Bengali document information retrieval using query expansion

Abstract

The goal of Information Retrieval (IR) is to retrieve documents from a vast document collection whose content fits a user query. Since most users struggle to create well-designed queries, query expansion is required to extract relevant information. Query expansion (QE) is an IR process that includes selecting and adding terms to a user's query in order to decrease query document mismatch and hence increase retrieval performance. Various QE techniques are commonly used to boost the efficiency of textual information retrieval systems. The amount of Indian language (IL) electronic documents has increased significantly as a result of globalization. As a result, the need for developing IR systems to deal with this growing collection is paramount. In this thesis , we propose a Bengali document Information Retrieval system using Query Expansion which implements the Vector Space Paradigm of Information Retrieval and uses the Pseudo Relevance Feedback approach of Query Expansion. Using our Information Retrieval System, we create Initial Search Results for a certain query. We re-analyze the top documents from the Initial Search Results and reformulate the initial query using our term selection algorithm. To obtain the Final Search Results, we take the reformulated query as the expanded query and rerun it through the same Information Retrieval System. For our experiments we have proposed seven comparison models. All experiments have been performed on Queries 100 to 125 of the FIRE 2010 dataset. analyzing the MAP of these said seven models against the MAP of the baseline model. Experimental results show that the MAP of our proposed method using QE improves over the MAP of the baseline Information Retrieval System without using QE . The hybrid collaboration model outperforms the baseline model on a higher level. The MAP of this model on the final search results is significantly higher than the MAP of our baseline model on initial search results without employing QE, demonstrating that the Pseudo Relevant Feedback approach of QE does actually aid in the retrieval of more relevant Bengali documents.