摘要
针对目前协同过滤方法存在的数据稀疏性、冷启动以及未能有效利用用户社交网络信息提高推荐质量等问题,提出一种融合用户社交网络信息的协同过滤方法,该方法以矩阵分解推荐模型为核心,可综合集成目标用户个人偏好以及社交网络中的关系用户偏好特征信息做出推荐.通过设计相应的推荐方法,并基于梯度下降法对用户以及商品特征矩阵的求解进行了优化运算.相关实验结果表明融合社交网络信息可在一定程度上提高协同过滤的推荐准确度以及缓解数据稀疏性、冷启动问题.
Collaborative filtering suffers the following problems: sparse data, cold start and ignoring the effect of the social network information among users. Aiming at improving the quality of collaborative filtering, A novel collaborative filtering method with social network information is proposed, which based on matrix factorization recommendation model and can fuse target user's and his friends' preference feature information in social network to make recommendation. The corresponding recommendation framework is designed and can factor the user-item feature matrix with optimized operation based on gradient descend method. Related experiment results show that our method can improve the accuracy of collaborative filte- ring and abate the problem of sparse data and cold start.
出处
《暨南大学学报(自然科学与医学版)》
CAS
CSCD
北大核心
2013年第3期243-248,252,共7页
Journal of Jinan University(Natural Science & Medicine Edition)
基金
国家自然科学基金项目(61272067)
国家科技支撑计划项目(SQ2011GX07E01500)
广东省自然科学基金团队研究项目(S2012030006242)
广东省重大科技专项计划项目(2012A080104019)
广东省高校优秀青年创新人才培养计划项目(2012LYM_0077)
关键词
协同过滤
社交网络
矩阵分解
梯度下降法
推荐系统
collaborative filtering
social network
matrix factorization
gradient descend meth-od
recommendation system