Large scale information retrieval using a Scatter Search based approach

Information retrieval (IR) has been widely investigated these last decades and significant results have been applied to several domains like e-commerce, e-library and automatic medical diagnostics. However, very few studies in this area deal with artificial intelligence or AI tools. Knowing the power of meta-heuristics in problem solving, we suggest exploring information retrieval with an evolutionary approach. It appears that the methodology that seems structurally well suited to this problem is undoubtedly scatter search. The concept of similarity in IR between a query and a document according to the vector space model is correlated to the concept of the distance between solutions in scatter search. In this paper, we design and implement a scatter search algorithm for information retrieval called SS-IR. The algorithm is tested on the well known smart collections and large data sets built for experimentation purposes. Numerical results are encouraging and show an interesting performance for the algorithm especially for large scale information retrieval.