基于增量矩阵分解的协同过滤推荐模型研究

Research on Incremental Matrix Factorization Collaborative Filtering for Recommender System

推荐系统能够有效解决用户在海量数据下个性化获取有效数据问题。传统推荐模型一般离线完成训练后再上线使用,不能实时更新模型以至随着时间的推移推荐结果并不精确。为了解决上述问题,提出一种基于增量矩阵分解的协同过滤推荐模型,该模型能够处理流式数据完成在线训练并实时更新模型。实验结果表明,该模型在保证较高召回率的同时其模型更新时间远快于其他模型。

Recommender system can solve the problem of personalized access to effective data under massive data effectively.The traditional recom⁃mendation model is used online after offline training,which cannot be updated in real time,so that the recommendation results are not accurate over time.To solve above problem,presents an incremental matrix factorization collaborative filtering model,which can train the model with streaming data update the model in real time.The experimental results show that our model have competitive recall,while being significantly faster.