Content area
This dissertation explores advanced representation learning techniques as an essential tool for enhancing management intelligence across diverse sectors. In today’s data-driven world, organizations are increasingly confronted with large, complex, and often unstructured datasets. To effectively harness this data for decision-making, sophisticated analytical techniques are crucial. Management challenges—ranging from optimizing human resources and assessing educational outcomes to streamlining industrial operations—often involve intricate data structures that traditional analytical methods struggle to decode. Representation learning, a key area of machine learning, focuses on automatically learning efficient representations from raw data, enabling more accurate predictions and insights. Unlike conventional methods, which rely on manually engineered features, representation learning techniques can discover the underlying structures in data, significantly improving the accuracy, scalability, and flexibility of analysis.
In management contexts, the ability to derive actionable insights from such complex data is critical for informed decision-making and strategic planning. The growing reliance on data for operational success has placed greater emphasis on machine learning, especially representation learning, which has seen rapid advances in its capacity to handle complex and large-scale datasets. These techniques help uncover deep, often hidden patterns in data, which can be leveraged for forecasting trends, improving processes, and ultimately making more intelligent, data-backed decisions. By focusing on methods that automatically adapt to diverse data structures, representation learning has the potential to revolutionize the way organizations approach problem-solving and decision-making.
This research specifically focuses on three core application areas where representation learning has the potential for significant impact: organizational networks, educational outcome assessments, and industrial operations. Each of these domains involves distinct data types and modalities, requiring tailored approaches to effectively address their unique challenges. In the case of organizational networks, where data is often relational and multi-relational, graph learning techniques are leveraged to capture complex relationships between employees, teams, and departments. For educational outcome assessments, where the data involves dynamic, longitudinal transitions, dynamic graph analysis is applied to model the evolving career paths of graduates and assess institutional performance. Finally, industrial operations, which are characterized by sequential and time-series data, are optimized using sequential data processing techniques that adapt to real-time changes in operational environments. The proposed methods are not only innovative in their respective domains but also adaptable, offering scalable solutions that can be applied to diverse management contexts, depending on the nature and structure of the data at hand.
