Abstract: Collaborative filtering is used extensively in personalized recommendation systems. With the development of E-commence, the magnitudes of users and commodities grow rapidly, resulting in the extreme sparseness of user rating data. To address the problem a collaborative filtering recommendation algorithm based on naive Bayesian method was proposed. The algorithm used improved weighted Bayesian method to predict the rating of unrated items. Through predicting unrated data, the sparseness of rating data problem had been alleviated and the accurate degree of searching nearest neighbor items had been improved simultaneously. The experiment shows that the measure provides better recommendation results for the system.

Key words: Collaborative filtering, Recommendation system, Naive Bayesian, Mutual information, Mean absolute error

摘要： 随着电子商务系统用户和商品数目的不断增加,导致整个项目空间上的用户评分数据极端稀疏,严重影响推荐系统的推荐质量。针对这一问题,提出了一种基于朴素贝叶斯方法的协同过滤推荐算法,采用改进的加权朴素贝叶斯方法对没有评分的数据进行预测。通过对未评分数据进行预测,缓解了数据稀疏性,提高了最近邻居项目搜索的准确度。实验结果表明,该算法在一定程度上提高系统的推荐质量。

关键词: 协同过滤, 推荐系统, 朴素贝叶斯, 互信息, 平均绝对误差