基于时序变化与用户聚类的提高总体多样性的方法

Aggregate Diversity Improvement Method Based on Time Series Change and User Clustering

摘要: 针对传统协同过滤算法数据稀疏性、相似度计算片面、多样性不足等问题，本文提出一种基于时序变化与用户聚类的提高总体多样性的方法(Aggregate diversity improvement method based on time series change and user clustering, ADI-TC, n = 3)。通过加入时间因子对预测评分进行加权提高用户评分实时性；采用奇异值分解法(Singular Value Decomposition, SVD)进行数据填充缓解数据稀疏性问题；根据用户偏好对用户进行聚类，结合用户评分和类别相似度计算用户综合相似度；通过跨类选取近邻的方式提高协同用户多样性进而提高推荐结果多样性。在MovieLens数据集实验表明本文方法相对于传统基于用户的协同过滤算法，在最近邻数为20时，MAE下降4.5%，总体多样性可以提高2%。说明本文提出的方法能在保证推荐准确性的前提下提高总体多样性。

Abstract: Aiming at the problems of traditional collaborative filtering algorithms, such as data sparsity, partial similarity calculation and insufficient diversity, this paper proposes an aggregate diversity improvement method based on time series change and user clustering (ADI-TC, n = 3) to improve the aggregate diversity. By adding time factor, the prediction rating is weighted to improve the real-time performance of user rating. Singular Value Decomposition (SVD) is used for data filling to alleviate the problem of data sparsity. Users were clustered according to their preferences, and user comprehensive similarity was calculated by combining user rating and category similarity. The nearest neighbor across classes methods were selected to improve the diversity of collaborative users and improve the diversity of recommendation results. Experiments on the Movielens data set show that compared with the traditional user-based collaborative filtering algorithm, the MAE of the proposed method decreases by 4.5% when the nearest neighbor number is 20, and the aggregate diversity can be improved by 2%. It shows that the method proposed in this paper can improve the aggregate diversity while ensuring the accuracy of recommendations.