Socio-demographic features meet interests: on subscription patterns and attention distribution in online social media

• Published in: Procedia computer science 2020, Vol.178, p.162-171
• Main Authors: Bardina, Mariia, Vaganov, Danila, Guleva, Valentina
• Format: Article
• Language: English
• Subjects: demographic features; Gini index; interests; social features; Social networks; statistical analysis; topic modelling
• ISSN: 1877-0509; 1877-0509
• DOI: 10.1016/j.procs.2020.11.018

This research is aimed to gain a better understanding of underlying connections between different demographic and social factors and interests as well as ways that can help to determine them. In contrast to existing studies of such correlations we focus on attention to specific topics of different socio-demographic classes. Interests are represented by topics that can be assigned to user’s subscriptions. As a measure of involvement in topics, we analyse interests heterogeneity and determine the most influential factors, associated with particular interests. Topic modelling is performed by ARTM; user’s attention to interests is measured by Gini Index and then related to socio-demographic factors. To investigate the influence of features on specific topics we trained an interpretable regression model (XGBoost and SHAP) and built a corresponding graph with clusters to analyze the results. To investigate further we scattered topics according to their socio-demographic profile and coloured according to clusters. Results show that patterns of user’s attention differ depending on socio-demographical features. We notice a shift in attention depending on age, and different patterns of attention for genders. Topics connected to gender mostly have a male audience, while age is more influential among topics with mostly female and mostly age-homogeneous audiences. We also suggest ways that can be used to improve interest prediction.