高级检索
    郭弘毅, 刘功申, 苏波, 孟魁. 融合社区结构和兴趣聚类的协同过滤推荐算法[J]. 计算机研究与发展, 2016, 53(8): 1664-1672. DOI: 10.7544/issn1000-1239.2016.20160175
    引用本文: 郭弘毅, 刘功申, 苏波, 孟魁. 融合社区结构和兴趣聚类的协同过滤推荐算法[J]. 计算机研究与发展, 2016, 53(8): 1664-1672. DOI: 10.7544/issn1000-1239.2016.20160175
    Guo Hongyi, Liu Gongshen, Su Bo, Meng Kui. Collaborative Filtering Recommendation Algorithm Combining Community Structure and Interest Clusters[J]. Journal of Computer Research and Development, 2016, 53(8): 1664-1672. DOI: 10.7544/issn1000-1239.2016.20160175
    Citation: Guo Hongyi, Liu Gongshen, Su Bo, Meng Kui. Collaborative Filtering Recommendation Algorithm Combining Community Structure and Interest Clusters[J]. Journal of Computer Research and Development, 2016, 53(8): 1664-1672. DOI: 10.7544/issn1000-1239.2016.20160175

    融合社区结构和兴趣聚类的协同过滤推荐算法

    Collaborative Filtering Recommendation Algorithm Combining Community Structure and Interest Clusters

    • 摘要: 传统的协同过滤推荐算法受限于数据稀疏性问题,导致推荐结果较差.用户的社交关系信息能够体现用户之间的相互影响,将其用于推荐算法能够提高推荐结果的准确度,目前的社交化推荐算法大多只考虑了用户的直接社交关系,没有利用到潜在的用户兴趣偏好信息以及群体聚类信息.针对上述情况,提出一种融合社区结构和兴趣聚类的协同过滤推荐算法.首先通过重叠社区发现算法挖掘用户社交网络中存在的社区结构,同时利用项目所属类别信息,设计模糊聚类算法挖掘用户兴趣偏好层面的聚类信息.然后将2种聚类信息融合到矩阵分解模型的优化分解过程中.在Yelp数据集上进行了新算法与其他算法的对比实验,结果表明,该算法能够有效提高推荐结果的准确度.

       

      Abstract: Traditional collaborative filtering recommendation algorithms suffer from data sparsity, which results in poor recommendation accuracy. Social connections among users can reflect their interactions, which can be mixed into recommendation algorithms to improve the accuracy. Only straightforward social connections have been used by most current social recommendation algorithms, while users’ latent interest and cluster information haven’t been considered. In response to these circumstances, this paper proposes a collaborative filtering recommendation algorithm combining community structure and interest clusters. Firstly, overlapping community detection algorithm is used to detect the community structure existed in user social network, thus users in the same community have certain common characteristics. Meanwhile, we design a customized fuzzy clustering algorithm to discover users’ interest clusters, which uses item-category relationship and users’ activity history as input. Users in the same cluster are similar in generalized interest. We quantify users’ preference for each social community and interest cluster they belong to respectively. Then, we combine this two types of user group information into matrix factorization model by adding a clustering-based regularization term to improve the objective function. Experiments conducted on the Yelp dataset show that, in comparison to other methods including both traditional and social recommendation algorithms, our approach gets better recommendation results in accuracy.

       

    /

    返回文章
    返回