Recommender System Based On The Technology Of Natural Language Processing

Recommender is a basic method of massive information filtering technologies widely used in information push and recommendation applications to solve sparsely data points and reach the recommended quasi-group and the diversity of purpose.Current approach focuses on SVD method to establish the two-dimensional relationship between users and items based on purchase history.Compared with deep learning technology,the SVD framework does not require massive amounts of data and expensive computing resources and has a better application in the mobile intelligent terminal.The research study used specifically the technology of natural language processing.The language model,,which is a time series model,can express the dependence of words in the time dimension.The dependency of the time dimension is reflected by the grammar of the language and the situation of the person within the context of the environment,which has similar dependency structure with the users purchased items in the special environment.The main contributions are:(1)The researchers extended the LDA model for the item recommendation.LDA model for items to capture the dependence of item in the time dimension was used to map the item to the topic space.Modeling:"item ID" as "word","a period of interaction with the user-item set" as "document",and captured the probability of the "corpus" data structure.LDA model items in the "topic space" automatic polymerization,we then introduced the real vector representation for items and semantic information of items.(2)Built the U_LDA model by integrating the user information and build the UB LDA model by integrating the user behavior information based on the LDA model.We implement to capture the dependence of user and user behavior in the time dimension,and at the same time introduce the real vectors of the user and user behavior within the hidden space.Additionally,We did an evaluation of the correlation between users and the correlation between items,and completed the recommendations based on similarity modules.The improved model eliminates the interference factors such as noise data,random behavior and so on,which enhances the robustness of the model and improves the ability to explore the hidden space.(3)Finally,We carried out an experiment on the optimization of topic models using a dataset from Taobao and did a comparative analysis of our model and the traditional collaborate filtering and the SVD method based on the precision e-value,recall value and F1 value.The experimental results showed that improved language model has the ability to explore the hidden space than the traditional SVD framework.