基于用户关系挖掘的多策略推荐算法

个性化推荐服务为解决网络信息过载问题提供了有效手段。传统的推荐方法大多只关注于如何提高推荐的准确性,而忽略了推荐多样性对用户体验的影响。文章将社会网络用户关系挖掘应用于用户偏好预测及推荐中,提出了一种基于用户关系挖掘的多策略推荐算法。采用信任传播模型挖掘用户间的信任度,计算用户偏好配置文件的余弦相似性获得用户间的相似度,并给出4种将用户信任度、相似度结合的策略,在定义用户偏好预测函数的基础上采用Topn原则为用户给出推荐结果。实验结果表明,文章方法不仅减少了数据稀疏性的影响,而且兼顾了推荐准确性与多样性指标,提高了推荐系统的整体性能。

Multi Strategy Recommendation Algorithm Based on User Relation Mining

Personalized recommendation service provides an effective means to resolve the information overload problem on the Internet. Traditional researches only pay attention to improving the accuracy of recommendation algorithm while neglect the impact of recommendation diversity on user experience. In this paper, relation mining techniques between social network users are applied to user preference prediction and personalized recommendation, and a multi-strategy recommendation algo- rithm based on user relation mining is proposed. The proposed method employs trust propagation model to mining trust relation between users, and calculates user similar relation based on cosine similarity of users＇ profile. Then four combination strategies of users＇ trust and similar relation are proposed. On the basis of defining user preference prediction function, the proposed method pres- ents recommendation results with Top principle. Experimental results confirm that our algorithm not only reduces the impact of data sparsity, but also balances recommendation accuracy with diversity, which improves the overall performance of recommendation system.