Research On Recommendation Model And Algorithm Based On Deep Learning

With the development of Internet technology,the era of information technology has gradually transitioned to the era of data technology,and data is playing an increasingly important role in our lives.In the face of ubiquitous data,how to obtain the required information from it becomes very difficult.This is the so-called ‘information overload' problem.The traditional recommendation system alleviates this problem to certain extent,analyzes the user's historical behavior and the user's preferences from the massive data,and the items with the user's preferences are recommended to the user.However,there are some problems in practical application,such as data sparsity and cold start,which will lead to low accuracy and single recommendation.In view of the above problems,this paper mainly makes the following work:Firstly,we propose a deep neural network recommendation model that combines Skip-gram and weighted loss function.This model used three layers ReLU layer structure to construct a deep neural network model,and combines Skip-gram for item embedding to obtain a dense representation vector,the weighted loss function is used to train the parameters of the deep neural network,which improves the accuracy of recommendation without using additional information,balances the popularity of recommended items,and ensures novelty.Secondly,the above model only uses the rating matrix.On this basis,we propose a deep model for the processing of review texts and rating matrices.The model extracts deep-level features and combines them to make rating predictions.Then,we use a pretrained Electra model to obtain the implicit expression of each comment.Next,we combine a deep sentiment analysis and attention mechanism to analyze the comment text at the context and semantics level,thereby solving the short text semantic analysis problem.User(item)reviews interact with a rating matrix to predict the user's rating of a product in the fusion layer module.Finally,we test two models.Experimental results on the APP dataset and the Last.FM dataset show that the DSM model has a certain improvement in accuracy and diversity compared with the existing methods when recommending applications and songs.ELM model was compared on 6 data sets,and the experimental results show that the model has better performance than other systems,and the average prediction error is reduced by 12.821% at most.Through the above experiments,the two models proposed in this paper are applicable to recommend accurate items to users.