Evolving General Term-Weighting Schemes for Information Retrieval: Tests on Larger Collections