| In recent years,with the rapid and large-scale development of the Internet,Micro-blog has become increasingly popular as a social networking platform,such as Sina Micro-blog,Tencent Micro-blog,and Twitter,etc.According to statistics,in the end of September 2017,the number of Micro-blog monthly active users has reached 376 million,which is an increasing of 27% over the same period of the previous year.The user profile is a process of describing the user's attributes.It will give Micro-blog user characteristic label obtained by analyzing existing real data.With more and more network devices,Micro-blog will become more and more popular as a social network platform.Data mining technologies for Micro-blog user profile will become more and more urgent,and can be widely used in various applications,such as marketing online,service research,and public opinion monitoring.This thesis takes the Sina Micro-blog data as the research object,and studies how to build Micro-blog user profile.It focuses on the classification methods of the ages and genders of Micro-blog users.First,this thesis proposes a user gender and age classification method that integrates Micro-blog user sentiment.A variety of basic classification methods are compared and analyzed,and then ensemble classification is used to improve the classification accuracy.Finally,the ensemble classification based on supervised clustering is proposed to further improve the classification accuracy.The main work of this thesis is as follows:1.Taking into account the sentiment differences between users of different genders and ages,sentiment tags can influence the gender and age classification quantity of users.This thesis proposes a gender and age classification method that fuses user sentiment of Micro-blogs.This method uses the idea of transfer learning by using LSTM model.The LSTM model predicts the sentiment poles of the Micro-blog data,that are integrated with the Micro-blog text features.Logistic Regression,Na?ve Bayes,k-Nearest Neighbor,Random Forest,SVM and other classifications have been studied in the experiment part.The results show that the classification method of integrating sentimental features is better than the traditional one,and the accuracy rate is increased by an average of 2.1%.2.Since each base classifier has the possibility of misclassification,if the base classifier is combined and voted,the ensemble classifier will be misclassified only when more than half of the base classifiers are in error.A relatively good classification method is bagging and stacking separately for better performing methods.The results show that the combination of the same method can improve the classification accuracy in the text classification of Micro-blog by dividing training data into several small parts.The accuracy of the ensemble classification is improved by about 0.8% compared with the traditional classification.3.In view of the ensemble classification of the same methods described above,it is that a random sampling of datasets is used to train the base classifiers,and therefore it does not know which of them will cause degraded classification performance.This thesis proposes an ensemble classification method based on supervised clustering.First,the data samples are clustered into several clusters within each category,and then those clusters are combined.Finally,the ensemble classification method which performs best in the previous experiments is used for classification.The results show that the ensemble classification with supervised clustering is improved by 1.79% and 0.67% for gender and age,respectively,compared to the general ensemble method.The classification method of user gender and age based on Micro-blog text studied in this thesis is applicable to other Micro-blog platforms.In addition,other labels for users,such as geographic,preference,or occupation,attribute can use the proposed methods in this thesis. |
