用户情感分类和话题推荐的图片语义应用实证

User/subject recommendation with sentiment classification:an empirical analysis of integrating image sentiment

用户/话题推荐的传统方法多基于用户热度,忽视了用户希望寻找志同道合伙伴的心理需求,推荐效果存在改进空间.尝试将融合图片语义后的微博情感分类结果用于计算,利用情感一致性和同意度,从而得出用户整体情感影响力和话题影响力.在此基础上结合聚类算法,计算并排序用户间的相对影响力,构建情感网络拓扑用于用户/话题推荐,推荐匹配特定用户的观点、态度和情感,提高推荐结果对用户的吸引力.实验结果表明,方案相比主流推荐算法,用户关注数、微博数等结果指标有一定提高.

Most user/recommendation researches focused on optimization of user-popularity-based algorithms while ignoring the psychological needs of users who want to find like-minded,and there is room for improvement in recommendation performance.Attempts are made in this paper with sentiment classification results of microblog based on integration of image sentiment to find out the emotional accordance and agreement degree scores,thus obtaining overall emotional influence levels and topic influence levels for different users,which are utilized in combination of clustering algorithms to figure out the relative influence levels between specific user pairs and to construct sentimental network topology.Then the results are used for user/subject recommendations to match the opinions,attitudes and emotions of specific users,improving the attractiveness of recommendation results.Experimental results show that our recommendation solution holds certain advantages in the aspects of follower count and microblog count,etc.