基于动态矩阵分解模型的电影推荐系统研究

A Research of Movie Recommendation Systems Based on a Dynamic Matrix Factorization Model

推荐系统已成为电子商务企业吸引客户、实现盈利的有效技术支持,它能够根据用户的网络点击数据预测其偏好,做出个性化推荐。研究了一个基于动态矩阵分解模型的NETFLIX电影推荐系统。该系统通过训练一个来自NETFLIX平台、包含9000部电影历史评分的数据集进行预测评分。核心算法包括运用矩阵分解(Matrix Factorization,MF)建立有效的数据处理模型,以及使用随机梯度下降(Stochastic Gradient Descent,SGD)训练该模型。数据集采用稀疏矩阵存储,以节省空间。在训练过程中,对预测评分增加了特定的偏向值。该系统与市场同类产品相比拥有更高的预测准确度,并向电影观众推荐符合他们喜好的电影,能极大地提高电影观看票房值。

Recommendation systems have become effective technical support for e-commerce enterprises to attract users and achieve benefits.A movie recommendation system for NETFLIX based on a dynamic matrix factorization(MF)model is developed.The system aims to train a NETFLIX dataset with history ratings of 9 thousand movies and predict user preferences on these movies.Core methods include MF,to construct an efficient data processing model,and stochastic gradient descent(SGD)algorithm,to train this model.For space saving,a sparse matrix structure is applied to store data.Specific biases are also added to predicted ratings during the training session.This system could perform with higher prediction accuracy compared to those in the market,recommend reliable movie information to users,and significantly enhancing the popularity of movies.