基于深度学习的多元信息嵌入推荐模型

传统的矩阵分解法只能简单提取低阶信息,特征组合单一,无法挖掘更多隐含信息。依赖稀疏的评分表已经无法满足个性化推荐。MovieLens数据集是研究推荐系统的经典数据集。通过改进神经协同过滤模型,使用多个嵌入层对MovieLens数据集所有信息进行嵌入分析,充分挖掘隐含信息,使得推荐算法更有实用意义。通过在MovieLens-100k和MovieLens-1m上设置多组对照实验表明,该改进模型不仅学习了更多信息,而且与神经协同过滤模型相比,其错误率显著减小。

MULTI-INFORMATION EMBEDDED RECOMMENDATION MODEL BASED ON DEEP LEARNING

The traditional Matrix factorization method can only extract low dimension information simply,and single feature combination can not extract more implicit information.The sparse scoring table has been unable to meet the personalized recommendations.MovieLens dataset is a classical dataset for recommendation system research.By improving the neural collaborative filtering model and using multi-embedding layers to embed all information of MovieLens,the hidden information was fully excavated,which made the recommendation model more practical.The experiments on MovieLens-100k and MovieLens-1m show that the new model not only learns more information,but also reduces the error rate significantly compared with the neural collaborative filtering model.