Search engines are supposed to return all relevant documents to users. The existing search engines read the text as a sequence of words without meaning. The search is then limited to find documents that contain the same words than the user‘s query that reduces the relevance in returned documents.
In this work, we propose a semantic search engine that represents user‘s information need by a pattern of relevance.
Our approach is characterized by the use of Latent Semantic Analysis (LSA) to find the words semantically correlated, semantic links are then created by using Discourse Representation Theory (DRT) that offers the possibility to translate a sentence from natural language to logical representation. The found links are saved in an ontology.
Information retrieval is executed by comparing the knowledge extracted from each document to the relevance pattern of user’s request. This allows a better identification of information’s needs, and restrict the set of documents returned.