摘要
随着社交网络的飞速发展引起了人们对推荐系统(RS)的广泛关注。针对社交网络中现有推荐方法仍存在冷启动问题以及未考虑用户所处的社交网络信息的情况,提出了在信任社交网络中基于图熵的个性化推荐算法(PRAGE)。首先,根据用户物品和它们之间的反馈信息建立用户物品图(UIG),同时引入信任机制建立用户信任图(UTG);其次,通过对两个图使用随机游走算法得到用户与物品的初始相似度和基于信任机制的新的用户物品相似度;重复随机游走过程直至相似度稳定到收敛值;然后,使用UIG和UTG的图熵对两组相似度进行加权并最终相应地得出目标用户的最终推荐列表。在真实的数据集Epinions和Film Trust上的实验结果表明,相比经典的基于随机游走算法,PRAGE的精确率分别提高了34. 7%和19. 4%,召回率分别提高了28. 9%和21. 1%,能够有效地缓解推荐的冷启动问题且在精确率和覆盖率指标上均优于对比算法。
Widespread attentions have been drawn to Recommendation Systems( RS) as rapid development of social networks. To solve the cold-start problem and neglect to user’s social network information in current recommendation algorithms, a novel Personalized Recommend Algorithm based on Graph Entropy( PRAGE) in trust social network was proposed. Firstly, a weighted User-Item Graph( UIG) was built based on feedback information, and a trust mechanism was introduced to establish a User Trust Graph( UTG). Secondly, by using random walk algorithm on two graphs, the initial similarity of user and item and new user-item similarity based on trust mechanism were obtained; the above algorithm process was repeated until the similarity value reaches convergence value. Then, a method to weight two sets of similarity values with graph entropies of both UIG and UTG was proposed and final recommendation list was accordingly created. The experimental results on two real-world datasets named as Epinions and Film Trust reveal that, compared with classical Random Walk algorithm, the accuracy of PRAGE is increased by about 34. 7% and 19. 4% respectively, and its recall is increased by28. 9% and 21. 1% respectively, which can alleviate cold start problem effectively and has better performance in accuracy and coverage.
作者
蔡永嘉
李冠宇
关皓元
CAI Yongjia;LI Guanyu;GUAN Haoyuan(College of Information Science and Technology,Dalian Maritime University,Dalian Liaoning 116026,China)
出处
《计算机应用》
CSCD
北大核心
2019年第1期176-180,共5页
journal of Computer Applications
基金
国家自然科学基金面上项目(61371090)~~
关键词
社交网络
信任机制
随机游走
图熵
推荐算法
social network
trust mechanism
random walk
graph entropy
recommendation algorithm
