融合标签和多元信息的个性化推荐算法研究

多数基于标签的推荐算法都存在推荐方式单一的问题,没有充分利用社会关系等其他信息。针对这一问题,提出了一种融合信任关系、时间因子和标签信息的个性化推荐算法TTLMF,该算法在现有基于标签的个性化推荐算法的基础上,充分利用了用户之间的信任关系和当前上下文的时间信息,使得推荐项目更加符合用户的需求。在公共数据集last.fm上进行了实验,结果表明TTLMF算法在准确率、召回率、Fmeasure以及覆盖率这四个指标上具有更好的推荐效果,一定程度上缓解了数据稀疏性和用户的冷启动问题。

Research on Personalized Recommendation Algorithm Based on Label and Multi-Information

Most of the label-based recommendation algorithms have a problem that the recommend approach is singleness, and do not make full use of other information such as social relations. Aiming at this problem, on the basis of existing algorithms, a matrix factorization personalized recommendation algorithm fusing label popularity, time weight and trust relationship（TTLMF） is proposed. TTLMF on the basis of the existing label-based personalized recommendation algorithm, makes full use of the trust relationship between users and the current context of the time information, which makes the recommended projects more in line with the needs of users. Experimental results in the dataset of Last.fm show that the TTLMF algorithm has a better recommendation effect on four evaluation metrics which are precision, recall, F-measure and coverage, and also alleviates the sparseness of data and the cold start problem of users to a certain degree.