基于用户认知水平的分类别推荐

Classification Recommendation Algorithm Based on User Cognitive Level

在线社会网络的出现,为进一步提高推荐质量提供了新的可能。基于信任的推荐不仅考虑了用户兴趣相似度,而且将信任关系融入推荐中,但却没有区分二者在推荐中的不同作用。通过对现实生活中的普遍规律及社会网络评分数据的分析,发现用户认知水平是影响信任关系和兴趣相似度在推荐中的作用的关键因素。给出用户认知水平的度量方式,并提出一种基于用户认知水平的分类别推荐。通过真实数据的实验表明,该算法较比传统协同过滤及基于信任的推荐可以获得更优的推荐质量。

Abstract. Because of the emergence of online social networks, it is possible to improve the quality of recommendation. Not only interest similarity between users is considered intrust-based recom- mendation, but also the user＇ strust relations. However the different effect between interest similarity and trust relations is not distinguished in trust-based recommendation. According to the analysisof gen- eral rules in real life, it ＇s found that the influence ofsurroundings is large to ＂cold start＂ users and active usersare relatively insensitive to the distance of the trust. Compared to active users, ＂cold start＂ users prefer the popular items to the less popular throughthe analysis of rating data in social network. Accord- ingly, the cognitive level of users is the key factor in the role of trust relationship and interest similarity in recommendation. Based on above description and discussion, the measurement of user cognitive level is provided. The cognitive level of user has a positive correlation with the number of items rated by the user, and is negatively related with the averagepopularity ofitems rated by the user. Considering the dif- ferenceamonguser cognitive level in multiplecategories, the measurement of user cognitive level isextend to differentcategories. Furthermore, with considerations ofuser cognitive level, interest similarity and trust relationship, akind of classificationrecommender based on user＇ s cognitive level is proposed. Ac- cording to the user cognitive level, interest similarity and trust relationshipplay different roles in the a- bove-mentioned recommendation algorithm. The experiments on real data show that, the recommendation quality of the proposedalgorithm is better than traditional collaborative filtering and trust-based recom- mendation.