Fuzzy Linguistic Query-based User Profile Learning by Multiobjective Genetic Algorithms

In this paper, a multiobjective genetic algorithm is proposed to automatically learn persistent fuzzy linguistic queries for text retrieval applications. These queries are able to represent user's long-term standing information needs in a more interpretable way than the classical "bag of words" user profile structure. Thanks to its multiobjective nature, the introduced genetic fuzzy system is able to build different queries for the same information need in a single run, with a different trade-off between precision and recall. The experiments performed on the classical CACM collection show that although the different queries obtained from our genetic fuzzy system are less accurate in the retrieval task than those derived by one state-of-the-art bag of words method, they compose more flexible, comprehensible and expressive user profiles