Proposing an Information Retrieval Model Using Interval Numbers

Recent expansions of web demands for more capable information retrieval systems that more accurately address the users' information needs. Weighting the words and terms in documents plays an important role in any information retrieval system. Various methods for weighting the words are proposed, however, it is not straightforward to assert which one is more effective than the others. In this paper, we have proposed a method that calculates the weights of the terms in documents and queries as interval numbers. The interval numbers are derived by aggregating the crisp weights that are calculated by exploiting the existing weighting methods. The proposed method, calculates an interval number as the overall relevancy of each document with the given query. We have discussed three approaches for ranking the interval relevancy numbers. In the experiments we have conducted on Cranfield and Medline datasets, we have studied the effects of weight normalization, use of variations of term and document frequency and have shown that appropriate selection of basic term weighting methods in conjunction with their aggregation into an interval number would considerably improve the information retrieval performance. Through appropriate selection of basic weighting methods we have reached the MAP of 0.43323 and 0.54580 on the datasets, respectively. Obtained results show that he proposed method, outperforms the use of any single basic weighting method and other existing complicated weighting methods.