IDUF: An active learning based scenario for relevance feedback query expansion

In usual Information Retrieval (IR) systems, the user query is represented in the form of a keyword set. Information resources are retrieved according to their similarities to this query. Consequently if query is not declared with appropriate terms, retrieved results would not be satisfactory. Therefore query refinement procedures are incorporated to improve the efficiency of the IR systems. In this paper, an active learning approach has been proposed for query expansion (QE) according to user feedbacks. A novel document selection procedure is used to acquire user feedbacks. In this procedure, firstly, the whole set of documents are classified according to existing feedbacks. Then a set of documents which are classified with low certainty and do not produce redundant information are selected as informative documents to get user feedbacks. In this scenario, the number of feedbacks is equal to customary relevance feedback methods but retrieval system would gain more useful information. Experimental results on Reuters-21578 full-text dataset demonstrate considerable improvement in the performance of retrieval system. It is shown experimentally that the proposed method can effectively employ user´s feedback in discovering the favorable hidden concepts too.