摘要
针对目前推荐系统广泛存在的数据稀疏性、预测评分不准确和神经网络模型复杂度较高等问题,提出深度神经网络模型ProfileDNN.模型借鉴自然语言处理中的Word2Vec表征学习方法预训练物品的嵌入向量,并利用物品的嵌入向量构建物品和用户画像,最后基于深度神经网络模型学习用户对物品的预测评分.基于3个公共数据集的对比实验表明,相比同类模型,ProfileDNN模型的复杂度更低,且推荐准确率最高提升达1.1%.
Score prediction has always been the core issue in the research process of recommender systems.In view of the widespread data sparsity problem in the current recommendation system,the inaccurate prediction score and the high complexity of the neural network model,a deep neural network model ProfileDNN is proposed.The model draws on the Word2Vec characterization learning method in natural language processing to pre train the embedding vector of the item,then uses the item embedding vector to construct the item and user portrait,and finally learns the user's predicted score for the item based on the deep neural network model.Comparative experimental studies on three public data sets show that the ProfileDNN model is less complex than similar models,and the recommendation accuracy rate is improved by up to 1.1%.
作者
金楠
王瑞琴
陆悦聪
JIN Nan;WANG Ruiqin;LU Yuecong(School of Information Engineering,Huzhou University,Huzhou 313000,China;Zhejiang Province Key Laboratory of Smart Management&Application of Modern Agricultural Resources,Huzhou 313000,China)
出处
《湖州师范学院学报》
2022年第8期45-54,共10页
Journal of Huzhou University
基金
浙江省自然科学基金项目(LY20F020006).