计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (16): 130-137.DOI: 10.3778/j.issn.1002-8331.1812-0400

• 模式识别与人工智能 • 上一篇    下一篇

融合时间隐语义填充和子群划分的推荐算法

王永康,袁卫华,张志军,温鹏   

  1. 山东建筑大学 计算机科学与技术学院,济南 250101
  • 出版日期:2019-08-15 发布日期:2019-08-13

Recommendation Algorithm Integrating Time Based Latent Semantic Completion Model with Subgroup Partitioning

WANG Yongkang, YUAN Weihua, ZHANG Zhijun, WEN Peng   

  1. School of Computer Science and Technology, Shandong Jianzhu University, Jinan 250101, China
  • Online:2019-08-15 Published:2019-08-13

摘要: 随着个性化推荐技术的发展,推荐系统面临着越来越多的挑战。传统的推荐算法通常存在数据稀疏性和推荐精度低等问题。针对以上问题,提出了一种融合时间隐语义填充和子群划分的推荐算法[K]-TLFM(Time Based Latent Factor Model Integrated with [k]-means)。该算法利用融合时间因素的隐语义模型对原始用户物品评分矩阵缺失项进行填充,避免了用全局平均值或者用户/物品平均值补全矩阵带来的误差,有效缓解了数据稀疏性问题,同时融合时间因素有效地刻画了用户偏好随时间的变化;完成评分矩阵缺失项填充后,基于二分[k]-means聚类算法将偏好、兴趣特征相似的对象划分到同一个子群中,在目标用户所属的子群中基于选定的协同过滤算法为用户产生推荐列表,提高了推荐效率和准确性。在MovieLens和Netflix数据集上对该算法的推荐性能进行了对比实验,结果表明该算法具有更高的推荐精度。

关键词: 时间因素, 隐语义模型, 聚类算法, 个性化推荐

Abstract: With the development of personalized recommendation technologies, recommender systems are facing increasingly more challenges. Traditional recommendation algorithms usually have the problems of data sparsity and low recommendation accuracy. To solve the above problems, this paper proposes a recommendation algorithm [K]-TLFM(Time Based Latent Factor Model Integrated with [k]-means), which combines time effect based latent semantic model with clustering methods for subgroup partitioning. Firstly, the missing items of the original rating matrix are filled by the proposed time based latent semantics model. This alleviates the errors caused by the filling with global average or user/item average, and effectively alleviates the problem of data sparsity. Meanwhile, the fusion of the time factor efficiently depicts the variations of user preferences over time. Secondly, after the completion of the missing items in the scoring matrix, items with similar preferences and interest characteristics are partitioned into the same clusters based on the bisecting [k]-means algorithm. There commendation lists are generated for targeted users by the selected collaborative filtering algorithms according to their subgroups. Lastly, this paper carries out several experiments to test the performance of the proposed [K]-TLFM with other state-of-the-art recommendation algorithms on the datasets of MovieLens and Netflix. The experimental results show both the efficiency and higher recommendation accuracy of the proposed algorithm.

Key words: time factor, latent factor model, clustering algorithms, personalized recommendation