AN INTERACTIVE INTERFACE FOR INSTILLING TRUST AND PROVIDING DIVERSE RECOMMENDATIONS

Abstract

Recommendation systems are software tools that provide suggestion of items that a user may like. In the thesis, we basically work under two disciplines of the recommender system- context aware recommender system and providing explanations in recommender approach. The incorporation of context in recommendation systems is important for providing good recommendations as it improves the performance. Context awareness is an important concept in many disciplines. There are various methods that have already been proposed for context aware recommendation systems and can be classified into pre-filter, post-filter and contextual generalization. However, we found that there are some problems in using a single approach. So, in the first phase of our thesis work, we propose a new approach by combining pre-filter and post-filter methods based on the importance of contextual attribute. We believe that the contextual attribute should be dynamically selected for every user, as we found that it varies according to the user behaviour. In our approach, in case of a few rating scenario a simple method is used to provide a solution. The strategy adopted is advantageous over the single approach and is effective in improving the accuracy for the context aware recommender system. In the second phase of our work we found that recommendation system ability to instill trust in its users and convince them about the recommendations provided is effective using suitable explanations. Although majority of existing research focus on the algorithm used to provide explanation, our algorithm focuses on the presentation of explanation interface for making the user understand the recommendations and its explanation better. The major contribution of our work is that it designs a complete model consisting of various visualization styles, where by each style is used for a specific purpose only. The explanation interface designed is for a hybrid recommender system where the explanation interface of individual recommender is re-used in a form that describes it best. In this model the user is held at an important position and using the explanations can incorporate diversity in the recommendations provided by the system. Results obtained shows that users found the system developed very interactive, appealing, assisting them to better understand the recommendations using various styles, low disappointment level and capable of accounting for diverse recommendations thus leading to overall satisfaction.