Research On Hybrid Collaborative Filtering Algorithm Based On Improved User Preferences And Item Features Topic

With the development of network technology,Internet data has grown exponentially,and all kinds of information are full of people's daily lives.Though the increasing data eases the lack of information,which greatly facilitates people's lives,with the increasing amount of data,how to make people accurately and effectively obtain information becomes an urgent problem to be solved.At present,most people use the search system such as Baidu,Google to obtain information,but when the amount of data is getting larger and larger,the disadvantages of the search system requiring stronger interaction are revealed,and the emergence of recommendation system effectively solves the search system's Insufficient.The recommendation system based on big data not only can reduce the necessary interaction between the user and the system,but also can customize the personalized solution for the user,and it can effectively improve the efficiency of the user to obtain information.By analyzing the principle of the traditional collaborative filtering algorithm,this paper recognizes the shortcomings of the traditional collaborative filtering algorithm when the user score data is missing or extremely sparse.In order to alleviate the data sparsity problem of collaborative filtering algorithm,this paper extracts the potential topics in the text data,combines the text topic with the scoring data,and builds a hybrid collaborative filtering algorithm based on improved user preferences and item features topic.In terms of algorithm improvement,this paper uses text mining technology to preprocess the text,and uses LDA topic model to extract potential topics in the text,and improves the defects of LDA topic model.In this paper,In the process of incorporating user comment text into the collaborative filtering algorithm,a series of preprocessing work are carried out,such as word segmentation,stop word elimination,word form restoration,part-of-speech tagging and text vectorization.In the process of part-of-speech tagging,this paper combines the existing literature and its own experience to eliminate the vocabulary that cannot reflect the subject in the text.The job of elimination can not only reduce the noise data in the text,but also improve the prediction accuracy of the model.After the text data preprocessing is completed,the paper uses the LDA topic model to extract the topics from the user and item levels,construct the user preference theme and the item feature theme,and use the constructed theme for the similarity calculation in the collaborative filtering algorithm,this job can effectively reduce the complexity of the algorithm's calculation.When using the topic distribution to perform similarity calculation,on the one hand,considering that the LDA topic model cannot distinguish the defects of the topic sentiment tendency,according to the user rating data,a comment attitude indicator is constructed to measure the user's preference for the item,so the user's preferred item will be recommend to the user;on the other hand,considering that the collaborative filtering algorithm ignores the impact of time on the recommendation results,this paper builds a time factor indicator based on the comment time data.This paper also proposes the user interest change indicator for the user interest change problem,and proposes the item heat attenuation index for the item heat attenuation problem.After constructing various indicators,a collaborative filtering algorithm based on improved user preference theme is constructed for the user level,and a collaborative filtering algorithm based on improved item feature theme is constructed for the item level.Finally,according to the evaluation indexes such as recall rate and promotion degree,through the iterative calculation method,the combination parameters of two improved collaborative filtering algorithm are determined.finally,this paper builds a hybrid collaborative filtering algorithm based on improved user preferences and item features topic.The experimental results show that the proposed hybrid collaborative filtering algorithm based on improved user preferences and item feature topics can alleviate data sparseness to a certain extent,compared with the traditional collaborative filtering algorithm.And it can improve the recommendation of collaborative filtering algorithms.