Abstract: Similarity calculation method in traditional collaborative filtering is inaccuracy due to the extreme sparsity of user rating data. To address this problem, a collaborative filtering algorithm based on item classification and user group interest is proposed. The algorithm first classifies items by item classification, and then calculates the user similarity matrix and user group interest of each item classification by rating data to construct the user weighted similarity matrix. It obtains the nearest neighbors of target user by user weighted similarity matrix, and generates recommendations. The experimental results show that this algorithm can effectively improve the quality of the recommendation of recommender systems, and get better recommend results.

Key words: recommender systems, collaborative filtering, item classification, user group interest, user weighted similarity

摘要： 由于用户评分数据在极端稀疏的情况下会导致传统协同过滤算法的推荐质量下降，针对该问题，提出一种基于项目分类和用户群体兴趣的协同过滤算法。该算法根据项目类别信息对项目进行分类，相同分类的项目具有较高的相似性；利用评分数据计算各个项目分类上的用户相似性矩阵，并计算用户群体在各个分类上的兴趣，通过二者构造加权的用户相似性矩阵；利用用户加权相似性矩阵寻找用户的最近邻以获得最佳的推荐效果。实验结果表明，该算法能有效提高推荐质量。

关键词: 推荐系统, 协同过滤, 项目分类, 用户群体兴趣, 用户加权相似性