Research And Construction On The User Profile Of Sina Weibo

User profiles describe users with many personalized features,give users a number of different tags,such as gender,age,occupation,place of residence,etc.,which help us to understand users more easily and more detailed.The construction of user profiles can be divided into many aspects,e.g.how to accurately predict a user's gender and determine the age of the user,and further tag the user's interests and concerns and so on.At present,the domestic researches on gender classification of Chinese users are still in infancy,and it is almost impossible to find the research on classification of user age.The study on user profiles of Chinese is still in a relatively backward level.Based on the data provided by Sina Weibo,this thesis presents a robust method and model to predict the user's gender and age.In this thesis,we first extract some new features,such as word embedding,and then use common machine learning and cutting-edge deep learning to model and train classifier,next,use ensemble methods like stacking,bagging to integrate features and the results of classifiers.In the case of controlling the dimension of features as much as possible,the accuracies of predictions should be still guaranteed.Finally,this thesis presents and compares the results of several related experiments.The accuracy of the gender prediction task is 89% and age estimation task is 67%,which validate the method and model proposed in this thesis.This framework can also be extended to Chinese platforms like news media,social network as a means of mining user profiles.