计算机与现代化

• 数据库与数据挖掘 • 上一篇    下一篇

基于改进余弦相似度的协同过滤推荐算法

  

  1. (1.中国科学院声学研究所国家网络新媒体工程技术研究中心,北京100190;
    2.中国科学院大学,北京100049)
  • 收稿日期:2019-04-10 出版日期:2020-02-13 发布日期:2020-02-13
  • 作者简介:李一野(1994-),男(蒙古族),内蒙古赤峰人,硕士研究生,研究方向:推荐系统与算法,E-mail: yiyeli@163.com; 邓浩江(1971-),男,研究员,博士,研究方向:多媒体通信和模式识别,E-mail: denghj@dsp.ac.cn。
  • 基金资助:
    中国科学院战略性科技先导专项基金资助项目(XDC02010701)

A Collaborative Filtering Recommendation Algorithm Based on Adjusted Cosine Similarity

  1. (1. National Network New Media Engineering Technology Research Center, Institute of Acoustics, Chinese Academy
    of Sciences, Beijing 100190, China; 2. University of Chinese Academy of Sciences, Beijing 100049, China)
  • Received:2019-04-10 Online:2020-02-13 Published:2020-02-13

摘要: 传统的协同过滤推荐算法存在数据稀疏情况下分类准确性低的问题,针对于此提出一种基于改进余弦相似度的协同过滤推荐算法,将数据经嵌入层转换为特征矩阵,将对其计算后得到的改进余弦相似度矩阵和单位矩阵之间的均方误差作为损失函数,从而提高推荐算法在数据稀疏情况下的分类准确性。实验结果表明,该算法的AUC和对数损失函数指标均优于基线模型FM、FFM和DeepFM模型。

关键词: 改进余弦相似度, 协同过滤, 推荐算法, 深度因子分解机

Abstract: In the traditional collaborative filtering recommendation algorithm, there exists a problem of low accuracy of classification in the case of sparse data. To solve this problem, a collaborative filtering recommendation algorithm based on adjusted cosine similarity is proposed. Converting the data to the feature matrix by the embedded layer, the mean square error between the adjusted cosine similarity matrix obtained by the calculation and the unit matrix is used as the loss function for improving the accuracy of classification of the proposed algorithm in the case of sparse data. Experimental results indicate that, compared to the FM, FFM and DeepFM models, AUC and the Logloss of the proposed collaborative filtering recommendation algorithm based on adjusted cosine similarity are better within the acceptable range of training time.

Key words: adjusted cosine similarity, collaborative filtering, recommendation algorithm, deep factorization machine

中图分类号: