Research On The Algorithm Of Words Feature Reconstruction In Social Network Environment

On social media platforms,there is a large number of multi-modal data published and shared by users.These data are published in various social media platforms in the form of text,video and pictures.Among them,text data as a widely existing data carrier appear in many forms,they are the basis of the implementation of user-theme modeling,interest mining and personalized recommendation.How to use the word features extracted from the text is a research hotspot in the field of social media data mining.At the same time,there are a lot of users in social media who contain only a small amount of text information and even lack text information,and it is hard to extract a valid text message from its historical data and many tasks can't be deployed.Therefore,the problem of reconstructing the word features of a cold-start users can make an important contribution to many data mining tasks.In view of the lack of textual information and even complete absence of text information in the social media platform,solutions have been put proposed.Firstly,the user's trust relationship,as the auxiliary information,is used to study how to construct the user-trust relationship matrix through the attention relationship among social media users,and to improve the accuracy of the word feature reconstruction of social media cold-start users.Secondly,considering the particularity of word features,we integrate the word correlation information as another auxiliary information,calculate the similarity between them,and use the similarity value as the weight to construct the word correlation matrix,and explore their hidden relationship.Then,we combine the user trust relationship matrix,word correlation matrix and user word frequency matrix to decompose the joint probability matrix,make full use of the trust relationship between users,the word correlation and the user's word frequency relation,and obtain the implicit feature matrix of the user's implicit feature matrix and word feature in the Shared low-dimensional feature space.In this way,the user-word frequency matrix is reconstructed by using the latent feature matrix of the user and the potential feature matrix of words,and then the word-feature reconstruction of the cold-start user of social media is completed.Thirdly,based on the reconstruction of the features of social media user words,this paper design and implement the user's word-feature reconstruction application system,in which the cold-start users of the lack of text information are built,and the social media cold launch user word feature reconstruction module is built,and the word cloud demonstration of the reconstruction results show the effectiveness of the algorithm.In summary,this paper revolves around the problem of reconstructing the words features of cold-start users in social media,and proposes a method of integrating trust relationships and a word feature reconstruction method that fuses trust relationships and word correlations.Through experiments,this method effectively solves cold-start user words.The feature reconstruction problem provides technical support for the data mining tasks of social media users based on text information.