Recommendation Model Of Deep Learning Integrates Rating Matrix And Review Text

With the arrival of the "Internet +" era,the data scale is sharply expanded.As an effective way to mitigate the information overload problem,the recommendation system will be born,and the user's interaction behavior is actively recommended to provide personalized customization services.As one of the research directions of the recommended model,the score prediction task measures the user's preference of the project,and has been widely spread in the field of e-commerce platforms,advertising marketing.However,the traditional recommendation system has generally existed data sparsity problems,and it is difficult to adapt to complex recommended scenarios,while deep learning technology has data fitting capacity and deep characteristic learning ability,which can effectively make up for problems such as sparse and interpretive,and improve the performance of scoring prediction system.Therefore,aiming at the scoring prediction task,taking convolution neural network and deep matrix decomposition network as the framework,this thesis proposes a hybrid recommendation model Deep-Con MF,which uses deep learning technology to integrate scoring data and comment text,fully mine the characteristics of users and items,extract the potential interests of target customers,optimize service quality,and provide more accurate scoring prediction information for e-commerce platforms to explore the market.The specific research work includes:(1)The auxiliary information of comment text is integrated into the deep recommendation system,and the effective combination of explicit and implicit interactive information is realized by using deep learning technology,which not only makes up for the deficiency of shallow recommendation model,but also weakens the negative impact of a large number of missing values contained in a single data source on recommendation performance.(2)Implementing the interaction of user characteristics and project characteristics in the introduction of factor decomposition machine in the fusion prediction layer.The correlation degree of any two features is fitted by the second-order interaction term between the modeling components,the user's score prediction of the project is realized,and the deep fitting capacity of the model is enhanced.(3)Select the Amazon Public Data Set to verify the model is effective and scientific.The root mean square error is used as a measurement index,compared the prediction performance of other six baseline models,and performing detailed analysis of the factors.The three data sets show that the performance of the Deep-Con MF model proposed in this thesis is better in the recommended rating task.