Abstract: The current tag recommendation methods mainly use the number of tags which appear in the object to represent the relationship among user, tag and item. Using the information of tags in this way is too simple that the precision and recall of the final recommendation result are relatively low. This paper proposes a personalized tag recommendation method using graph model, which converts the relationship among user, tag and item to an undirected tripartite graph. An integrated weight measure is used to compute the adjacent vertices in the graph, while the nonadjacent vertices adopt the thought of shortest path. Tag recommendation method considers not only the relationship between tag and user, but also tag and item. This method is compared to the existing algorithms. The experiment data is collected from CiteULike. The experiment shows that the method can significantly improve the performance of recall and precision.

Key words: social tagging, tag recommendation, graph model, shortest path

摘要： 现有的标签推荐方法大多根据标签在对象中出现的次数来表示用户，标签与资源之间的关系。这种方法对标签信息的利用过于简单，导致最终的推荐结果的准确度和召回率不高。基于这个问题，提出一种采用图模型的个性化标签推荐方法，将用户、标签和资源三者的关系转换成一个三元无向图。对图中相邻顶点的处理采用一种综合的权重衡量方法，而不相邻顶点的关系采用最短路径思想得出。既考虑标签与用户的关系，又考虑标签与资源的关系给出最后的标签推荐方法。将该方法与现存的标签推荐方法做比较。实验采用的数据来自CiteULike。实验结果表明，该方法能够显著地提高推荐结果的召回率，准确性等。

关键词: 社会化标注, 标签推荐, 图模型, 最短路径