A supervised term-weighting technique for topic-based retrieval

This article presents a technique for term weighting that relies on a collection of documents labeled as relevant or irrelevant to a topic of interest. The proposed technique weights terms based on two factors representing the descriptive and discriminating power of the terms. These factors are combined through the use of an adjustable parameter into a general measure that allows for the selection of query terms that independently favor different aspects of retrieval. The proposed weighting technique is applied in the development of topic-based retrieval strategies that can favor precision, recall or a balance between both. The strategies are analyzed using a collection of news from the The Guardian newspaper, labeled by the authors, the 20 newsgroups data set, and the Reuters-21578 collection. Finally, a comparative evaluation of retrieval effectiveness for different retrieval goals is completed. The results demonstrate that, despite its simplicity, our method is competitive with state-of-the-art methods and has the important advantage of offering flexibility at the moment of adjusting to specific task goals.