NEW INFORMATION RETRIEVAL APPROACH BASED ON SEMANTIC INDEXING BY MEANING

Abstract

An Information Retrieval System (IRS) offers a number of tools and techniques, which enable to locate and visualize the relevant information needed. This information, is expressed by the user in the form of a query natural language. However, the representation of documents and the query in a traditional IRS lead to a lexical-centered relevance estimation which is, in fact, less efficient than a semantic-focused estimation. As a consequence, the documents that are actually relevant are not being recovered if they do not share words with the query, while the documents non relevant, which are words in common with the query, are recovered even though at times they do not have the meaning intended. This paper tackles this problem while suggesting a solution in the level of indexation of an IRS allowing it to improve its performance. To be more precise, we suggest a new approach of semantic indexation allowing to lead to the exact meaning of each term in a document or query undergoing a contextual analysis at the sentence level. In fact, if the system is able to comprehend the need of the user, then consequently it is perfectly capable to respond to it. Add to that, we suggest a simple method allowing to apply any model of IR on our new index table without changing its original bases making it faster. In order to validate this proposed approach, this new created system is evaluated base on numerous collections naming “TIME” , “BBC” , “The Guardian” and “BigThink” . The results based on the experiments indicate the efficacy of our hypothesis compared to traditional IR approaches.