融合用户隐含偏好的社会化推荐算法

Social Recommendation Algorithm Integrating Users’ Implicit Preferences

协同过滤算法的基本思想是利用兴趣相投、拥有共同经验之群体的喜好来推荐用户感兴趣的信息.目前大部分算法对于相似用户的分析都是基于用户的显式偏好,没有对用户的隐含偏好进行分析与利用.用户的偏好不仅仅体现在对产品种类的喜好上,对于产品各个属性的喜好程度、评分偏好和由偏好相似而建立的信任关系等,都反映了用户在交互时所隐含的偏好.本文提出了一种融合用户隐含偏好的社会化推荐算法:通过对评分矩阵进行分解得到用户和产品的潜在特征向量,利用用户的潜在特征向量进行用户隐含属性偏好相似度的计算;为了缓解推荐系统中常见的冷启动问题,本文引入了信任关系,并将其与评分信息相联系,量化出带有用户偏好的信任关系并将其融入到算法模型之中;最后,使用动态的权重计算用户间的推荐权重.该算法在FilmTrust和Epinions数据集上进行了测试和对比,结果证实了该算法能够更加有效地预测用户评分,提高推荐精度.

Collaborative filtering is a method of making automatic predictions about the interests of a user by collecting preferences or taste information from the users which have similar interests or share same experiences. At present most of algorithms for similar user analysis are based on user’s explicit preference,lacking analysis user’s implicit preference. The user’s preferences not only embody in the product categories,preferences of product attributes,rating preferences and trust relationship based on preference similarity but also show users’ preferences implicitly. In this paper,we propose a recommendation algorithm integrating users’ implicit preferences. First,we factorize the rating matrix to get user latent feature vector and item latent feature vector,then use the user latent feature vector to calculate the similarity of preferences between users. In addition,in order to solve the common problem of cold start in recommendation system,this algorithm introduce trust relationship. We fully explore the connection between rating information to measure the trust relationship with user preference and integrate it into the algorithm model. After that,dynamic weight is used to balance the recommendation weight among users. The algorithm is tested and compared with other recommendation algorithms on FilmTrust and Epinions datasets. Experiments showit gains a higher accuracy of recommendation.