Multi-Objective Optimisation for Information Access Tasks

Information access tasks involve setting a large number of model parameters (such as term weighting functions or the particular terms from the document collection to use in the model); in addition, information access tasks frequently attempt to satisfy multiple competing objectives, such as precision and recall. We investigate the use of multi-objective evolutionary algorithms for finding model parameters that optimise both precision and recall. We present information retrieval experiments using the TREC 7 and 8 collections for ad hoc query problems and the Cora collection for document classification problems. In particular, we compare the following optimised term weighting functions in retrieval and classification tasks: , , , and BM25. We also optimise the range of terms used for document classification. We show that multi-objective optimisation can yield significantly enhanced performance beyond the default parameter settings and indicates that more terms than current stop-word lists should be exluded. This study shows the wide range of possibilities using multi-objective optimisation for information access tasks.