Abstract: Social recommendation alleviates the data sparse problem in recommendation to some extent, while it usually only involves the local influence between neighbors. Taking full account of local and global influence, this paper proposes a social recommendation algorithm based on a user influence walk model. The algorithm first calculates the local influence based on neighbors’ trust relations and users’ historic behaviors, and explores the global influence by measuring users’ quality of rating. Then, exploit local and global influence together to compute the transition probability between each node in the random walk model. A lot of experiments is done based on RMSE（Root Mean Squared Error）, coverage rate and F-Measure, the results show that the proposed algorithm improves performance for recommendation in some degree.

Key words: local influence, global influence, random walk model, social recommendation, collaborative filtering

摘要： 社会化推荐在一定程度上缓解了推荐中的数据稀疏性问题，但是通常仅考虑了社交网络中用户间的局部影响关系。综合考虑用户的局部影响力和全局影响力，提出了基于用户影响力游走模型的社会化推荐算法，该算法根据用户信任关系和历史行为分析用户的局部影响力，通过评估用户的评分质量研究用户的全局影响力，然后将二者有机结合计算随机游走模型中各节点之间的转移概率。通过与以往的算法在均方根误差、覆盖率和F-Measure等指标的实验结果表明，提出的算法在一定程度上提高了推荐的性能。

关键词: 局部影响力, 全局影响力, 随机游走模型, 社会化推荐, 协同过滤