Implicit user behaviours to improve post-retrieval document relevancy

The collection of user feedback as indications of users' interests resulted in a growing interest in improving users' search experiences. In this article, we describe a method that integrates multiple implicit feedback approaches to unobtrusively monitor users' interactions to improve document search results relevancy. The study gathered users' feedback based on the dwell time, click-through data, page review, and also text selection. An experiment was conducted to assess the performance of the proposed integrated model. Collected data were analysed and compared at three ranking levels, that is, top 10, 15 and 25. Both the mean average precisions and normalised discounted cumulative gain values indicate the integrated model to significantly outperform the baseline (TF-IDF) at each of the varying levels. Moreover, a comparison across all the models also show the integrated model to have the best search performance further indicating that merging multiple feedback techniques improves the overall document relevancy. Results also show page review and text selection have the lowest and highest precisions, respectively among all the four implicit feedback models, however the differences are insignificant. Overall it can be concluded that integrated implicit feedback significantly improves post-retrieval document relevancy compared to stand-alone feedback, and also when no feedback is available.