基于电商用户评分权重协同过滤的推荐系统算法

Recommendation System Algorithm Based on Weighted Collaborative Filtering for E-Commerce User Ratings

摘要: 为解决传统协同过滤算法在用户评分波动性与数据稀疏性下相似度计算失真、推荐效果下降的问题，本研究提出一种基于用户评分信息熵权重的协同过滤方法。该方法利用评分分布的信息熵衡量用户评分的稳定性，并将其作为权重融入相似度计算过程，以减弱高噪声用户对推荐结果的干扰。基于MovieLens数据集的实验表明，与传统协同过滤相比，熵权模型在RMSE、MAE等预测误差上显著降低，同时在Precision@10、Recall@10与MAP@10等Top-K推荐指标上均取得提升。研究结果验证了熵权机制在增强推荐准确性与模型鲁棒性方面的有效性。

Abstract: To address the distortion of similarity computation and the decline in recommendation performance caused by user rating volatility and data sparsity in traditional collaborative filtering algorithms, this study proposes a collaborative filtering method based on users’ rating information entropy weights. The method uses the information entropy of rating distributions to measure the stability of users’ rating behaviors and incorporates it as a weight into the similarity calculation, thereby reducing the influence of high-noise users on recommendation results. Experiments conducted on the MovieLens dataset show that, compared with traditional collaborative filtering, the entropy-weighted model significantly reduces prediction errors such as RMSE and MAE, and achieves improvements in Top-K recommendation metrics, including Precision@10, Recall@10, and MAP@10. The findings validate the effectiveness of the entropy-weight mechanism in enhancing recommendation accuracy and model robustness.