Research And Application Of Sentiment Analysis And Interests Based On Social Network

With the rapid development and widespread application of Internet technology,Internet data is showing explosive growth,and netizens are participating in this new type of information interaction platform in various ways.Under this background,more and more scholars have begun to participate in the research and analysis of social network data mining.Among them,user emotion analysis and interest recognition based on Internet data are two important research contents.Early scholars tried bag-of-word models and simple machine learning algorithms.With the development of language models,the accuracy of related tasks has much room for improvement.In addition,although many scholars have tried on the research of word embedding and word similarity,few scholars have conducted research on the similarity between interests.The main research contents and contributions of this paper are as follows:(1)A sentiment analysis model based on BERT is proposed.The paper first preprocesses the corpus text,and then carries out word embedding and neural network training.By using the BERT language model to achieve word embedding,it is then classified by the recurrent neural network LSTM.To explore the impact of BERT as a word embedding model on sentiment classification tasks,this paper chose word2vec as the word embedding benchmark model.In addition,this paper uses three different types of data sets.The experimental results on these real data sets show that the BERT-based language expression model combined with neural networks has a significant improvement over traditional neural network.(2)Research on interest similarity based on word embedding.An improved word similarity calculation method-Distance Combination Similarity(DCS)is proposed,which is compared with the benchmark method and the performance of the method under public data sets is observed.By collecting real personal data of LinkedIn users,using standardized methods to organize interest items in various expressions into standardized interest items,and mining high-frequency interest items from it as a research object,mining similar sets of interest items through word embedding and DCS similarity calculation method.(3)Explore the application of research results.An interest mining method based on interest similarity is proposed,which combines word embedding and DSC similarity calculation method and applies it to social network user interest mining.Design comparison experiments verify that this method can solve the problem of insufficient generalization ability of traditional methods and improve the hit rate of interest mining.This article also introduces a user analysis system based on large-scale social network data and explains the project background and system functions.By applying the two main research contents of this article to the system,functions including sentiment analysis,member recognition,interest recognition,and user relationship analysis are realized.The research results of this paper can quickly determine the user's emotional polarity in various types of comment information on the network and can prove that there is an effective similarity between human interests.The research results of this paper are not only useful for the analysis of the user's emotional characteristics in social networks,but also have great application value in the business field related to the interest recommendation system.