基于用户相似度的随机游走社交网络事件推荐算法

针对基于社交网络的事件推荐覆盖性和准确性不够高的问题，提出了基于用户相似度Si-user Walker算法.该算法利用基于事件的社交网络特征，将线上用户群组数据抽象为图，以重启随机游走算法为基础，改变了传统的完全基于图的拓扑结构进行随机游走的策略.根据地理位置划分事件类型，提出了新的用户相似度计算方法，然后根据用户相似度矩阵作为随机游走的转移概率，既保留了图的传递性，又保证了图节点游走的真实性.与其他推荐算法在真实的数据集上实验表明，该算法在均方根误差、准确率及覆盖率上均得到提升.

Social Network Event Recommendation Algorithms Based on User Similarity Random Walk

Aimed at the problem of insufficient coverage and accuracy of event recommendation based on social network， a user similarity-based Si-user Walker algorithm was proposed. The algorithm abstracts online user group data into graphs through event-based social network features and changes the traditional strategy of random walk based on graph-based topology structure on the basis of restart random walk algorithm. According to geographical location， event types were divided and a new calculation method of user similarity was proposed. Then， the user similarity matrix works as transition probability of random walk. Transfer probability not only preserves the transitivity of graph， but also guarantees the authenticity of graph node walk. Compared with other recommendation algorithms， it is shown that the proposed algorithm experiments on real data sets can improve root mean square error， accuracy and coverage.