Content area
Translation from original language as provided by author
With the rapid development of information technology, people hope to communicate with computer in natural language as human use. Natural language understanding is an interesting and challenging task. From the view of computer science, especially artificial intelligence, the task of natural language understanding is to build a computer model which can understand ,analyze and answer questions as human usually do. Chinese natural language processing is the core technology in enabling computer to understand Chinese. During the process of semantic understanding, automatic word segmentation is the initial and basic step. Word segmentation can extract key words from sequential characters for further semantic analysis. Providing adequate words and filtrating redundant noise is guarantee of quality and speedy in subsequent processing. Named entity recognition is a subtask of information extraction aiming at recognizing names of persons, organizations or locations etc. in text. Usually these entities are unique in real world. So named entity recognition plays an important role in information processing systems such as information retrieval, information extraction. Good performance of named entity recognition benefits information processing system much. Both Chinese word segmentation and named entity recognition could be regarded as sequential labeling task. Conditional Random Fields is a popular sequential labeling statistical language model. Deriving from Markov Random in graphical model, conditional random fields represents the distribution of variables through structure of graph and probability theory. Linear-chain conditional random fields, a simple form of conditional random fields, would calculate joint probability distribution of variables related in graph and get the final label sequence by maximizing a posteriori criterion.In this thesis original principle of conditional random fields was presented. Compared with other traditional statistical language model in theory and practice, Conditional Random Fields is suitable for sequential labeling task with excellent performance. Under the same circumstance conditional random fields out-performs other traditional language model. In a word Conditional Random Fields is a state-of-art sequential labeling language