一种融合信任度和相似度的推荐算法

A method to incorporate trust and similarity in recommender systems

推荐系统是一种能够帮助用户在面对大量信息时,能快速、有效地获取有用资源的工具。协同过滤是目前广泛使用的一种推荐技术,该技术通过相似邻居对项目的评分为源用户产生推荐,但面临数据稀疏性和冷启动的问题。基于信任模型的推荐系统虽然在一定程度上缓解了上述问题,却仍然需要进一步提高。针对这些困难,提出了一种融合了信任度和相似度的算法。该算法利用用户间的信任信息,将源用户的信任邻居对项目的评分作为该用户的个人喜好,同时根据基于物质扩散的协同过滤算法找出源用户的相似邻居,利用信任邻居和相似邻居为该用户产生推荐。在2个真实数据集上的实验结果显示,融合算法对冷启动用户的准确性比协同过滤算法分别提高了19%和37%,覆盖率分别提高27%和42%。

Recommender system is a tool to help users who are faced with a large amount of information to gain useful re- source quickly and effectively. Collaborative filtering is a widely used technique to provide recommendations based on ratings of similar users. But it still suffers from two main problems that are data sparsity and cold start. Although the trust-aware recommender system can alleviate above problems, but it still need to be further improved. To address these issues, an algorithm merging trust and similarity is presented. The algorithm uses the trust information between users and incorpo- rates ratings of source user＇ s trusted neighbors to represent the preference of the source user, then find the similar users ac- cording to the massive diffusion collaborative fihering for generating recommendations. Finally, the recommendation is gen- erated for source users through the trust and similar users. Experimental results on two real dataset show that the accuracy of merge method for cold star are improved by 19% and 37% and coverage are increased by 27% and 42%.