协同过滤的相似度融合改进算法.PDFVIP

2017 年 第 26 卷 第 1 期 计 算 机 系 统 应 用 协同过滤的相似度融合改进算法① 于世彩, 谢颖华, 王 巧 (东华大学 信息科学与技术学院, 上海 201620) 摘 要: 针对传统协同过滤推荐在数据稀疏性条件下性能不佳的问题, 在相似度计算上做出了优化, 提出了一种 基于项目类别和用户兴趣相似度融合的协同过滤算法, 算法将相似度的计算分解为两个方面进行: 用户-项目类 别评分相似度和用户-项目类别兴趣相似度, 将两者用合适的权值加以融合得到最终相似度, 参与最终预测评分 的计算. 利用 MovieLens 公用数据集对改进前后的算法进行对比. 结果表明, 基于项目类别和用户兴趣的协同过 滤改进算法有效地缓解了数据稀疏性问题的影响, 提高了推荐的准确性. 关键词: 协同过滤; 数据稀疏性; 项目类别; 用户兴趣; 相似度融合 Improved Collaborative Filtering Algorithm of Similarity Integration YU Shi-Cai, XIE Ying-Hua, WANG Qiao (School of Information Science and Technology, Donghua University, Shanghai 201620, China) Abstract: Aiming at the poor recommendation quality due to the data sparsity problem of traditional collaborative filtering recommendation, this paper puts forward an improved collaborative filtering algorithm. The improved algorithm proposes a collaborative filtering algorithm based on the similarity integration of item categories and user interests to make optimization on the similarity calculation. The algorithm does not simply concentrate on similarity calculation, but divides it into two aspects: users-item category interest similarity and users-item category rating similarity, which will finally be integrated with appropriate weights to get the final similarity. After a series of verification and comparison carried out on the MovieLens public data set, it is concluded that the improved algorithm based on data sparsity of collaborative filtering indeed plays a positive role in reducing the influence caused by data sparsity and improves the accuracy of recommendation. Key words: collaborative filtering; data sparsity; item category; user interest; similarity integration 互联网的发展带来了便利的同时, 也造成了信息 推荐过程依赖于 3 个要素: 推荐候选对象、用户 量的急速增长和膨胀, 出现了“信息过载”[1]现象. 互联 和推荐方法[2]. 推荐系统使用户摆脱了单向的搜索服 网信息过