User Age Recognition In Social Network

Automatic data analysis in social network is an important task in several research communities,such as natural language processing and social network analysis.One basic and fundamental technology is known as age prediction which aims to determine the age information of online users with user generated textual and social information.Age prediction of social media users from evidence benefits several applications,such as intelligent marketing,online advertising,and personality analysis.Although there are various studies which focus their researches on age classification,few of them model age prediction as an age regression problem.In this paper,we mainly focus on regression models for age prediction.In details,our studies mainly include the following three aspects:First,this paper proposes a novel approach to user age regression with active learning.The main idea is first to divide the whole feature space into several disjoint feature subspaces and then leverage them to learn a committee of regressors.Second,given the committee of regressors,we apply a query by committee(QBC)method to select unconfident samples in the unlabeled data for manual annotation and then add them to the labeled sample set.Third,using the updated labeled sample set to build ultimate model for age regression.Experimental results show that the proposed method can effectively reduce the annotation cost and obtain better age regression performance than random selection strategy.Second,this paper proposes a cross-media age regression approach with textual adaptation.The main idea is first to obtain two views generated by random subspace generation(RSG)approach with the labeled data from the source media in a co-training algorithm.Second,training regressors with above two views to automatically label samples in the target media and then add the most confident sample to the labeled sample set by the query by committee(QBC)approach.Third,building the ultimate regressor with the updated labeled sample set to predict user age in target social media.Empirical studies demonstrate that our approach can effectively alleviate the problem that the textual features in different social media are sometimes differently distributed and improve the performance of cross-media age regression.Third,this paper proposes an age prediction approach by combining classification and regression.First,in order to learn long relevant connection between input values,we build the LSTM(Long Short-Term Memory)model of age regression and age classification respectively for age identification.Then,we combine the result of the age classifier and age regressor linearly as the final result of age identification.Experiments demonstrate that the hybrid model we proposed can utilize advantages of classification models and regression models simultaneously,and achieves better performance than either classification models or regression models.