Query Recommendation based terms and relevant documents using language Models

The query submitted by the user is only a partial and often ambiguous expression of his need. This implies that it is essential to suggest to the users the most related queries to the context of their queries. However, the notion of context is quite broad and includes all the elements related to the query (Its field, its environment, the user profile, his preferences and his search history). In this paper, we extract the environment of a user’s query in order to use it later in its query recommendation process. For this purpose, three different methods of query recommendation are proposed, and then compared based on the quality of the extracted environments, by calculating the Average Internal Similarity (AIS) of each built environment. The results show that the information of documents relevance influence the similarity between queries better than the information of existence of terms for all the proposed approaches. The final experiment was a comparison between the three approaches, and it shows that for short and long queries the highest value of AIS is reached by the TLM approach using Language Models based on common terms and relevant documents. Key-Words: Information Retrieval, Query Recommendation, Language Model, Recommendation Algorithm, Query’s context.