Twitter Demographic Classification Using Deep Multi-modal Multi-task Learning

Abstract

Twitter should be an ideal place to get a fresh read on how different issues are playing with the public, one that's potentially more reflective of democracy in this new media age than traditional polls. Pollsters typically ask people a fixed set of questions, while in social media people use their own voices to speak about whatever is on their minds. However, the demographic distribution of users on Twitter is not representative of the general population. In this paper, we present a demographic classifier for gender, age, political orientation and location on Twitter. We collected and curated a robust Twitter demographic dataset for this task. Our classifier uses a deep multi-modal multitask learning architecture to reach a stateof-the-art performance, achieving an F1-score of 0.89, 0.82, 0.86, and 0.68 for gender, age, political orientation, and location respectively.

Listening between the Lines: Learning Personal Attributes from Conversations

“…Social Media User Profiling: The rapid growth of social media has led to a massive volume of user-generated informal text, which sometimes mimics conversational utterances. A great deal of work has been dedicated to automatically identify latent demographic features of online users, including age and gender [3,4,8,9,17,[34][35][36]41], political orientation and ethnicity [26,[32][33][34]41], regional origin [8,34], personality [14,36], as well as occupational class that can be mapped to income [10,31]. Most of these works focus on user-generated content from Twitter, with a few exceptions that explore Facebook [35,36] or Reddit [8,14] posts.…”

“…Kim et al employ Graph Recursive Neural Networks (GRNNs) to infer demographic characteristics of users [17]. Vijayaraghavan et al exploit attention-based models to identify demographic attributes of Twitter users given multi-modal features extracted from users' profiles (e.g., name, profile picture, and description), social network, and tweets [41].…”

“…On the other hand, many efforts have considered the problem of profiling social media users in order to predict latent attributes such as age, gender, or regional origin (e.g., [3,4,8,9,17,[34][35][36]41]). While social media posts and utterances are similar in that both are informal, the former can be associated with many non-textual features that are unavailable outside of the social media domain (e.g., social-network friends, likes, etc.…”

Me, My Echo Chamber, and I

“…We are interested in exploring the following overarching question: to what extent does making social media users aware of their online political echo chambers affect their beliefs and future platform engagement patterns? In particular, we design an intervention to help us measure changes in the following response variables for each participant p: For R2, we compute political diversity using Shannon's entropy and infer the political ideology of each account p follows using a state-of-the-art classifier presented in [26]. R3 is measured using political alignment scores inferred as a part of [3] for different web domains.…”

Inferring Social-Demographics of Travellers based on Smart Card Data

“…However, researchers mainly concentrate on people's online behaviour, such as web browsing (Hu, Zeng, Li, Niu, & Chen, 2007;Saste, Bedekar, & Kosamkar, 2017) and social network (Rao, Yarowsky, Shreevats, & Gupta, 2010;Vijayaraghavan, Vosoughi, & Roy, 2017). The discriminative power of people's mobility in the physical world has been overlooked, especially the travel behaviour via public transit.…”