面向场景的协同过滤推荐算法

Context Based Collaborative Filtering Recommendation Algorithm

推荐系统是电子商务系统中最重要的技术之一。用户相似性度量方法是影响推荐算法准确率高低的关键因素,针对传统相似性度量方法存在的不足,利用云模型在定性知识表示以及定性、定量知识转换时的桥梁作用,提出一种在知识层面比较用户相似度的方法,克服了传统基于向量的相似度比较方法严格匹配对象属性的不足。进而以该方法为核心,提出一种面向场景的协同过滤推荐算法,该算法能够充分利用项目的分类信息,避免了传统算法把用户的整体打分作为单个向量的弊端。实验结果表明,算法可以在用户评分数据极端稀疏的情况下,仍能取得较高的推荐质量。

Similarity measuring methods is fundamental to collaborative filtering algorithm while the traditional ways are inefficacy especially when the user rating data are extremely sparse. A new kind of similarity measuring method was proposed, which compares the similarity of two users on knowledge level. Then based on this method, a context based collaborative filtering recommendation algorithm was put forward, which is well integrated with the classification information of items and avoids the shortcoming of treating all voting data of a user as a whole vector. Instead of trying to finding out users’ nearest neighbors on all items, the algorithm searches out classified friends according to each context. Experiments show the excellent performance of this approach.