Abstract: To address the problem of item cold-start in collaborative filtering systems, this paper advances a new method that using content based prediction before collaborative filtering to get the predictive ratings of items for users. It points out two fine grained parameters to guarantee the accuracy of the predictions. After the content based filtering, collaborative filtering algorithm is used to generate predictions for users. Experimental results show that the method is superior than traditional collaborative filtering algorithm in the coverage which indicates that the item cold-start problem is alleviated.

Key words: collaborative filtering, cold-start, content based prediction, hybrid recommendation

摘要： 研究协同过滤推荐系统中的冷启动问题，运用基于内容预测的方法，对系统内未被用户评价过的项目进行评分预测，应用2种优化步骤，过滤掉预测不准确的用户的评分。在此基础上用协同过滤的方法产生推荐，使传统推荐算法中无法推荐给用户的项目得到推荐机会。通过一系列实验证明，该混合推荐算法能保证推荐准确性，提高了新项目的推荐概率。

关键词: 协同过滤, 冷启动, 基于内容的预测, 混合推荐

CLC Number: 

• TP391