作　　者： (李倩）; (李诗瑾）; (徐桂琼）;

机构地区： 上海大学管理学院,上海200444

出　　处： 《计算机应用研究》 2017年第10期2905-2908,共4页

摘　　要： 针对传统协同过滤算法面临数据稀疏、忽略用户时间上下文信息及对兴趣物品偏好程度等问题,提出基于谱聚类与多因子融合的协同过滤推荐算法。首先将FCM聚类融入到谱聚类算法的关键步骤,并通过聚类有效性指数对用户聚类个数进行优化,以降低生成最近邻的时耗;然后将Salton因子、时间衰减因子、用户偏好因子进行融合,从而对相似度进行改进;最后获取系统当前时间为目标用户生成推荐列表。Movie Lens上的实验结果表明,该算法在推荐精度、覆盖率及新颖度指标上有较大改善,提升了推荐性能。 Due to the problems of traditional collaborative filtering recommendation algorithm, included the data sparsity, ignored the users＇ time context information and preference for interest items, this paper proposed a collaborative filtering recommendation algorithm based on spectral clustering and multiple factors. Firstly, it integrated FCM into the key step of the spectral clustering, and determined the cluster number via cluster validity index, which could reduce the cost to generate a set of the nea- rest neighbors. Then, it improved the similarity measure by combing the Salton factor, time decay factor and user pre-ference factor. Finally, it generated the recommendation list for the objective user combining the system＇ s current time. The experimental results on MovieLens show that the proposed algorithm improves recommendation quality in accuracy, cove-rage and novelty.

关 键 词： 协同过滤 谱聚类 因子 时间衰减因子 用户偏好因子