基于用户评分权重的矩阵分解推荐算法

Matrix Factorization Recommendation Algorithm Based on User Rating Weight

在传统的推荐算法中存在数据评分稀疏的问题,同时,在建立预测模型时默认每个用户评分都是真实可信的。但实际评分中存在某些数据不符合用户的整体评分趋势和偏好。为了解决上述问题,对每项用户评分的真实性进行计算,在进行评分预测时,使符合用户整体评分趋势的评分数据获得更高的权重,让推荐算法更精准的把握用户和项目的特征信息,提升推荐系统的整体性能。经过在Movie Lens 100k数据集上与其它三种经典算法的对比实验表明,本文提出的改进算法能更好地把握用户真实喜好,提高预测的准确性。

In the traditional recommendation algorithm,there is a problem of sparse data scores.At the same time,when building a prediction model,it is assumed that each user's score is true and credible.However,there are some data in actual ratings that do not conform to the user's overall rating trends and preferences.In order to solve the above problems,this paper calculates the authenticity of each user's rating.When making rating predictions,the rating data that conforms to the user's overall rating trend is given higher weight,so that the recommendation algorithm can more accurately grasp the characteristics of users and items so as to improve the overall performance of the recommendation system.The comparative experiments on the Movie Lens 100k dataset with three other classic algorithms show that the improved algorithm proposed in this paper can better get the real preferences of users and improve the accuracy of prediction.