A Novel Approach to Trust Based Recommender Systems Leveraged by Context Attributes

摘要

Internet has undergone many fold expansion in last couple of decades or so, the pitfall of that is the data overloading problem which has become extremely intricate for retrieving useful information from internet. Users searching for intended contents have endless number of Web pages to navigate and require enormous efforts, requires judgmental aptitude and intuitiveness to extract meaningful information from almost unlimited number of pages and huge content. Recommender systems are meant to be an important solution to the data overload and skewed information problem that persists today in World Wide Web. Very recent research trend in Recommender systems encourages towards consideration of Context awareness along with trust based filtering.One of the major challenges in the context aware recommender system is the selection of relevant contexts and appropriately weighting the relevant contexts for prediction calculations. Also dynamic nature of trust puts practical challenges in using trust based recommender system. The selection of a few most relevant contexts and using them with proper importance factors incorporates aspects of dynamic behaviour of trust and are vital for enhanced accuracy in the recommender output, as irrelevant and inappropriate contexts assimilation decreases the accuracy of recommender output, creates data sparseness and also increases the computational complexity. There are various ways to infuse the weightages of relevant contexts in recommender systems. While doing the neighbourhood formation, trust propagation and predictions, context weighting plays a pivotal role towards increasing the accuracy of Recommender Systems. In this paper, we propose an approach that incorporates the relative weightages of relevant contexts in trust calculations and neighbourhood formation. Trust network thus formed is leveraged by the context attributes. This approach is advantageous in terms of increased recommender accuracy and also overcome data sparseness of hard context filtering methods.