Research On Query Expansion And Related Technologies Of Information Retrieval

With the rapid development of Internet technology, the information on the Internet increases exponentially. An important research fouses on how to deal with these great capacities of information and acquire relevant data that we need. Techniques for query expansion have been extensively studied in information retrieval research as a means of processing the word mismatch between queries and documents or defective query expression. These techniques can re-construct or expand the query terms. Researches on query expansion have become hotspot in information retrieval domain, for theoretic importance and practical meaning. Studies mainly including in the thesis are as follows:(1) This paper compares the traditional retrieval models, such as Boolean Model, Vector Space Model and Probabilistic Model. And we proposes a N-level Vector Space Model. In the process of information retrieval, the document is divided into N levels by this model. The related and irrelated documents are well distinguished by dispatching different proportional factors to the words in different levels according to the importance of each level. Experimental results show the precision of the retrieval system has been improved by using N-level Vector Space Model.(2) This paper analyses the advantages and limitations of Relevance Feedback method, Global Analysis method and Local Analysis method. We proposes a novel query expansion method based on local co-ocurrence information, which combines the advantages of the methods of Global Analysis and Local Analysis. It selects most appropriate expansion terms by utilizing the local co-occurrence informtion in top-ranked documents and the global statistical information in the whole collection. Experimental results show that our method offers more effective and robust retrieval performances.(3)A network information retrieval experimental system has been designed and implemented. N-level Vector Space Model and query expansion method based on local co-occurrence information are implemented in the system to improve the recall and precision.The system has rapid response time and good expansibility.