Micro-blog User Profiling: A Supervised Clustering based Approach for Age and Gender Classification

Abstract

User profiling is the process of tagging user attributes, such as gender, age, location and so on. Currently the popular way is to do classification by treating attributes as labels. In addition, ensemble classifications are used to future improve classification quality. However, the base classifier of traditional ensemble classification is trained by using random sampling, which cannot guarantee a constantly good classification performance. This paper proposes a supervised clustering of ensemble classification approach to tag micro-blog users with age and gender attributes. The proposed approach is to divide the training data with a same label into multiple clusters and then combine different clusters from training data with different labels into training data subsets. Each subset is used to train a base classifier. Experimental results show that the proposed approach can improve the gender and age classification accuracy classification by 1.79 % and 0.67 % respectively, compared with the traditional ensemble classification approach.