Research On Micro-blog User Modeling Technology

Micro-blog has gradually become the largest and representative social media in China through its development of many years.A great quantity of users browse news,publish micro-blog texts,and interact with others through micro-blog every day.However,with the widely use of micro-blog,it also has corresponding problems.Users receive a large amount of meaningless information,and the cost of information dissemination also increases because of the contradiction between data redundancy caused by massive information and users' actual data requirements.Micro-blog user modeling is conducive to accurate delivery of all kinds of business information.It also plays a very important role in monitoring social consensus and reflecting people livelihood.This thesis mainly researchs the imperfect feature extraction in micro-blog user modeling tasks,and the low accuracy of gender classification and interest recognition.The main work is as follows:1.Aiming at the problem that the feature extraction in Chinese micro-blog gender classification needs to be improved and the accuracy of gender classification still needs to be improved,a new micro-blog classifier based on fusion strategy is proposed.Firstly,micro-blog text features are used to construct classifier and get the classification results;Then the convolutional neural network model is used to classify the user gender.Finally the XGBoost model is used to combine the two classification models to obtain the final result.The experimental results show that the method proposed by this study has better classification results compared with a series of contrast methods.2.Aiming at the problem of insufficient used of corpus and unsatisfactory recognition results in user interest recognition of micro-blog,a three-layer modeling method of user interest is proposed.Firstly,the traditional classification method is used to classify the user interest to obtain the classification result.The threshold is used to identify the users correctly determined and the users misjudged;Secondly,for the user who has tag information in the misjudged users,the similarity between user tag words and interest category keywords is got.And for the user without tag information,the similarity between user document keywords and the interest class keywords is got.Finally the user interest categories could be got by synthesizing the results of three layers.The experimental results show that the method proposed by this study can improve the accuracy of users interest recognition.