Browsing by Subject "recommender system"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Enhancing User Experience With Recommender Systems Beyond Prediction Accuracies
    (2016-08) Nguyen, Tien
    In this dissertation, we examine to improve the user experience with recommender systems beyond prediction accuracy. We focus on the following aspects of the user experience. In chapter 3 we examine if a recommender system exposes users to less diverse contents over time. In chapter 4 we look at the relationships between user personality and user preferences for recommendation diversity, popularity, and serendipity. In chapter 5 we investigate the relations between the self-reported user satisfaction and the three recommendation properties with the inferred user recommendation consumption. In chapter 6 we look at four different rating inter- faces and evaluated how these interfaces affected the user rating experience. We find that over time a recommender system exposes users to less-diverse contents and that users rate less-diverse items. However, users who took recommendations were exposed to more diverse recommendations than those who did not. Furthermore, users with different personalities have different preferences for recommendation diversity, popularity, and serendipity (e.g. some users prefer more diverse recommendations, while others prefer similar ones). We also find that user satisfaction with recommendation popularity and serendipity measured with survey questions strongly relate to user recommendation consumption inferred with logged data. We then propose a way to get better signals about user preferences and help users rate items in the recommendation systems more consistently. That is providing exemplars to users at the time they rate the items improved the consistency of users’ ratings. Our results suggest several ways recommender system practitioners and re- searchers can enrich the user experience. For example, by integrating users’ personality into recommendation frameworks, we can help recommender systems deliver recommendations with the preferred levels of diversity, popularity, and serendipity to individual users. We can also facilitate the rating process by integrating a set of proven rating-support techniques into the systems’ interfaces.