Deep Neural Network Based Recommendation Technology Integrating Rating Matrix And Comment Text

With the rapid development of information technology,the amount of information is growing exponentially,which has a great impact on people's life.On the one hand,more and more information is available.On the other hand,it is getting more and more difficult to find the information users need.However,Recommendation System can help users quickly find the most interesting content in complex data,and provide users with personalized services.Recommender System has been successfully applied in many fields,and has received extensive attention in industry and academia.Among many recommendation algorithms,collaborative filtering algorithm is one of the most widely utilized algorithms,but the sparsity of data and cold start problems limit the improvement of model performance.In recent years,deep learning has been successfully applied in many fields,which has brought new research hotspots to Recommendation System.This paper mainly studies the extraction and application of data features in the recommendation process based on deep neural networks.At the same time,it combines neural networks and attention mechanisms to effectively extract features from text information,and further integrates scoring features and text features to improve recommendation performance.The main contents of this paper are as follows:(1)Aiming at the problem of data sparsity,this paper proposes a deep learning recommendation model based on the fusion of rating matrix and comment text.Text information is used to alleviate the impact of data sparsity,and user preferences and item characteristics are obtained.The scoring data contains potential associations between users and items.Convolution neural network will be used to process the comment text,and attention mechanism will be introduced to extract the representative comments in the comment information,as to obtain high-quality user and item feature representation.The scoring data is processed by deep neural network,and the depth features are extracted,and then the features are fused to predict the user's rating of the project.(2)Aiming at the problem of model interpretability,a recommendation model based on hierarchical attention is proposed.The recommendation model based on review information only consider the important single words in the review,while ignoring the truly valuable reviews will affect the performance of the model.Firstly,it is necessary to use the bi-directional GRU to process the text information to obtain the vector representation of the text,and then introduce the attention mechanism to extract the important words in the comments,and finally get the final feature representation of the review level.Secondly,the bi-directional GRU will further combine the representations obtained by the attention layer to obtain the document-level representation of the comments.Then,the attention layer of the comment-level attention layer will extract representative comments in the document to better present the user's preferences and project characteristics.This model uses both word level and comment level to evaluate the importance of comments,which improves the interpretability of recommendation results.(3)In order to verify the performance of the two models mentioned above,this paper selects data from different fields in the Amazon data set for experiments.The results show that the model proposed in this paper has achieved better results on different data sets,verifying the effectiveness of the proposed model on the task of scoring prediction.