融合信息熵和加权相似度的协同过滤算法研究

Research on Collaborative Filtering Algorithm Based on Information Entropy and Weighted-similarity

针对传统协同过滤算法存在的稀疏性问题以及只考虑用户间共同评分项目数量而忽略具体评分值的问题,提出了融合信息熵和加权相似度的算法模型。该模型在信息熵的基础上引入了差异信息熵,即考虑到用户对共同评分项目的具体评分值的影响,有效缓解了由于原始评分数据稀疏而造成的推荐质量欠佳问题。另外在此基础上,通过引入调节因子将传统的相似度计量方法和差异信息熵通过加权平均,使用加权相似度计算用户之间的相似性,获得更好的近邻,提高了发现邻居的精确度。最后将该算法应用于Movie Lens数据集,并与传统的协同过滤算法进行比较,结果表明,该算法具有更好的推荐效果,证明了该模型的有效性和可行性。

Aiming at the problem of sparsity and only considering the number of common items to score between users and ignoring the specific score in traditional collaborative filtering, we propose a new algorithm model based on information entropy and weighted similari- ty, where we introduce the difference information entropy based on information entropy, which mainly consider the influence of specific score to common items, effectively alleviating the poor recommendation equality caused by sparse original rating matrix. In addition, by introducing the adjusted factor for weighted average for traditional similarity measurement method and the difference information entropy, we apply the weighted similarity to compute the similarity between the users and get better neighbors, to improve the precision of the neighbors. Last,it is applied to MovieLens dataset and compared with traditional collaborative filtering algorithm, and the experiments show that we can get better recommendation results, proving its effectiveness and feasibility.