摘要
推荐系统可以帮助人们在海量的数据中发现所需的有价值的信息。传统的协同过滤推荐算法根据历史数据中用户对项目的各种行为操作构建用户-项目评分矩阵,进而计算相似度,从而预测用户对项目的偏好程度进行推荐。但因为评分数据通常较为稀疏,使得推荐的准确性不高,从而不能很好地对用户进行推荐。针对这个问题,提出一种结合场论理论的随机游走歌曲推荐算法,融合歌曲评分相似度和歌曲基本信息相似度,降低歌曲间综合相似度矩阵的稀疏性,并将物理学中的场论理论和歌曲的重要度结合,构造转移概率矩阵,从而实现歌曲推荐。实验表明,该算法较协同过滤算法的推荐准确性更佳。
The recommendation system can help people find the valuable information needed in the huge data.The traditional collaborative filtering recommendation algorithm constructs a user-item scoring matrix based on various behaviors of users in the historical data,and then calculates the corresponding similarity,predicts the user′s preference for the project.However,because the scoring data is usually sparse,the accuracy of the recommendation is not high,so the user cannot be recommended very well.Aiming at this problem,this paper proposes a random walk song recommendation algorithm combined with field theory,which combines the basic information similarity and song score similarity of songs,reduces the sparsity of the comprehensive similarity matrix between songs,and combines the theory of field theory in physics with the importance of songs,constructing a state transition matrix to reduce the time complexity of the algorithm,so achieving song recommendation.Experiments show that the proposed algorithm is more accurate than the collaborative filtering algorithm.
作者
吴雪君
米红娟
李欣
Wu Xuejun;Mi Hongjuan;Li Xin(Department of Information Engineering,Lanzhou University of Finance and Economics,Lanzhou 730020,China)
出处
《信息技术与网络安全》
2019年第10期35-39,共5页
Information Technology and Network Security
关键词
推荐算法
协同过滤
随机游走
场论
recommendation
collaborative filtering
random walk
field theory
