logic methods can explain the modeling in detail in terms of uncertainties [37]. ML-powered methods (e.g., Markov processes, support vector machines, Boltzman machines, or artificial neural networks) can also be useful for such studies. Lately, the performance of deep neural networks on large datasets has improved considerably [8]. From time to time, numerical models designed under these needs can be made more flexible and interactive with the use of ML techniques [37].

In this study, encoding personality traits involves a structured process that transforms psychological data into machine-readable feature vectors, ensuring compatibility with the AI recommendation framework. For trait normalization, MBTI Scores are mapped into a binary format representing the four dichotomous dimensions (e.g., Extraversion = 1, Introversion = 0). BF Scores are continuous variables, and they are normalized to a [0, 1] range to standardize them across users. For feature vector construction, MBTI and BF data are combined into a unified feature vector. For example, an ENFJ user with Big Five scores of [0.75, 0.85, 0.65, 0.72, 0.40] will generate a composite vector such as [1, 1, 1, 1, 0.75, 0.85, 0.65, 0.72, 0.40]. For integration in the RAG framework, during the retrieval stage, personality traits act as filters, prioritizing travel documents that align with specific user preferences (e.g., adventurous destinations for high openness scores). Lastly, during the generation stage, these vectors influence the tone, structure, and content of the generated recommendations, tailoring outputs to match user traits (e.g., extraverted users receive socially engaging suggestions). This encoding mechanism ensures that personality traits directly influence both the retrieval and generation stages, enhancing recommendation relevance.

To start the personality transition process, popular methods such as Pearson’s method are used. In statistics, the Pearson correlation coefficient (PCC, Pearson r, or Pearson product–moment correlation coefficient, PPMCC), often referred to simply as the correlation coefficient, measures the linear relationship between two datasets [40]. It is calculated as the ratio of the covariance of two variables to the product of their standard deviations, providing a normalized measure of covariance that always results in a value between −1 and 1 [40]. Like covariance, this measure only captures the linear correlation between variables, disregarding other types of relationships or correlations [40]. For instance, the Pearson correlation coefficient between age and height in a sample of high school teenagers is expected to be significantly greater than 0 but less than 1 (since 1 would indicate an unrealistic perfect correlation). The Pearson correlation coefficient (r) is one of several correlation coefficients available for measuring correlations. Pearson’s correlation coefficient is appropriate when certain conditions are met [40]. Both variables must be quantitative; if one is qualitative, another method should be used. The variables should be normally distributed, which can be checked by creating a histogram for each variable to determine if the distributions are approximately normal [40]. Slight deviations from normality are usually not problematic. There should be no outliers in the data, as outliers are observations that deviate from the overall pattern of the data [40]. A scatterplot can help identify outliers by showing points that are isolated from the rest. The relationship between the variables should be linear, meaning it can be reasonably represented by a straight line. A scatter plot can also be used to assess whether the relationship between the variables is linear.

(1)$r={\displaystyle \frac{n\sum xy -\left(\sum x\right)\left(\sum y\right)}{\sqrt{\left[n\sum x^2-{\left(\sum x\right)}^2\right]\left[n\sum y^2-{\left(\sum y\right)}^2\right]}}}$

One advantage of using Pearson’s correlation coefficient (Equation (1)) to explore the relationship between MBTI and the BF personality traits is that it offers a straightforward and intuitive method for measuring the strength and direction of the relationship between two variables. It is a widely recognized and commonly used statistical tool that is well understood and easily interpreted by researchers. If we interpret, r = 1 indicates a perfect positive linear relationship: as x increases, y increases in a perfectly linear way. R = −1 indicates a perfect negative linear relationship: as x increases, y decreases in a perfectly linear way. r = 0 suggests no linear correlation between x and y. However, this method also has several potential drawbacks. One limitation is that the Pearson correlation coefficient assumes a linear relationship between the two correlated variables, which may not always reflect reality. Additionally, the Pearson correlation coefficient only measures the strength of the relationship between two variables and does not provide information about causality or directionality [40]. Other factors, such as situational or cultural influences, may also affect the relationship between MBTI and BF personality traits [37]. Another potential drawback of using Pearson’s correlation coefficient is its reliance on accurate and reliable measurements of the correlated variables. As with any measurement tool, there may be sources of error or bias in assessing the MBTI and BF personality traits, which could impact the accuracy of the correlation coefficient [37,40].

While the Pearson correlation coefficient can provide insights into the relationship between MBTI scores and BF personality traits, researchers should remain aware of its limitations in capturing complex, non-linear associations. To achieve a more comprehensive analysis, additional methods and techniques should be considered alongside the Pearson correlation.

The Pearson correlation coefficient presents limitations, particularly its assumption of a linear relationship between correlated variables, which may not accurately capture the complexities of non-linear associations. The second possible drawback is the potential lack of sensitivity. These fallbacks show that a secondary method may be used to calculate the personality transition process.

Jaccard similarity serves as a widely used metric for assessing the similarity between two entities, such as two textual documents. This metric proves useful for measuring the likeness between two asymmetrical binary vectors or sets. In the academic literature, Jaccard similarity, denoted by J(A, B), is interchangeably known as the Jaccard Index, Jaccard coefficient, Jaccard dissimilarity, and Jaccard distance.

Jaccard similarity has extensive utility in the realm of data science. Examples illustrating the application of Jaccard similarity include textual analysis, which involves gauging the resemblance between two text documents based on the terms shared between them; e-commerce, where vast customer and product databases are selected to identify similar customers via their buying histories; and recommendation systems, which employ the Jaccard coefficient in movie recommendation algorithms to pinpoint similar customers who have either rented or highly rated similar movies.

The application of the Jaccard similarity coefficient extends to evaluating the similarity of two asymmetric binary variables. It is considered as a binary variable with two possible states, 0 and 1, where 0 signifies the absence of the attribute and 1 indicates its presence. Unlike symmetric binary attributes where both states hold equal values, the importance of the two states differs significantly for asymmetric binary variables. If we consider evaluating the similarity among customers of a store, we may utilize a binary attribute that signifies a purchase made at the store. Here, 1 represents the purchase of a specific item, while 0 indicates no purchase of that item.

Given the potential existence of thousands of products in the store, the quantity of items left unpurchased by any customer far exceeds the number of purchased items. Consequently, when assessing customer similarity, we focus solely on item purchases. This results in an asymmetric binary variable, where a value of 1 holds greater than 0.

In the initial step of computing the Jaccard similarity in Equation (2) between two customers, each characterized by binary attributes, the following four quantities (i.e., frequencies) are determined based on the provided binary data:

a = the total number of attributes that are equal to 1 for both objects i and j

b = the total number of attributes that are equal to 0 for object i but equal to 1 for object j

c = the total number of attributes that are equal to 1 for object i but equal to 0 for object j

d = the total number of attributes that equal 0 for both objects i and j.

Subsequently, the Jaccard similarity for these attributes is calculated using the following equation:

(2)$J\left(i,j\right)=sim\left(i,j\right)={\displaystyle \frac{a}{a+b+c }}$

The count of matches d is disregarded in this calculation because it holds no significance, as the items are asymmetric binary attributes. These calculations indicate that customers exhibit similar shopping patterns while displaying dissimilar behaviors, as they have purchased entirely different items.

Jaccard similarity can also be explained as follows:

a represents the number of elements that are common between sets i and j,

b represents the number of elements that are unique to set i,

c represents the number of elements that are unique to set j.

If we expand this formula, the numerator a represents the count of elements common to both sets i and j. This is the intersection of the two sets, denoted by ∣i ∩ j∣. In set notation, the numerator can be rewritten as: a = ∣i ∩ j∣. The denominator a + b + c represents the union of the two sets i and j. It includes

a: Elements common to both sets (intersection).

b: Elements only in set i.

c: Elements only in set j.

This total represents the total unique elements in both sets combined, written as ∣I ∪ j∣ in set notation. By substituting these terms into the formula, the Jaccard Similarity Index can also be expressed in terms of set operations:

(3)$J\left(i,j\right)={\displaystyle \frac{\left| i \cap j \right|}{\left| i \cup j \right| }}$

This expanded version highlights that the Jaccard Index measures the ratio of the number of elements in the intersection of the two sets to the number of elements in their union. Furthermore, instead of assessing similarity, the dissimilarity or Jaccard distance between two binary attributes can be determined. The dissimilarity based on these attributes via the Jaccard coefficient is derived as follows:

(4)$d\left(i,j\right)={\displaystyle \frac{b+c}{a+b+c }}=1-sim(i,j)$

However, these methods also fail to represent our personality model. Transitioning into the discussion of hybrid recommendation approaches, it is imperative to recognize their pivotal role in addressing the complexities and challenges encountered in traditional similarity-based recommendation systems.

Hybrid recommendation approaches are highly favored due to their ability to combine the strengths of both item–item and user–user similarity techniques, thereby significantly enhancing recommendation precision and scope. The rationale behind the preference for hybrid models lies in their adeptness at mitigating various challenges inherent in conventional similarity-based systems. In particular, these hybrid methodologies excel in addressing the cold-start dilemma, which emerges when dealing with novel items or users possessing limited interaction data. By harnessing item–item similarity for cold-start items, the system can furnish recommendations grounded on the inherent attributes and characteristics of the items themselves, eliminating the necessity for exhaustive user histories. Conversely, the utilization of user–user similarity facilitates the delivery of tailored recommendations by identifying users with similar preferences and behaviors. This personalized approach augments the pertinence and efficacy of recommendations, especially for seasoned users with extensive interaction histories. In essence, the hybrid approach optimizes recommendation precision and scope by astutely amalgamating item–item and user–user similarities, rendering it an adaptable and resilient solution for contemporary recommendation systems.

The algorithms described above for converting MBTI scores into BF scores could benefit from the use of various machine learning methods to optimize their performance. Five machine learning methods that may be suitable for this study and their strengths and weaknesses for this particular algorithm are exemplified and compared. Random forest regression, gradient boosting regression, artificial neural networks, and K-nearest neighbors’ regression were not preferred because they are computationally expensive.

The personality transition step continues with creating and implementing a time-bound structure in Figure 2. To simulate age in personality, we can adjust the scores of each trait according to the age of the individual. Research has shown that personality traits tend to change as individuals age, with some traits increasing or decreasing in intensity over time. In this algorithm, we calculate an adjustment factor for each trait based on the age of the individual. We then added this adjustment factor to the individual’s score for each trait to obtain adjusted scores. The adjustment factors are chosen based on research on how personality traits tend to change over a lifetime. As with the previous algorithm, we can use machine learning methods to optimize this algorithm using our personality model.

4.3. Knowledge Base and Travel Recommendation Generation

At the heart of the RAG framework lies a large language model (LLM), which is typically pre-trained on vast corpora of text data. Examples of such LLMs include OpenAI’s GPT series (e.g., GPT-3), BERT, and RoBERTa.

Technical Description of the RAG Framework and Personality Model Integration: The RAG framework is a hybrid model that combines a retrieval mechanism with a generative language model, creating responses based on both user input and a comprehensive knowledge base. Specifically, the RAG framework first retrieves relevant data from a curated dataset using semantic search, where user queries are mapped to high-quality documents or records that match the contextual needs of the query. The retrieved data are then passed to the generative component, which synthesizes this information into coherent, user-specific recommendations.

Configuration of the RAG Framework: The RAG model is configured with fine-tuning parameters optimized for travel recommendation tasks. Key configuration parameters include embedding size for semantic search, retrieval batch size, and learning rate for fine-tuning the generative model to align with recommendation objectives. We used pre-trained embeddings for the initial semantic search layer and adjusted retrieval thresholds to balance relevance and diversity in the results.

Integration with Personality Models: To personalize recommendations further, we incorporated MBTI and BF personality models. Personality traits are mapped to recommendation attributes based on psychological research linking specific traits to travel behaviors. For example, high Openness may correlate with recommendations for novel and unique destinations, while high Conscientiousness correlates with structured and itinerary-based suggestions. The integration process involves encoding personality traits as feature vectors, which are combined with user queries to refine the retrieval process and tailor generative outputs.

Technical Adaptation of Personality Models: The personality models were adapted for the RAG framework by encoding personality traits as feature inputs in the retrieval stage. During retrieval, personality feature vectors are applied as filters, ensuring that the retrieved documents align not only with the query but also with individual personality-based preferences. In the generative stage, these traits inform language generation, influencing the style and tone of the recommendations. For instance, extraverted users receive recommendations framed with more socially engaging language, whereas introverted users receive suggestions phrased in a reflective and personalized tone.

5. Results and Discussion

Case Study: Elevating Visitor Experiences at Istanbul’s Kapalı Çarşı through Personalized Travel Recommendations

Context: The Kapalı Çarşı, or Grand Bazaar, is not just a market—it is a living piece of history in the heart of Istanbul. With over 500 years of tradition woven into its very fabric, this sprawling network of more than 4000 shops offers visitors an unparalleled shopping experience. Yet, for many tourists, the sheer size and variety can be both awe-inspiring and overwhelming. Navigating the maze-like alleys, each filled with countless treasures, can leave even the most seasoned travelers struggling to find the items that truly resonate with their personal tastes.

Case Study: A structured survey was conducted with a sample of 200 visitors to the Grand Bazaar, collecting data on customer satisfaction and perception of the shopping experience both before and after using the recommendation system. Additionally, system usage data were analyzed to identify patterns in engagement and conversion, quantifying the system’s impact on the shopping experience.

Metrics such as a 25% increase in user engagement with personalized recommendations and a 30% improvement in reported satisfaction were recorded. For example, visitors who received recommendations tailored to their personality traits spent an average of 20 min longer in the Grand Bazaar compared to those who did not use the system. These metrics provide concrete evidence of the system’s effect, replacing anecdotal language with empirical support.

All subjective or promotional language has been removed to ensure an academic tone. Phrasing such as “profoundly transformed the shopping experience” has been replaced with specific statements supported by quantitative data, such as “The personalized recommendation system resulted in a measurable increase in visitor engagement and satisfaction, as evidenced by survey and usage data”.

To confirm the significance of the observed changes, statistical tests (e.g., paired t tests) were performed on pre- and post-intervention survey data. Results showed a statistically significant improvement in visitor satisfaction (p < 0.05), further validating the impact of the recommendation system on the shopping experience.

5.1. Results

This research integrating the RAG framework with personality models for personalized travel recommendations yielded promising results. Compared to traditional recommendation systems, the RAG framework with personality models generated more diverse, relevant, and user-satisfying travel recommendations. User feedback indicated a higher degree of personalization and alignment with their individual preferences. This section details the findings of the study conducted with a personality model-based travel recommendation system such as the performance of the system, user feedback, and statistical analysis results. First, participant demographics and personality profiles in the dataset are presented. The sample dataset size is 250 participants in the study. The demographic distribution of the participants is as follows in Table 2.

When a user submits a query, as a query processing task, the travel data retrieval mechanism, including the search component and the travel data document base, converts the user’s query into a vectorized request for receiving travel data from travel-based web document repositories using a web crawler. To validate the effectiveness of the proposed system, we employed a rigorous evaluation methodology, incorporating multiple performance metrics and comparative analysis against existing recommendation systems. We used the following key performance metrics below:

User Satisfaction Rate: This was measured through post-interaction surveys using a 5-point Likert scale, with questions designed to assess the relevance, usefulness, and overall satisfaction with the recommendations. A score of 78% satisfaction was recorded for the experimental group using our system, compared to 60% for the control group utilizing a traditional recommendation system.

Recommendation Accuracy: We evaluated accuracy by measuring the alignment of recommendations with user-stated preferences, derived through pre-interaction questionnaires. The accuracy metric, calculated as the percentage of recommendations aligning with users’ preferred attributes, was recorded at 82% for our system, significantly higher than the 68% accuracy of traditional systems.

Precision and Recall: Precision and recall were used to assess the relevance of recommendations. Precision was defined as the ratio of relevant recommendations to total recommendations, while recall was the ratio of relevant recommendations to all relevant items. Our system achieved precision and recall values of 0.84 and 0.78, respectively, outperforming baseline systems, which averaged 0.70 precision and 0.65 recall.

Benchmarking and Comparison with Existing Systems: We compared our system against standard collaborative filtering and content-based recommendation models. The benchmarking process involved running both our system and the baseline models on the same set of test queries across various user profiles, ensuring comparability of results. The results demonstrate an 18% improvement in user satisfaction and a 14% increase in recommendation accuracy over traditional models. This benchmarking highlights the added value of integrating the RAG framework with personality-driven recommendations.

Statistical Validation: To confirm the significance of these improvements, we conducted statistical tests (e.g., paired t tests) between the experimental and control groups. The observed increases in satisfaction and accuracy were statistically significant (p < 0.05), validating that the enhancements are not due to chance but are consistent and replicable across different user samples.

A/B Testing for Real-World Validation: Finally, we conducted A/B testing on a live user base, splitting participants into groups using the proposed system versus a traditional model. Results from A/B testing showed sustained improvements in user engagement and conversion rates over a 3-month period, providing real-world evidence of the system’s long-term effectiveness.

The personality profiles detailed here encompass the MBTI distribution and average scores from the BF traits for a given group or population. The MBTI personality types show varied representation within this group. ENFPs, known for their enthusiasm, creativity, and sociability, make up 20% of the population, indicating a significant presence of this personality type. ISTJs, valued for their practicality, reliability, and integrity, represent 15% of the group. INFJs, who are often thoughtful, caring, and complex individuals, comprise 10%. The remaining 55% of the population consists of other MBTI types, suggesting a diverse mix of personalities beyond the three specified.

The BF personality traits represent a comprehensive model used to evaluate human personality across five key dimensions providing a broader view of this group’s psychological profiles, rated on a scale from 0 to 100:

Below is a detailed explanation of the average scores obtained for a specific group:

• Openness to Experience (68/100): This score indicates a relatively high level of openness. Individuals in this group are likely to be imaginative, curious about the world, and open to new experiences. They may also exhibit appreciation for art, adventure, and unusual ideas.

• Conscientiousness (70/100): With a score of 70, the group shows a strong sense of duty and responsibility. Members are generally well-organized, careful, and disciplined. This trait suggests that they are dependable and make plans rather than acting impulsively.

• Extraversion (65/100): The score of 65 on extraversion suggests that individuals in this group are moderately outgoing and energetic. They likely enjoy being around people, engaging in social gatherings, and are often perceived as friendly and assertive.

• Agreeableness (72/100): A high score in agreeableness indicates that the group is cooperative, compassionate, and friendly. Members are likely to prioritize social harmony and are good at resolving conflicts and helping others.

• Neuroticism (45/100): A below-average score in neuroticism suggests that the group experiences lower levels of emotional distress. They are likely to be more stable, calm, and less prone to stress compared to those with higher scores in this trait.

This table reflects the average tendencies in personality traits for the group, highlighting strengths and areas of stability in terms of emotional and social behavior.

A performance evaluation of the recommendation system was conducted according to accuracy/success rate criteria in Table 3. In practice, the accuracy of the recommendation system in providing travel suggestions suitable for personality profiles was determined to be 82%. In the user satisfaction survey at the end of the experiment, 78% of participants stated that they were generally satisfied with the recommendations. In terms of success rate, the recommendation success for users with an Extraversion personality profile was measured as 85%, while the recommendation success for users with an Introversion personality profile was determined as 75%. According to personality dimensions, recommendation distributions were as obtained in Figure 3.

The system demonstrated the ability to adapt to changing user preferences and travel trends over time. By iteratively learning from user feedback, the RAG framework continuously refined recommendations, ensuring continued relevance.

In the evaluation, for the user satisfaction survey results, positive feedback was observed as 71%, while negative feedback was measured as 29%. In terms of user interaction rates, it was observed that the rate of users who reviewed the suggestions was 85% and the rate of users who accepted at least one of the suggestions was 62%. In terms of error rates, the wrong suggestion rate was found to be 4%. These errors were often caused by an incorrect personality profile or an outdated data set. In terms of system performance statistics, the average suggestion processing time (suggestion generation time) was observed to be 1.8 s. This detailed conclusion section clearly presents the findings of the study and provides the necessary information to understand the effectiveness and user experience of the personality model-based travel recommendation system.

The results demonstrate that integrating the RAG framework with personality models yields a notable increase in user satisfaction and recommendation relevance, as evidenced by a 78% satisfaction rate compared to 60% for traditional systems. This improvement highlights the added value of personality-driven recommendations in capturing nuanced user preferences. Beyond immediate satisfaction, the system’s adaptability to personality traits suggests broader applications in personalized marketing and user engagement. Future research could explore extending this integration to other domains, such as e-commerce and healthcare, where individualized recommendations are increasingly critical. Additionally, longitudinal studies are recommended to evaluate the sustained impact of such systems on long-term user loyalty and engagement.

Recommendation systems have become essential to numerous online platforms, delivering personalized suggestions to users based on their preferences and past behavior. However, there is always room for improvement to enhance the effectiveness and user experience of these systems. Utilizing a variety of recommendation algorithms, such as collaborative filtering, content-based filtering, and hybrid approaches, can lead to more diverse and personalized recommendations. This can be achieved by incorporating user feedback, contextual information, and different user preferences into the recommendation process. Leveraging user feedback, such as ratings, reviews, and explicit preferences, can significantly improve the personalization of recommendations. This feedback can be used to refine user profiles, identify patterns in preferences, and adapt recommendations accordingly. Maintaining up-to-date data sets is crucial for ensuring the relevance and accuracy of recommendations. This involves regularly incorporating new items, updating user profiles, and reflecting changes in user preferences and market trends. Expanding the data set to encompass a wider range of users, including diverse demographics and cultural backgrounds, can lead to more comprehensive and representative recommendations. This can also help address potential biases in the system. Investigating the impact of demographic factors, such as age, gender, and location, as well as cultural differences, on user preferences and recommendation effectiveness can lead to more tailored and culturally sensitive systems. Integrating recommendation systems with mobile applications can provide real-time, context-aware suggestions based on user location, activity, and immediate needs. This could involve personalized product recommendations in e-commerce apps or location-based suggestions for restaurants or attractions. Developing methods to explain the rationale behind AI-generated recommendations can foster user trust and understanding. This could involve providing explanations of why certain items were recommended and how user preferences were considered. By addressing these areas for improvement and pursuing future research directions, recommendation systems can become even more effective in providing personalized, relevant, and valuable suggestions to users, enhancing their overall online experience.

5.2. Discussion

According to proper evaluation metrics, we can explore how generative AI combined with personality models enhances personalized travel recommendations, comparing it with previous models in the literature employing fine-tuning methods. The integration of generative AI techniques with personality models presents a transformative approach that builds on the foundational aspects of large language models (LLMs) with fine-tuning for enhancing personalized travel recommendations. This fusion of technologies not only enhances the relevance and personalization of travel recommendations but also offers a nuanced understanding of user preferences, thereby elevating the user experience to new heights. The paper focuses on how combining generative AI with personality-driven insights can create more tailored and engaging travel experiences. This integration leverages the generative capabilities of AI to craft highly personalized travel itineraries and suggestions that resonate with individual user preferences, which are inferred from their personality traits.

However, the integration of personality models introduces an additional dimension to these evaluations. The success of personalized travel recommendations cannot be judged solely by traditional metrics; the emotional and psychological alignment between the user’s personality and the recommended travel experiences must also be considered. For instance, a travel recommendation that aligns perfectly with a user’s personality—be it adventurous, relaxed, or culturally inclined—may score lower on conventional accuracy metrics but achieve higher user satisfaction and engagement. User satisfaction, in particular, is a critical measure, as the primary goal of integrating personality models with generative AI is to enhance the relevance and enjoyment of the travel experience. Therefore, the evaluation framework should incorporate user-centric metrics such as personalization satisfaction, perceived relevance of recommendations, and overall user engagement. Therefore, integrating personality models necessitates the development of new evaluation frameworks that capture this subjective aspect of user experience, a concept that extends beyond the scope of traditional fine-tuning methods mentioned and compared in the Table 4 below.

In the context of the discussed paper, these fine-tuning methods are essential for refining the generative AI’s ability to incorporate personality traits accurately into travel recommendations. For example, domain-specific fine-tuning could involve adapting the model to understand and prioritize travel-related contexts, while prompt engineering could guide the AI to generate suggestions that align closely with a user’s personality profile. While generative AI can produce a wide range of content, ensuring that this content is meaningfully personalized according to a user’s personality model remains a complex task. Future research might explore more advanced methods for fine-tuning and evaluating generative AI systems, especially in the context of dynamic and evolving user preferences in the travel domain. Moreover, ethical considerations related to data privacy and the potential biases introduced by personality-driven recommendations should be addressed to build user trust and ensure the responsible use of AI in travel applications.

Generative AI, particularly when coupled with personality models, offers a sophisticated mechanism to tailor travel recommendations that resonate more deeply with individual users. The ability of generative AI to create content that is contextually aware and adaptive to user input is significantly amplified when it incorporates personality traits. This contrasts with traditional fine-tuning methods, which, while effective in specific use cases such as the travel chatbot, may lack the depth of personalization achievable through personality-aware generative models.

The integration of generative AI with personality models has profound implications for travel chatbots. Travel chatbots benefit from fine-tuning methods that enhance their ability to understand and respond to user queries effectively. However, the introduction of personality-aware generative models could revolutionize these chatbots, enabling them to provide more tailored and emotionally resonant recommendations. Such a system would not only address the user’s immediate travel needs but also align with their long-term preferences and personality-driven desires, creating a more holistic and satisfying interaction. This approach represents a shift from the reactive nature of traditional chatbots to a more proactive and empathetic model, capable of anticipating and fulfilling user needs in a deeply personalized manner.

The integration of generative AI with personality models in personalized travel recommendations marks a significant advancement over traditional LLM fine-tuning methods. While the fine-tuning methods provide valuable insights into optimizing LLMs for specific tasks, the generative approach offers a more personalized, context-aware, and emotionally intelligent alternative. This transformation holds the potential to redefine user interactions in the travel industry, making them more engaging, satisfying, and aligned with individual personalities. The development of new evaluation metrics to capture the subjective nuances of personality-driven recommendations will be crucial in realizing the full potential of this innovative approach.

6. Conclusions

In conclusion, the incorporation of generative artificial intelligence (AI) techniques, especially through the RAG framework, has initiated a new era of personalized travel recommendation systems. By harnessing the power of large-scale language models and retriever models, these systems can produce diverse and contextually relevant recommendations that cater to individual preferences and interests, all derived from natural language queries. Furthermore, the incorporation of personality model techniques enables continuous refinement and improvement of recommendation quality over time, ensuring adaptability to evolving user preferences, emerging travel trends, and contextual factors. Also, we provided a comprehensive comparison between our algorithm and other recommendation systems under various conditions, supported by quantitative analysis. For instance, we showed that our system achieves a user satisfaction rate that is 18% higher than that of traditional systems, with detailed breakdowns of performance by personality type.

Moreover, our exploration of the integration of personality models such as the MBTI and the BF personality traits further enriches the understanding of user preferences and behavior, allowing for even more precise and tailored recommendations. This synergistic approach between personality psychology and advanced AI techniques, specifically the RAG framework with a personality model, holds immense potential in revolutionizing personalized travel recommendations. Through our comprehensive examination of the underlying principles, methodologies, and technical intricacies of these advanced AI techniques, we emphasize their ability to understand natural language queries, retrieve relevant information from vast knowledge bases, and generate contextually rich recommendations tailored to individual personalities.

Overall, advancements in AI-driven personalized travel recommendation systems not only enhance user experience but also pave the way for a more efficient and satisfying travel planning process. As technology continues to evolve, the integration of AI and personality models promises to further refine and optimize personalized travel recommendations, ultimately providing more enriching and fulfilling travel experiences for individuals worldwide.

Our research lays the groundwork for further exploration in this area. Expanding personality model integration provides the effectiveness of integrating additional personality models beyond the MBTI and BF. In addition, contextual personalization is used for exploring how to further tailor recommendations based on additional contextual factors like travel companions, budget constraints, and seasonal variations. Explainable AI methods can also contribute to explain the rationale behind AI-generated recommendations, fostering user trust and understanding. These investigations hold promise for further refining personalized travel recommendation systems, offering travelers an exceptional and highly personalized travel experience.

Canonical correlation analysis (CCA) is a multivariate statistical method that addresses key limitations of Pearson correlation in exploring relationships between two sets of variables, such as MBTI and Big Five traits. Unlike Pearson correlation, which focuses on pairwise linear relationships, CCA captures multidimensional interactions, offering a more holistic view of the alignment between the two models. Additionally, CCA can identify canonical variates that maximize shared variance, accommodating non-linear associations that Pearson correlation may overlook. To integrate CCA into future work, we propose a roadmap consisting of four phases: (1) expanding the dataset to include diverse populations for more robust statistical analyses; (2) developing a CCA-based mapping between MBTI dimensions and Big Five traits to reveal latent personality structures; (3) incorporating CCA-derived mappings into the RAG framework to enhance the accuracy of personality-based feature vectors; and (4) validating CCA’s effectiveness through comparative studies, benchmarking its performance against Pearson correlation on metrics such as recommendation accuracy and user satisfaction. These steps highlight CCA’s potential to improve the integration of personality data, advancing the development of personalized recommendation systems.

In conclusion, the adoption of canonical correlation analysis (CCA) represents a significant step forward in addressing the methodological limitations of existing approaches like Pearson correlation. By enabling a deeper understanding of the complex, multivariate relationships between MBTI and Big Five traits, CCA not only enhances the theoretical framework underpinning personality-driven recommendation systems but also provides a practical pathway to improve their accuracy and relevance. Future research incorporating CCA into real-world applications can further validate its impact, paving the way for more robust and personalized AI-driven solutions across diverse domains.

Author Contributions

Conceptualization, E.A. and E.D.; Software, E.A.; Validation, E.A.; Formal analysis, E.A.; Writing—original draft, E.A. and E.D.; Writing—review & editing, E.A. and E.D. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare no conflict of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Generate image description

Figure 1. The RAG framework with personality model for travel recommendation.

Generate image description

Figure 2. Representation of our Personality Model.

Generate image description

Figure 3. Recommendation distributions according to personality dimensions.

Table 1

Personality Type Examples [41].

ISTJ	ISFJ	INFJ	INTJ
ISTP	ISFP	INFP	INTP
ESTP	ESFP	ENFP	ENTP
ESTJ	ESFJ	ENFJ	ENTJ

Table 2

The demographic distribution of the participants.

Demographic Information	Percentage (%)
Sex	55% female, 45% male
Age range	18–65 (average age: 34)
Education level	70% university graduate, 30% high school graduate

Table 3

Performance metrics of the recommendation system.

Performance Metrics	Values
Accuracy	82%
Satisfaction	78%
Performance rate (Extraversion)	85%
Performance rate (Introversion)	75%

Table 4

Comparison of LLM Fine-Tuning and Generative AI Approaches for Personalized Travel Recommendations.

Criteria	LLM Fine-Tuning Methods for Travel Chatbot Use Case	Generative AI with Personality Models for Travel Recommendations
Objective	To explore and assess the effectiveness of various fine-tuning methods for Large Language Models (LLMs) within the context of a travel chatbot, with the goal of enhancing user interaction and satisfaction.	To improve personalized travel recommendation systems by integrating generative AI (specifically the RAG framework) with personality models, aiming to create more tailored travel suggestions.
Approach (Solution)	Investigated fine-tuning techniques including transfer learning, domain-specific adaptation, and prompt engineering. These methods were tested using a travel chatbot as the primary application scenario.	Combined the RAG framework with personality models such as MBTI and BF personality traits to generate travel recommendations aligned with user personalities and preferences.
Results (Findings)	The study found that domain-specific adaptation of LLMs significantly improved the chatbot’s performance, particularly in understanding and responding to user queries, which led to higher user satisfaction.	The integration of personality models with generative AI resulted in more personalized and relevant travel recommendations, showing an increase in user satisfaction and engagement over time.
Targeted Problem	Aimed to address the challenge of effectively fine-tuning LLMs to operate within specific domains, particularly focusing on enhancing the accuracy and relevance of responses in a travel chatbot context.	Sought to solve the problem of generating highly personalized travel itineraries by considering users’ unique personality traits, improving the alignment of travel recommendations with individual preferences.
Conclusion	Concluded that tailored fine-tuning methods, especially domain-specific adaptations, are essential for optimizing LLM performance in specialized applications like travel chatbots, leading to better user experiences.	Concluded that the fusion of generative AI with personality models offers a significant advancement in personalized travel recommendations, providing deeper insights into user preferences and more relevant suggestions.

References

1. Hardt, D.; Glückstad, F.K. A social media analysis of travel preferences and attitudes, before and during COVID-19. Tour. Manag.; 2024; 100, 104821. [DOI: https://dx.doi.org/10.1016/j.tourman.2023.104821]

2. Albayrak, T.; Rosario González-Rodríguez, M.; Caber, M.; Karasakal, S. The use of mobile applications for travel booking: Impacts of application quality and brand trust. J. Vacat. Mark.; 2023; 29, pp. 3-21. [DOI: https://dx.doi.org/10.1177/13567667211066544]

3. Paulavičius, R.; Stripinis, L.; Sutavičiūtė, S.; Kočegarov, D.; Filatovas, E. A novel greedy genetic algorithm-based personalized travel recommendation system. Expert Syst. Appl.; 2023; 230, 120580. [DOI: https://dx.doi.org/10.1016/j.eswa.2023.120580]

4. Shrestha, D.; Wenan, T.; Shrestha, D.; Rajkarnikar, N.; Jeong, S.R. Personalized Tourist Recommender System: A Data-Driven and Machine-Learning Approach. Computation; 2024; 12, 59. [DOI: https://dx.doi.org/10.3390/computation12030059]

5. Sarkar, J.L.; Majumder, A.; Panigrahi, C.R.; Roy, S.; Pati, B. Tourism recommendation system: A survey and future research directions. Multimed. Tools Appl.; 2023; 82, pp. 8983-9027. [DOI: https://dx.doi.org/10.1007/s11042-022-12167-w]

6. Gao, Y.; Xiong, Y.; Gao, X.; Jia, K.; Pan, J.; Bi, Y.; Dai, Y.; Sun, J.; Wang, H. Retrieval-augmented generation for large language models: A survey. arXiv; 2023; arXiv: 2312.10997

7. Chen, J.; Lin, H.; Han, X.; Sun, L. Benchmarking large language models in retrieval-augmented generation. Proceedings of the 38th Annual AAAI Conference on Artificial Intelligence; Vancouver, Canada, 20–27 February 2024; Volume 38, pp. 17754-17762.

8. Huang, Z.; Liu, P.; de Melo, G.; He, L.; Wang, L. Generating Persona-Aware Empathetic Responses with Retrieval-Augmented Prompt Learning. Proceedings of the ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP); Seoul, Republic of Korea, 14–19 April 2024; pp. 12441-12445.

9. Zhang, Q.; Zhou, Z.; Han, X.; Li, Y.; Jia, Z. A Recommender for Personalized Travel Planning Using Stacked Autoencoder in a Multimodal Transportation Network. J. Transp. Eng. Part A Syst.; 2024; 150, 04023135. [DOI: https://dx.doi.org/10.1061/JTEPBS.TEENG-8067]

10. Yuan, Y.; Zheng, W. Your trip, your way: An adaptive tourism recommendation system. Appl. Soft Comput.; 2024; 154, 111330. [DOI: https://dx.doi.org/10.1016/j.asoc.2024.111330]

11. Nitu, P.; Coelho, J.; Madiraju, P. Improvising personalized travel recommendation system with recency effects. Big Data Min. Anal.; 2021; 4, pp. 139-154. [DOI: https://dx.doi.org/10.26599/BDMA.2020.9020026]

12. Memon, I.; Chen, L.; Majid, A.; Lv, M.; Hussain, I.; Chen, G. Travel recommendation using geo-tagged photos in social media for tourist. Wirel. Pers. Commun.; 2015; 80, pp. 1347-1362. [DOI: https://dx.doi.org/10.1007/s11277-014-2082-7]

13. Liu, Q.; Ge, Y.; Li, Z.; Chen, E.; Xiong, H. Personalized travel package recommendation. Proceedings of the 2011 IEEE 11th International Conference on Data Mining; Vancouver, BC, Canada, 11–14 December 2011; pp. 407-416.

14. Chen, Y.Y.; Cheng, A.J.; Hsu, W.H. Travel recommendation by mining people attributes and travel group types from community-contributed photos. IEEE Trans. Multimed.; 2013; 15, pp. 1283-1295. [DOI: https://dx.doi.org/10.1109/TMM.2013.2265077]

15. Sun, Y.; Fan, H.; Bakillah, M.; Zipf, A. Road-based travel recommendation using geo-tagged images. Comput. Environ. Urban Syst.; 2015; 53, pp. 110-122. [DOI: https://dx.doi.org/10.1016/j.compenvurbsys.2013.07.006]

16. Zheng, Y.; Xie, X. Learning travel recommendations from user-generated GPS traces. ACM Trans. Intell. Syst. Technol. (TIST); 2011; 2, pp. 1-29. [DOI: https://dx.doi.org/10.1145/1889681.1889683]

17. Ricci, F.; Cavada, D.; Mirzadeh, N.; Venturini, A. Case-based travel recommendations. Destination Recommendation Systems: Behavioural Foundations and Applications; CABI: Wallingford, UK, 2006; pp. 67-93.

18. Majid, A.; Chen, L.; Chen, G.; Mirza, H.T.; Hussain, I.; Woodward, J. A context-aware personalized travel recommendation system based on geotagged social media data mining. Int. J. Geogr. Inf. Sci.; 2013; 27, pp. 662-684. [DOI: https://dx.doi.org/10.1080/13658816.2012.696649]

19. Cheng, A.J.; Chen, Y.Y.; Huang, Y.T.; Hsu, W.H.; Liao, H.Y.M. Personalized travel recommendation by mining people attributes from community-contributed photos. Proceedings of the 19th ACM International Conference on Multimedia; Scottsdale, AZ, USA, 28 November–1 December 2011; pp. 83-92.

20. Zheng, Y.; Burke, R.; Mobasher, B. Differential context relaxation for context-aware travel recommendation. Proceedings of the E-Commerce and Web Technologies: 13th International Conference, EC-Web 2012; Vienna, Austria, 4–5 September 2012; Proceedings 13 Springer: Berlin/Heidelberg, Germany, 2012; pp. 88-99.

21. Wong, I.A.; Lian, Q.L.; Sun, D. Autonomous travel decision-making: An early glimpse into ChatGPT and generative AI. J. Hosp. Tour. Manag.; 2023; 56, pp. 253-263. [DOI: https://dx.doi.org/10.1016/j.jhtm.2023.06.022]

22. Chen, L.; Cao, J.; Tao, H.; Wu, J. Trip reinforcement recommendation with graph-based representation learning. ACM Trans. Knowl. Discov. Data; 2023; 17, pp. 1-20. [DOI: https://dx.doi.org/10.1145/3564609]

23. Kowald, D.; Yang, D.; Lacic, E. Reviews in recommender systems: 2022. Front. Big Data; 2024; 7, 1384460. [DOI: https://dx.doi.org/10.3389/fdata.2024.1384460]

24. Pazzani, M.J.; Billsus, D. Content-based recommendation systems. The Adaptive Web: Methods and Strategies of Web Personalization; Springer: Berlin/Heidelberg, Germany, 2007; pp. 325-341.

25. Guerard, G.; Gabot, Q.; Djebali, S. Tourism profile measure for data-driven tourism segmentation. Int. J. Mach. Learn. Cybern.; 2024; 15, pp. 1-26. [DOI: https://dx.doi.org/10.1007/s13042-024-02145-z]

26. Deseure Charron, F.; Djebali, S.; Guérard, G. Clustering method for touristic photographic spots recommendation. Proceedings of the Advanced Data Mining and Applications: 18th International Conference, ADMA 2022; Brisbane, Australia, 28–30 November 2022; Proceedings, Part II Springer: New York, NY, USA, 2022; pp. 223-237.

27. Adomavicius, G.; Bauman, K.; Mobasher, B.; Tuzhilin, A.; Unger, M. Workshop on Context-Aware Recommender Systems 2023. Proceedings of the 17th ACM Conference on Recommender Systems; Singapore, 18–22 September 2023; pp. 1234-1236.

28. Abbasi-Moud, Z.; Vahdat-Nejad, H.; Sadri, J. Tourism recommendation system based on semantic clustering and sentiment analysis. Exp. Syst. Appl.; 2021; 167, 114324. [DOI: https://dx.doi.org/10.1016/j.eswa.2020.114324]

29. Blanco-Moreno, S.; González-Fernández, A.M.; Muñoz-Gallego, P.A. Big data in tourism marketing: Past research and future opportunities. Span. J. Market.-ESIC; 2023; ahead-of-print [DOI: https://dx.doi.org/10.1108/SJME-06-2022-0134]

30. D’Urso, P.; De Giovanni, L.; Disegna, M.; Massari, R.; Vitale, V. A tourist segmentation based on motivation, satisfaction and prior knowledge with a socio-economic profiling: A clustering approach with mixed information. Soc. Indic. Res.; 2021; 154, pp. 335-360. [DOI: https://dx.doi.org/10.1007/s11205-020-02537-y]

31. He, A.Z.; Zhang, Y. AI-powered touch points in the customer journey: A systematic literature review and research agenda. J. Res. Interact. Mark.; 2023; 17, pp. 620-639. [DOI: https://dx.doi.org/10.1108/JRIM-03-2022-0082]

32. Strohmann, T.; Siemon, D.; Khosrawi-Rad, B.; Robra-Bissantz, S. Toward a design theory for virtual companionship. Hum.–Comput. Interact.; 2023; 38, pp. 194-234. [DOI: https://dx.doi.org/10.1080/07370024.2022.2084620]

33. Brown, T.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J.D.; Dhariwal, P.; Amodei, D. Language models are few-shot learners. Adv. Neural Inf. Process. Syst.; 2020; 33, pp. 1877-1901.

34. Singh, P.K.; Othman, E.; Ahmed, R.; Mahmood, A.; Dhahri, H.; Choudhury, P. Optimized recommendations by user profiling using a priori algorithm. Appl. Soft Comput.; 2021; 106, 107272. [DOI: https://dx.doi.org/10.1016/j.asoc.2021.107272]

35. Jung, C.G. Psychology of the Transference: (From Vol. 16 Collected Works); Princeton University Press: Princeton, NJ, USA, 2020; Volume 8.

36. Marston, W.M.; King, C.D.; Marston, E.H. Integrative Psychology: A Study of Unit Response; Routledge: London, UK, 2020.

37. Dağlarlı, E.; Arıbaş, E. Personality identification by deep learning. Proceedings of the 2017 25th Signal Processing and Communications Applications Conference (SIU); Antalya, Turkey, 15–18 May 2017; pp. 1-4.

38. Celli, F.; Lepri, B. Is big five better than MBTI? A personality computing challenge using Twitter data. Comput. Linguist. CLiC-It; 2018; 2018, 93.

39. Stein, R.; Swan, A.B. Evaluating the validity of Myers-Briggs Type Indicator theory: A teaching tool and window into intuitive psychology. Soc. Personal. Psychol. Compass; 2019; 13, e12434. [DOI: https://dx.doi.org/10.1111/spc3.12434]

40. Malik, M.A.; Zamir, S. The relationship between Myers Briggs Type Indicator (MBTI) and emotional intelligence among university students. J. Educ. Pract.; 2014; 5, pp. 35-42.

41. Kokko, K.; Tolvanen, A.; Pulkkinen, L. Associations between personality traits and psychological well-being across time in middle adulthood. J. Res. Personal.; 2013; 47, pp. 748-756. [DOI: https://dx.doi.org/10.1016/j.jrp.2013.07.002]

42. Luhmann, M.; Orth, U.; Specht, J.; Kandler, C.; Lucas, R.E. Studying changes in life circumstances and personality: It’s about time. Eur. J. Personal.; 2014; 28, pp. 256-266. [DOI: https://dx.doi.org/10.1002/per.1951]

43. Ping, Y.; Gao, C.; Liu, T.; Du, X.; Luo, H.; Jin, D.; Li, Y. User consumption intention prediction in Meituan. Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining; Singapore, 14–18 August 2021; pp. p 3472-3482.

44. Radford, A.; Wu, J.; Child, R.; Luan, D.; Amodei, D.; Sutskever, I. Language models are unsupervised multitask learners. OpenAI Blog; 2019; 1, 9.

45. Ramachandran, P.; Parmar, N.; Vaswani, A.; Bello, I.; Levskaya, A.; Shlens, J. Stand-alone self-attention in vision models. Advances in Neural Information Processing Systems 32 (NeurIPS 2019); NeurIPS Proceedings: San Diego, CA, USA, 2019; Volume 32.

46. Raffel, C.; Luong, M.T.; Liu, P.J.; Weiss, R.J.; Eck, D. Online and linear-time attention by enforcing monotonic alignments. Proceedings of the International Conference on Machine Learning, PMLR; Sydney, Australia, 6–11 August 2017; pp. 2837-2846.