结合灰色预测的动态概率矩阵分解

Dynamic probabilistic matrix factorization with grey forecast

中文关键词  推荐系统  概率矩阵分解  灰色预测模型

英文关键词  recommender system  probabilistic matrix factorization  grey forecast model

中文摘要

      推荐系统的目标是找出符合用户喜好的物品, 但是用户的喜好和物品的特征是动态变化的, 这种变化会影响推荐系统的准确性. 很多推荐系统只是简单的使用概率矩阵分解模型, 缺乏对这个问题的有效解决. 本文利用灰色系统理论中的灰色预测模型对用户和物品的动态性建模, 继而提出了一个基于概率矩阵分解和灰色预测模型的动态推荐系统.首先, 利用概率矩阵分解模型生成各个连续时间窗中用户和物品的隐式向量. 接着, 利用灰色预测模型得到未来时间窗中用户和物品的隐式向量, 继而进行推荐. 实验结果说明本文的算法能够有效地对用户和商品的动态性进行建模, 且优于一些现存的最好的算法.

英文摘要

      The goal of recommender system is to find out the items which meet the users’ preferences. However users’ preferences and items’ features change over time that can affect the accuracy of recommender system. Many recommender systems simply employ probabilistic matrix factorization (PMF) model without addressing this issue. Motivated by the grey system theory, in this paper, the dynamics of both users and items are modeled by utilizing the grey forecast (GF) model. Accordingly, a new dynamic recommender system based on probabilistic matrix factorization and grey forecast model (DPMF-GF) is developed. Firstly, the probabilistic matrix factorization (PMF) model is used to produce user’s and item’s latent vectors between consecutive time windows. Next, the grey forecast model is used to predict user’s and item’s latent vectors in the following timestamp. The experimental results show that our model can effectively model users’ dynamics and items’ dynamics, and outperforms the existing state-of-the-art recommendation algorithms.