Abstract: User-based collaborative filtering algorithm find a set of similar users by analyzing user behavior, and its core is to establish interest model and calculate similarity between the users. Traditional user recommend algorithm produces personalized recommendation based on user ratings, or the information of items and other user behavior data, the accuracy rate of it is relatively low. Giving full consideration to online reviews for the influence of similarity between user interests, the accurate user interest model is built by the sentiment analysis of reviews, if the higher the degree of similarity the user manifested in the reviews, it means more similar interest among users. Experimental results show that, compared to the traditional user-based collaborative filtering recommendation algorithm, the accuracy and the recall are both significantly improved in collaborative filtering recommendation algorithm based on the sentiment analysis of reviews.

Key words: collaborative filtering, online reviews, interest model, sentiment analysis

摘要： 基于用户的协同过滤推荐算法是通过分析用户行为寻找相似用户的集合，其核心是用户兴趣模型的建立以及用户间相似度的计算。传统的用户推荐算法是根据用户评分或者物品信息等行为数据进行个性化推荐，准确率比较低。充分考虑在线评论对于用户之间兴趣相似度的作用，通过对评论的情感分析，构建准确的用户兴趣模型，若用户在评论中表现出来的相似度越高，则表示用户之间的兴趣越相似。实验表明，和传统的基于用户的协同过滤推荐算法相比，基于评论情感分析的协同过滤推荐算法，无论准确率还是召回率都有明显提高。

关键词: 协同过滤, 在线评论, 兴趣模型, 情感分析