Coupled social media content representation for predicting individual socioeconomic status. (15th July 2022)

• Record Type: Journal Article
• Authors: Zhao, Tao; Tang, Lu; Huang, Jinfeng; Fu, Xiaoming
• Is Part Of: Expert systems with applications. Volume 198(2022)
• Publication Date: 2022-07-15
• Subjects: Socioeconomic status -- Coupled social media content representation -- Structure-aware social media text representation -- Coupled attribute representation; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.116744
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Predicting individual socioeconomic status (SES) from social media content benefits various applications in economic and social fields. Most previous works adopt machine learning methods with predefined features to infer SES. Nevertheless, they ignore some important information of social media content, such as order, structure and relation information, which leads to limited performance. In this paper, we propose a COupled social media content REpresentation model (CORE) for individual SES prediction, which efficiently exploits latent complex couplings of social media content. CORE devises a structure-aware social media text representation method to incorporate the order and the hierarchy of social media text, and leverages a coupled attribute representation method to take into account intra-coupled and inter-coupled interaction relationships among user level attributes. Our experiments on a real data set of a Chinese microblogging platform demonstrate that our approach significantly outperforms benchmark methods, which validates its efficiency and robustness. The proposed model could be applied to improve the SES prediction and other user profiling tasks. Highlights: A coupled social media content representation model for predicting individual SES is proposed. A structure-aware social media text representation method is presented. A coupled attribute representation method exploring intrinsic relationships is devised. Experiments prove that the proposed model greatly outperforms state-of-the-art models.