In a collaborative filtering system, for a target user, the potential evaluation of an object is estimated ac-cording to the ratings from users similar to the target user. Thus, the definition of the similarity between users is of significance. Traditional collaborative filtering algorithms do not consider the preference of users when defining the similarity. To conquer this problem, a collaborative filtering algorithm based on bi-clustering is presented. The pref-erences of users are identified by bi-clustering, and the preferential similarity is defined by the algorithm. When the attributive information of users can be acquired, the common features of users sharing the same preference can be extracted. Furthermore, similarity based on attributions is proposed. The recommendations are given by combining the attributive similarity and the rating similarity. Our algorithm is applied to MovieLens data to validate its accuracy. Experiments demonstrate that compared with other methods, our algorithm can deal with extremely sparse data, predict ratings more accurately, and suggest more movies which the users really like in the top part of the recommendation list.%协同过滤利用与目标用户相似性较高的邻居对其他产品的评价来预测目标用户对特定产品的喜好程度,用户间的相似性定义至关重要.传统协同过滤算法定义相似性时不考虑用户偏好,为了解决这一问题,本文提出基于联合聚类的协同过滤算法.该算法利用联合聚类识别用户偏好,定义用户偏好相似性.当可用数据还包括用户的属性信息时,算法提取有共同偏好的用户的公共特征,进一步定义基于属性的相似性,结合属性相似性与打分相似性产生推荐.实验用MovieLens数据验证推荐算法的准确性,实验结果表明本文算法可以处理极度稀疏数据,且预测的打分更加准确,推荐排名靠前的电影更受用户喜爱.
展开▼
