An Entropy-Based Approach for Preserving Diversity in Evolutionary Topical Search

CECILIA BAGGIO; ROCÍO L. CECCHINI; CARLOS M. LORENZETTI; ANA MAGUITMAN

Buenos Aires

Simposio; Simposio Argentino de Inteligencia Artificial (ASAI 2016) - JAIIO 45 (Tres de Febrero, 2016); 2016

Topic-based information retrieval is the process of matching a topic of interest against the resources that are indexed. An approach for retrieving topic relevant resources is to generate queries that are able to reflect the topic of interest.Multi-objective Evolutionary Algorithms have demonstrated great potential to deal with the problem of topical query generation. In an evolutionary approach to topic-based information retrieval the topic of interest is used to generate an initial population of queries, which is evolved towards successively better candidate queries. A common problem with such an approach is poor recall due to loss of genetic diversity. This work proposes a novel strategy inspired on the information theoretic notion of entropy to favor population diversity with the aim of attaining good global recall. Preliminary experiments conducted on a large dataset of labeled documents show the effectiveness of the proposed strategy.