Query Expansion Based On The Concept Of User Interest

Since the precision and recall of web information retrieval sysrems is very low now, we present a new method for automatic query expansion to improve the performance of retrieval. Techniques for automatic query expansion have been extensively studied as a means for addressing words mismatch between queries and documents. However, these techniques always have many limitations; for instance, they can't solve synonymity and ambiguity in natural language synchronously. After analyzing the limitations of trational techniques of query expansion, we describe a new technique of query expansion based on concept and user profile. Firstly, we set up the ontology knowledge bases in a prededined set of domains based on Yahoo classified category, and retrieval documents using conceptual query expansion to replace traditional methods based on key words. Secondly, we utilize datamining and machine learning to learn user's interest profile automaticly. For example, with the IE browser history, user favorite and log file, we can mine user's browse habits. By making the individual query expansion, we can eliminate ambiguity of query words and expand words. The paper mainly includes some researches as following:Researchs on foundation: After investigating the merits and defeats of traditional query expansion, we present a new method of query expansion based on concept and user profile. By constructing the ontology knowledge bases in specifically domain and mining user's interest profile automaticly, our method can eliminate synonymity and ambiguity of query words and expand words synchronously, and improve the performance of information retrieval.Constructing predefined domain ontology knowledge base: Firstly we analysis each document in the Yahoo category and extract all terms. Then a concept can be formed by mapping the given category and the terms extracted form documents in that category. The terms in that concept can be used to expand the original query. Additionally, to avoid synonymy phenomena, we use the thesaurus in wordnet to expand the concept.Modeling user profile: After constructing ontology knowledge base, we can train SVM classification using this konwlodge base. By classifying the pages in IE browser history and user favorite with the trained SVM, we can mine the user profile.Mechanisms of query expansion: by combining the ontology knowledge base and user profile, we present two mechanisms of query expansion.Experiment result and evaluation: we simply introduce the system mudules and the Java classes contained in the modules of the prototype of query expansion system based on user profile andknowledge base, and the steps to implement the system. Experiment is carried out in our information retrieval system based on vector spase model. At last we evaluate the performance of our method of query expansion with traditional simple retrieval method and the local context analysis.