面向隐式反馈的推荐系统研究现状与趋势

Research Status and Future Trends of Recommender Systems for Implicit Feedback

推荐系统作为解决信息过载的一种有效手段,已成为工业界和学术界的研究热点,它依据用户的显式或隐式反馈信息推测其需求、兴趣等,将其偏好的信息、产品等推荐给他们。面向显式反馈信息的推荐方法是目前的主流,而隐式反馈信息的普遍性使得基于此类的推荐方法具有更广的适用性,但是,隐式反馈信息并不能直接反映用户的偏好,因而利用它进行推荐具有很大的挑战。首先阐述了隐式反馈的特性以及基于此类信息进行推荐的必要性和所面临的问题;然后对面向隐式反馈的推荐算法给出了全面的、系统的分类,在此基础上比较了各类隐式反馈的推荐方法的优、缺点,并进一步分析了适用于隐式反馈推荐方法的多种评价指标;最后讨论了面向隐式反馈推荐方法的未来发展方向。

As an effective approach addressing information overloading problem, recommender system has been a hot- spot in both industry and academia, which infers users＇ potential requirements and interests by utilizing their explicit/ implicit feedbacks, and then recommends them with preferable information or products. The recommendation methods based on explicit feedbacks are the mainstream approaches in this area, however, because of the prevalence of implicit feedbacks,the recommendation methods for implicit feedbacks can be more widely applied. Because the implicit feed- backs cannot reflect users＇ preferences directly, to recommend products relying on implicit feedbacks is a more challen- ging task. The characteristics of implicit feedback, necessity and problems of recommendation for implicit feedbacks were illustrated firstly. Then a systematic taxonomy for various recommendation methods on implicit feedback was pro- posed. On this basis, the strength/weakness of these approaches and evaluation metrics for implicit feedback oriented recommendation were analyzed. Lastly, the possible directions of implicit feedback oriented recommendation in the fu- ture were discussed.