Research On Recommendation Algorithm Based On Deep Learning And Feature Embedding

In today's information overload,people can have a variety of ways and means of obtaining information through the Internet,but what most needs people to spend time is to find the information of interest in the vast information ocean.The recommendation system came into being under these conditions.For users,recommendation systems can help make decisions and discover new things.For businesses,the recommendation system can provide personalized services,increase trust,stickiness and revenue.The traditional recommendation algorithm usually analyzes the user's rating data of the item,and then performs preference recommendation.However,if the user's behavior information on the item is too little,making the rating data too sparse,the recommendation's effect will be worse.In order to predict the blank information in the scoring data,the researchers proposed a scoring prediction algorithm.The scoring prediction algorithm studies the attributes of items and constructs a model to predict and complete the blank in the user's item scoring matrix.Common scoring prediction models include average method,matrix factorization model and so on.But these methods have problems such as difficulty in extracting deep features and single data processing methods.These problems will lead to a low prediction accuracy and an unreasonable results.The models in deep learning have strong adaptability and are widely used in many fields.Compared with the embedding method,the deeplearning model can analyzes the connection between the data more fully,and the features learned from the data are more representative,which can indirectly improve the prediction accuracy.In response to the problems above,this paper improves the feature embedding recommendation algorithm(FE),uses deep learning model,including convolutional neural networks and long and short text memory models,to process the text feature of the project name,extracts the deep information of the item name attribute and processes feature vector.The other attributes are processed into a vector form by an embedding method.Then,each feature vector of the user and the item is added to obtain a user feature matrix and a item feature matrix,and finally the prediction score is obtained by multiplying the two feature matrices.The research work in this paper consists of the following three parts:First,by reading the literature,a deep understanding of the research status of traditional recommendation algorithms,models related to deep learning,and embedded models is analyzed.The advantages and disadvantages of each algorithm are analyzed,and several deep learning-based recommendation algorithms are described.Secondly,in view of the problems of insufficient feature extraction by embedding model,low score prediction accuracy and data sparseness in the datset,this paper introduces a model that combine convolutional neural network into the feature embedding-based recommendation algorithm,and proposes a combination algorithm of convolutional neural network and feature embedding-based recommendation algorithm(Feature Embedding-Convolutional Neural Network,FE-CNN).The model uses the embedding method to process the user's and item's label feature to obtain the user's and item's feature matrix.According to the text information of the item's name,the convolutional neural network isintegrated to process the text information in the item's name to achieve fuller mining of features.Multiply the user's and item's feature matrix processed by the model to get the prediction score.Experimental results confirm that the algorithm FE-CNN has lower prediction error on the two sets of datasets on Movie Lens than the comparison algorithm.Thirdly,in view of the problem that the accuracy of feature extraction is limited by the size of the sliding window in convolutional neural networks,a long and short text memory model is used to process information of item's name.A recommendation algorithm based on long and short text memory models and feature embedding(Feature Embedding-Long Short Text Memory,FE-LSTM)is proposed.The algorithm uses the LSTM model to process the feature of the item's name.The other attributes are processed by the embedding model to build the feature matrix of the user and the item.Finally,the feature matrix is multiplied to obtain the prediction score.It is verified through experiments that the FE-LSTM algorithm performs better on the MAE and RMSE values of the prediction results,and at the same time,it can alleviate the problem of data sparsity and improve the recommendation accuracy.