The Information Retrieval Model Based On Markov Concept

With the dynamic increase of Web information resource, how to get exact information rapidly and expediently is becoming a problem need be solved urgently. The ambiguity of natural language and the random and dynamic of users'information needs result in the retrieval inefficient. Therefore, query expansion is an indispensable technique for solving this problem.Query expansion is an effective approach to improve retrieval efficiency. In traditional approaches of query expansion, a term will be chosen if it associates strongly with a certain query term in a given query. However, query concept is rarely taken into account. In reality, terms, which are strongly similar to the query topic rather than a single query term, are more important and benefit to improve efficiency .Therefore, such kind of terms should be added to original query. Therefore, an information retrieval model based on Markov concept is proposed in this paper. A Markov Network is an undirected graphical network that is capable of efficiently representing relevance in knowledge and can be easily gotten from training data with strong learning and inferring capability.In this paper, the models learn from document set, and extract relationship between index terms and between query terms to construct a Markov knowledge network. Shapes of concept mined from Markov Network are added into query process. The experiments show that our models make significant improvements in retrieval performance.New points in this thesis:1. The shapes of concept mined from Markov Network, such as clique and Markov concept graph, are added into query process. The Markov Network information retrieval model based on clique strengthens the simple relationships between terms. It considers clique as a concept and adds it as a whole into query expansion. However, the information retrieval model based on Markov concept graph focus on dependency between query terms. It makes the relationship of query terms transfer to index terms and integrates candidate pruning into query process. In this way, some noisy candidates will be pruned, and the key terms relevant to query topic are added into. Therefore, it is helpful to improve efficiency.2. Analysis and verify the performance of the information retrieval model based on Markov concept graph, compare the result with other information retrieval model and other Markov network information retrieval models. The models proposed in this paper perform outstanding and make significant improvements. In additional, the model based on Markov concept graph has the best performance and decrease computational overhead in query process.