On The Reuse of Past Searches in Information Retrieval: Study of Two Probabilistic Algorithms

When using information retrieval systems, information related to searches is typically stored in files, which are well known as log files. By contrast, past search results of previously submitted queries are ignored most of the time. Nevertheless, past search results can be profitable for new searches. Some approaches in Information Retrieval exploit the previous searches in a customizable way for a single user. On the contrary, approaches that deal with past searches collectively are less common. This paper deals with such an approach, by using past results of similar past queries submitted by other users, to build the answers for new submitted queries. It proposes two Monte Carlo algorithms to build the result for a new query by selecting relevant documents associated to the most similar past query. Experiments were carried out to evaluate the effectiveness of the proposed algorithms using several dataset variants. These algorithms were also compared with the baseline approach based on the cosine measure, from which they reuse past results. Simulated datasets were designed for the experiments, following the Cranfield paradigm, well established in the Information Retrieval domain. The empirical results show the interest of our approach