AI for social good: social media mining of migration discourse

Abstract

The number of international migrants has steadily increased over the years, and it has become one of the pressing issues in today’s globalized world. Our bibliometric review of around 400 articles on Scopus platform indicates an increased interest in migration-related research in recent times but the extant research is scattered at best. AI-based opinion mining research has predominantly noted negative sentiments across various social media platforms. Additionally, we note that prior studies have mostly considered social media data in the context of a particular event or a specific context. These studies offered a nuanced view of the societal opinions regarding that specific event, but this approach might miss the forest for the trees. Hence, this dissertation makes an attempt to go beyond simplistic opinion mining to identify various latent themes of migrant-related social media discourse. The first essay draws insights from the social psychology literature to investigate two facets of Twitter discourse, i.e., perceptions about migrants and behaviors toward migrants. We identified two prevailing perceptions (i.e., sympathy and antipathy) and two dominant behaviors (i.e., solidarity and animosity) of social media users toward migrants. Additionally, this essay has also fine-tuned the binary hate speech detection task, specifically in the context of migrants, by highlighting the granular differences between the perceptual and behavioral aspects of hate speech. The second essay investigates the journey of migrants or refugees from their home to the host country. We draw insights from Gennep's seminal book, i.e., Les Rites de Passage, to identify four phases of their journey: Arrival of Refugees, Temporal stay at Asylums, Rehabilitation, and Integration of Refugees into the host nation. We consider multimodal tweets for this essay. We find that our proposed theoretical framework was relevant for the 2022 Ukrainian refugee crisis – as a use-case. Our third essay points out that a limited sample of annotated data does not provide insights regarding the prevailing societal-level opinions. Hence, this essay employs unsupervised approaches on large-scale societal datasets to explore the prevailing societal-level sentiments on YouTube platform. Specifically, it probes whether negative comments about migrants get endorsed by other users. If yes, does it depend on who the migrants are – especially if they are cultural others? To address these questions, we consider two datasets: YouTube comments before the 2022 Ukrainian refugee crisis, and during the crisis. Second dataset confirms the Cultural Us hypothesis, and our findings are inconclusive for the first dataset. Our final or fourth essay probes social integration of migrants. The first part of this essay probed the unheard and faint voices of migrants to understand their struggle to settle down in the host economy. The second part of this chapter explored the viability of social media platforms as a viable alternative to expensive commercial job portals for vulnerable migrants. Finally, in our concluding chapter, we elucidated the potential of explainable AI, and briefly pointed out the inherent biases of transformer-based models in the context of migrant-related discourse. To sum up, the importance of migration was recognized as one of the essential topics in the United Nation’s Sustainable Development Goals (SDGs). Thus, this dissertation has attempted to make an incremental contribution to the AI for Social Good discourse.