Research On A Deep Recommendation Model Fused With Contextual Information

With the development of science and technology,recommendation system has been widely used in people's daily life to alleviate the problem of information overload.In recent years,the recommendation algorithm based on collaborative filtering technology has been widely used,and attracted much attention in Netflix competition.However,with the increasing number of users,the sparsity of user item rating matrix will lead to the decline of recommendation accuracy.Some recommendation technologies use auxiliary information such as t ext information to improve the recommendation accuracy.In many hybrid recommendation systems using text information,convolution matrix factorization(Conv MF)can extract the context information of the text by using convolution neural network,which achie ves better recommendation effect than other hybrid recommendation algorithms based on bag of words model,However,this method still relies on the shallow model when extracting user information.When the rating data is large,the recommendation effect is often not ideal.In order to improve the recommendation accuracy and alleviate the problem of data sparsity in recommendation system,a deep fusion model(Deep FM)is proposed.The model is composed of two neural networks.One neural network uses multi-layer perceptron to extract the row vectors in the scoring matrix to obtain the potential feature vectors of users,and the other uses multi-layer perceptron and convolutional neural network to jointly model to obtain additional text information to form the potential feature vectors of items.Finally,by constructing the fusion layer,the user feature information and project feature information are fused to get the prediction score.Through the comparative experiments on Movie Lens data set and Amazon data set,the performance of Deep FM model is better than the baseline model.Then,in order to extract the text information better,the convolution neural network structure in Deep FM is further improved.Firstly,the attention mechanism is introduced to expand the word vector,so as to highlight the important information of the text.Secondly,cross channel fusion is carried out after convolution calculation,so as to learn the relationship between different channels and generate more rich text feature information.The experimental results on Movie Lens data set show that the improved method has higher recommendation accuracy.