Research On Deep Recommendation Model With Review Information

In the era of Big Data,there is a geometric explosion in the information that people can access.What brings with it is the problem of Information Overload.As an important means of solving Information Overload,the recommendation system can help people get information that may be of interest from complicated data.The recommended technology has developed rapidly in the past few decades,and has been widely used in e-commerce,information retrieval,news push and other fields.But there are still many historical problems,such as data sparseness and cold start.To solve these historical problems,incorporating comment information into the recommendation system has proven to be a very effective method to provide more support for user preference product attribute feature mining.At the same time,with the rapid development of deep learning,its powerful potential to deal with complex problems and the ability to handle large-scale multidimensional data has also brought new breakthroughs to the recommendation system.In order to improve the recommendation performance,the main problem is to solve the following two problems:one is how to extract the characteristics of the user/commodity from the complex comment information,and the second is how to apply the feature information to the recommendation system.For the first challenge,we compare text representation models,such as bag of words,probabilistic topic model,word vector language model,etc.Finally,we utilize the word vector language model built by neural network to model the comment text information,and integrate it into the recommendation system.We represent each word as a form of word embedding and update the word vector dynamically with model training.Compared with the existing topic model using bag of words,the feature attributes extracted from the text can be better applied to score prediction and recommendati on.To address the second challenge,we propose a B-NGMM model(Neural Gaussian Mixture Model with Bert for Review-based Rating Prediction),which imitates the rating behavior of users.The model can learn user preferences and product attributes from commentary texts,construct a gaussian mixture layer in the upper layer of parallel neural networks to capture the interaction between users and products,and learn the ratings and weights over different factors.Finally,we tested our model on the product review datasets of five real Amazon users to verify the recommendation performance of the in-depth recommendation model built in this paper.The experimental results show that our B-NGMM model performs well in scoring and prediction tasks based on comment information and is superior to the most advanced methods.