%0 Journal Article %T 一种利用多群组智慧的协同推荐算法<br>A Collaborative Filtering Recommendation Algorithm Using Multiple Groups Intelligence %A 郑修猛 %A 陈福才 %A 吴奇 %A 朱宇航 %A 黄瑞阳 %J 西安交通大学学报 %D 2016 %R 10.7652/xjtuxb201610018 %X 针对当前基于社会网络的推荐系统大多数采用一般的启发式方法,存在节点复杂路径选择和信任弱传递现象导致推荐精确度不高的问题,以及针对推荐系统固有的冷启动问题,提出了一种利用多群组智慧的协同推荐算法。该算法首先根据用户的社会属性和社会信任关系信息进行群组划分,将用户分为多个不同的群组;然后分析群组中用户的社会活动和社会关系等,建立一种利用多群组的评分预测模型,并利用群组评分预测新用户的评分。该算法通过对社会网络进行深层次的群组挖掘,利用多群组智慧可以有效提高推荐效果,利用群组评分可改善对冷启动用户的推荐。仿真实验表明,该算法相比传统的协同推荐算法在效果评分上提高了约0.2,相比其他社会化推荐算法进一步提高了约0？？02,并有效解决了冷启动问题。<br>A collaborative filtering recommendation algorithm using the multiple groups intelligence is proposed to address the problems that most of the current recommendation systems that are based？？on social networks use the general heuristic methods and have drawbacks of choice of complex paths and weak transferring of trust phenomenon which leads to low recommendation precision, and there exists the inherent cold start problem in recommendation systems. The proposed algorithm divides users into several different groups from their social attribution and social trust relationship information Then predictive models are built based on multi？？group through analyzing the user’s social activities and social relationships in the groups and the evaluated group scores are used to predict ratings of new users The algorithm uses a deep group mining in a social network, the multi group intelligence to effectively improve the effect of the recommendation, and the evaluated group score to improve the recommendation of the cold start users. Simulation results and comparisons with the traditional collaborative recommendation algorithm and other social recommendation algorithms show that the recommendation effects of the proposed algorithm improve by about 0.2 and 0.22, respectively, and that the problem of cold start is effectively solved %K 协同推荐 %K 社会网络 %K 群组智慧 %K 冷启动< %K br> %K collaborative filtering recommendation %K social network %K group intelligence %K cold start %U http://zkxb.xjtu.edu.cn/oa/DArticle.aspx?type=view&id=201610018