Research On Collaborative Filtering Algorithm Based On Deep Learning

Collaborative filtering algorithm is one of the most important and effective algorithms in recommendation system,and it is widely used in various commercial recommendation scenarios.The traditional collaborative filtering algorithm has problems such as poor fitting ability to sparse data.The collaborative learning algorithm based on deep learning has attracted the attention of researchers in recent years by virtue of its ability to overcome problems such as data sparsity.However,the collaborative learning algorithm based on deep learning still has a lot of optimization space in the network structure and input data level.In this context,this paper studies the collaborative filtering algorithm based on deep learning.The following is the main research content of this paper.And contributions:First,for the simple neural network structure,it is difficult to capture the product relationship between vectors.This paper proposes an Inception-Resnet network-based external product interaction collaborative filtering algorithm(IncONCF),which uses Inception-Resnet network structure to capture users and objects.The data relationship of different orders between the hidden vectors greatly enhances the feature expression of the user and the hidden vector of the object,and uses the outer product interaction to map the embedded layer relationship between the user and the item,and expands the hidden vector of the user and the item.The relationship between data interactions.The simulation experiments of the model on the real data set confirmed that the addition of the Inception-Resnet network structure improves the recommendation accuracy,and the performance of the model using the simple CNN network structure is better.Second,the IncONCF model uses a complex network to map the relationship of vector outer product interactions,but in the interaction layer lacks the interaction of vector linear combination.This paper proposes the idea of modeling the element product interaction based on matrix decomposition.Based on the IncONCF model,Added an embed layer based on element product interaction.In this paper,the optimized IncONCF model is called IncEONCF.The expression of the linear combination between the user and the hidden vector of the IncEONCF model is greatly improved,which improves the accuracy of the recommendation and accelerates the training of the model to some extent.Thirdly,adding the auxiliary feature data of the item in the collaborative filtering algorithm and modeling it with the deep neural network is an important means to improve the recommendation effect.Based on this idea,this paper proposes a multi-view structured collaborative filtering algorithm(MV-DMF)based on depth matrix decomposition,using the hidden vector similarity between the object perspective and the user perspective as the predicted score in the recommendation,and adding the attention mechanism module.The difference iin the degree of user attention to the item is captured,which greatly enhances the feature expression of the model.Using the real data with the image information of the item to compare the model,the improvement of the recommended accuracy of the MV-DMF's innovative structure is verified.