| 摘 要: 为了改善经典协同过滤个性化推荐算法准确性不高、推荐列表排序性能和偏好分类预测精确性低的缺点，综合考虑了用户共同评分物品数量及质量，通过降低共同评分物品数量较少或质量不高的邻居用户权重，设计了一种用户相似性度量方法进行协同过滤和评分预测。通过实验，在MovieLens电影评分数据集上与领域内几个典型协同过滤算法对比，研究发现本文设计的融合用户共同评分数量和质量的协同过滤个性化推荐算法可以将预测误差降低8.41%，将推荐列表排序性能提高10.21%，将偏好分类预测准确率提高2.55%。 |
| 关键词: 推荐算法；评分预测；相似性；共同评分 |
|
中图分类号: TP182
文献标识码: A
|
| 基金项目: 国家自然科学基金项目(61803264). |
|
| A Collaborative Filtering Personalized Recommendation Algorithm Integrating the Quantity and Quality of Common Rating |
|
CAI Yifang, AI Jun, SU Zhan
|
(School of Optical -Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China )
worldchinacai@126.com; aijun@outlook.com; suzhan@foxmail.com
|
| Abstract: Traditional collaborative filtering personalized recommendation algorithm has the shortcomings of low accuracy, low ranking performance of recommendation list and low accuracy of preference classification prediction. Aiming at these problems, this paper proposes to design a user similarity measurement method for collaborative filtering and rating prediction by reducing the weight of neighbor users with small number of common rating items or low quality of common rating, after comprehensively considering the quantity and quality of users' common rating items. Through experiments, compared with several typical collaborative filtering algorithms in the field on MovieLens rating dataset, it is found that the proposed collaborative filtering personalized recommendation algorithm, which incorporates the quantity and quality of user co-ratings, can reduce the prediction error by 8.41%, improve the performance of recommendation list sorting by 10.21%, and improve the accuracy of preference classification prediction by 2.55%. |
| Keywords: recommendation algorithm; rating prediction; similarity; common rating |
