Study On Recommended Method Based On Content Analysis And Personalized Modeling

With the rapid development of the Internet,a large amount of information is constantly emerging,and the recommendation system is playing a vital role,and each user cares and interests different content.Therefore,providing personalized recommendation content for each user has become an important research direction of recommendation systems.However,even if extremely personalized recommendations are made,there will still be the problem of information redundancy.Therefore,the text summarization algorithm can be used to compress the article information,remove redundant information and facilitate users to read.Finally,new words are constantly being produced due to the new content generated every moment,and lexical analysis as a basic task provides an analytical basis for the upper layer algorithms and applications,so it is especially important to identify new words.At the same time,the recognized new words can be added to the text summarization,and the summarization content can also provide more concise article content for the content analysis based recommendation algorithm,which is beneficial to the recommendation system.This paper mainly analyzes the content of new words and automatic summarizations,and then conducts related research on personalized recommendation through the fusion recommendation system.The specific research contents are as follows:1)A new word discovery method based on temporal features is proposed.On the basis of mutual information and information entropy of candidate words,this method scores the changes of these two features in time series.Finally,according to the score of candidate words,the words with high score are taken as the dictionary of new words and added to the lexical analysis system,which provides the basis for the downstream analysis work,and the experimental results show that the effect of word segmentation has been obviously improved.2)A summarization extraction algorithm based on neural network and combining multiple semantic features is proposed.A variety of semantic features include similarity between sentences,similarity between sentences and article topics,feature of sentence positions,and sentence features extracted by recurrent neural network.The classification model of neural network is used for training,and sentences containing important information and having semantic similarity with article topics are extracted to generate summarizations.3)A recommendation algorithm based on content analysis and user personalization model is proposed.The algorithm mainly uses the classified records of news browsed by users,carries out personalized modeling,and then fuses articles containing abstracts for joint training to obtain article features and user features.Finally,two methods,the content-based and the user-based collaborative filtering,are used to generate recommendation lists.The experimental results show that the content-based recommendation algorithm has obvious effect on the analysis of summarization content.Experiments show that the method proposed in this paper is effective.For content-based recommendations,new word discovery and text summarization can be used to conduct in-depth analysis of the content of the article,capture the core content of the article,and further improve the performance of the recommendation system.Classification tags based on user browsing history can provide personalized recommendations for users.At the same time,the summarization task can also simplify the content of the article,and the discovery of new words can also provide more accurate lexical analysis for summarization and recommendation.Finally,the shortcomings of this paper and the next research work are explained.