一种结合聚集图嵌入的社会化推荐算法

社交网络信息已被广泛的应用到传统的推荐上,一定程度上减轻了数据稀疏和冷启动问题.随着表示学习的兴起,出现了利用表示学习进行推荐的算法研究.然而社交网络过大,表示学习可扩展性差,难以在有限内存中进行计算.聚集图通过空间压缩,保留了关键的结构关系,去除次要或噪音的结构数据,便于表示学习能够有效学习图结构,从而更好地找到相似用户进行推荐.首先,利用图聚集算法同时考虑分组间及分组内的结构得到最终的聚集图;其次,在聚集图上计算随机游走的转移概率,然后选择每个具有偏差概率的后继节点并生成节点序列;最后将节点序列输入到skip-gram学习用户的潜在表示,获得节点的表示向量整合其信息到贝叶斯个性化排序模型(BPR)来解决项目排名问题.实验结果表明,该方法相比于社会化贝叶斯个性化排序(SBPR)、协同用户网络嵌入(CUNE)等基线方法在推荐任务中保持时间效率的同时有效提升了准确率、召回率和平均精度均值.

Social Recommendation with Embedding of Summarized Graphs

In recent years,social network information has been widely applied to traditional recommendation,which alleviates the problem of data sparsity and cold start to a certain extent.With the rise of representation learning,there are researches on algorithms for recommendation using representation learning.However,the social network is too large,indicating that the learning scalability is poor and it is difficult to perform calculations in limited memory.Aggregation graphs are compressed through space,retaining key structural relationships,removing minor or noisy structural data,which is convenient for representation learning to effectively learn graph structure,and better find similar users for recommendation.First,using the graph aggregation algorithm to consider the structure betw een and w ithin groups to obtain the final aggregation graph.Second,calculating the random walk transition probability on the aggregation graph,and then selecting each subsequent node with a bias probability and generate a node sequence;Finally,the node sequence input to the skip-gram to learn the potential representation of the user,and the node’s representation vector is obtained to integrate its information into the Bayesian personalized ranking model(BPR)to solve the problem of item ranking.Experimental results show that this method can effectively improve accuracy,recall and average compared with baseline methods such as socialized Bayesian personalized ranking(SBPR)and collaborative user netw ork embedding(CUNE)while maintaining time efficiency in recommendation tasks.