A Pareto Local Search Algorithm for the Bi-Objective Composite Retrieval Problem

PAULA ZABALA; GUILLERMO CABRERA-GUERRERO; MAURICIO MOYANO CASTILLO

Dublin

Conferencia; 30th European Conference on Operational Research EURO 2019; 2019

Association of European Operational Research Societies

In information retrieval, traditional search strategies only consider one attribute to build up a ranking list of results which depends exclusively on the considered attribute. However, often one needs to rethink the original query to accomplish the right solution as these search strategies do not consider the existing relations among all the other attributes. As a response to this behaviour, composite retrieval (CR) of diverse and complementary bundles has been proposed. Its objective is to group elements into bundles, in which the items are related to each other under both criteria: similarity and complementarity of bundles. This bundles should satisfy users' expectations without the needing of a new intervention, improving the searching experience.In spite of its inherent multi-objective nature, so far, the CR problem has been only addressed from a single-objective point of view. Grouping algorithms, as well as local search algorithms, have been proposed to solve this problem. Unlike these single-objective approaches, in this study, we propose a Pareto local search algorithm to solve the bi-objective CR problem, where both similarity and complementarity criteria must be maximised.Results show that our PLS algorithm obtains a high-quality set of (locally) non-dominated bundles in a reasonable time.