Research On Application Of User's Avatar Information In Microblog Retrieval

With the rapid development of the Internet,social media,such as Sina,Twitter,Facebook,etc,has risen rapidly,becoming an important platform for people's self-expression and interpersonal communication,gradually replacing the traditional information media,becoming one of the most important sources of information for people to obtain news and current affairs.In a social network,people are not only the consumers of information,but also the producers and disseminators of information.Data dissemination is rapid and the amount of data is unprecedented huge.The large data traffic and short text characteristics of social media,such as length limitation,special character usage and spoken expression,launch new challenges to high-quality microblog retrieval.Through observation,we find that there are a large number of pictures in social media,which contain very rich information to be explored.Among the numerous picture information,the user's avatar is the most intuitive and generalized representative of a user,and it is the concentrated express of the user's personality and preferences.Users can set pictures with different styles according to their preferences as personal avatars,the avatars of different users are also different,and a certain type of user avatars are always related to specific topics,such as women and cosmetics,men and sports,etc.How to use the abundant image information in social media to help users search related topics quickly and effectively has become the focus of attention.Unfortunately,although the research on information retrieval models(such as vector space model,probability model,language model,etc.)is very mature now,most of these methods and optimization are based on text,that is,classifying,clustering and retrieval of microblog itself and other texts,ignoring the connection between the user's avatar and the topic.In this paper,we propose a microblog retrieval method based on user's avatar classification,and explore the application of user's avatar information in microblog retrieval.We add user's avatar classification information and adjust the ranking of retrieval results to improve the performance of microblog retrieval.The research work and main contributions of this paper are as follows:1.This paper presents a classification method of microblog users' avatars.According to the content of user's avatar,we divide it into nine categories: male avatar,female avatar,more than one person,animation/cartoon character,animal,landscape,item,letter/logo and default avatar.Also,we extract the 90 D,GIST,SIFT,HOG,LBP features from the avatar,using Bayes,KNN,SVM,random forest,training classifiers and detecting classification performance.2.This paper completed the modeling and solving of the user avatar sparse group constraints.On the basis of the traditional vector space model,group lasso is used to constrain the sparse group of different user avatars,select group features from the perspective of user avatar classification.We use the block coordinate descent method to solve the model parameters,and the complex model is treated equivalently.The feasibility of this method mentioned above is verified by experiments.Based on the data,this paper proves that the classification information of user's avatar can improve the retrieval performance through the comparative experiment of randomrearrangement,hard-rearrangement,correlation-rearrangement.By analyzing and comparing the retrieval performance of different topics,we find the topics that are more sensitive to user's avatar.