【摘要】 在大众分类网络中,允许用户使用个性化标签对资源进行标注,标签可以使用户方便地表达的自己的兴趣与偏好.但是,标签自由、松散的分类方式使标签存在冗余、歧义以及一词多义的问题,使用户难以发现自己需要的资源,因此在基于标签的推荐系统中,推荐精确性低,用户体验差,社区发现(聚簇)技术是解决这一问题的重要手段.本文从构建标签共现图入手,采用标签共现图的重叠社区发现技术来理解标注的正确含义、减少冗余歧义标签带来的噪声.在此基础上设计了完整的个性化推荐方案,经过真实标签网络数据验证表明标签重叠社区检测能够提高推荐质量,算法在精确性和多样性上均有较好的改进.更多还原

【Abstract】 In folksonomy based networks,users are allowed to annotate resources with personalized tags,which can facilitate users conveying the user’s interest and preference information.However,this flexibility and loosening method of classification brings with it certain costs: redundant,ambiguous and polysemy,which can render resource discovery difficult.So,in tag-based recommendation,the recommended result of precision and diversity is low and has a poor user experience.Communities detection(clustering) provides a means to remedy these problems.Starting from a tagging co-occurrence network,we leverage overlapping communities detection method in tagging co-occurrence network to comprehend the proper meaning of the tags and reduce tagging noise.Based on overlapping communities detection,a complete scheme of personalized recommendation was presented.We validate this approach through evaluation of proposed personalization algorithm using data from a real collaborative tagging Web site,the result demonstrates that overlapping communities detection could considerably improve the precision and diversity of recommendations.更多还原