Research On Recommendation Model Based On Deep Learning

As an effective way to solve the problem of information overload,recommendation model has been widely used in various fields in recent years.Due to the shallow structure design,the traditional recommendation model cannot extract the deep level of features of users and items.Simultaneously,the traditional recommendation model cannot reasonably model the serialized user behavior data,and it is difficult to learn the before and after dependencies of the serial data.The existing recommendation models based on deep learning mostly focus on the learning of users' historical behaviors or the analysis of the internal relationships of items,accordingly,it is difficult to take into account the effective extraction of deep user features and item features.In view of this,this thesis explores the combinatorial construction method based on deep learning recommendation model,selects the deep learning method that is more suitable for the features of users and item data,carries out more targeted feature extraction,and ultimately constructs two combinatorial recommendation models based on deep learning.The specific research contents of this thesis are as follows:(1)Aiming at the problem that the existing recommendation model fails to take into account the effective extraction of deep-seated user features and item features,a combined recommendation model based on LSTM and CNN was researched and constructed.In this model,local relevant features of the item are extracted by CNN and user rating data are processed by LSTM to obtain user features.Afterwards the model integrates relevant features to get a predicted rating and generate recommendations for users.Compared with other recommendation models based on deep learning and traditional recommendation models,the results show that the MSE and MAE loss values of the constructed model are smaller.(2)Aiming at the problem that the distant internal dependencies of user features and item features cannot be accurately expressed,a combined recommendation model of GRU and CNN with self-attention mechanism is constructed.In this model,CNN was used to extract item features and GRU was used to analyze user rating data.Then,through the introduction of self-attention mechanism,the weighted item features and user features were obtained.Finally,the model gets prediction rating and a list of recommendations.Through simulation experiments,the performance of each model is evaluated reasonably,and the results show that the constructed model performs better.This thesis researchs two recommendation models based on deep learning,and the combinatorial construction method based on deep learning recommendation model is further explored,providing feasible ideas for the design of other recommendation models.