The COVID-19 pandemic has caused damage not only in the numerical aspects of the global economy but has also had severe impacts on the mental well-being of the Thai population. Assoc. Prof. Dr. Suppawong Tuarob, Head of the Machine Intelligence and Knowledge Engineering (MIKE) research group at the Faculty of ICT, Mahidol University, along with the research team, addresses this concern and aims to help solve this problem using Artificial iIntelligence framework called ‘FAST.’
“This work initiated from the COVID-19 situation, where it was observed that government policies, such as financial aid and lockdowns, had significant repercussions on the mental health of the Thai population. Instances of depression and increased suicide rates were noted, likely resulting from job loss and mandatory isolation. Collaborating with the Puey Ungphakorn Institute for Economic Research, the team sought ways to assess the mental well-being of the Thai people beyond traditional survey methods, which are limited by their frequency and scope.”
The team turned to Thai social media as a rich source for studying mental health during the COVID-19 pandemic. The team began collecting data from Thai users’ tweets on the Twitter platform (currently X) and used sentiment analysis models to extract emotional signals. The analysis provided emotional dimensions, trends in self-harm, and sentiment scores (positive/negative), aiming to understand if these signals could predict future trends.
“We took these signals collected over several months and correlated them with the number of self-harm incidents in the country. We found a high correlation between negative emotional signals and the number of suicides. However, we wanted to expand on these signals and predict actual self-harm incidents in the future. If successful, this work could empower government policymakers to design intervention mechanisms to aid affected individuals. Thus, we conceptualized the ‘FAST’ framework.”
‘FAST’ is a framework designed to forecast self-harm patterns at the national level by analyzing mental signals obtained from a large volume of social media data. We collected data from Twitter, extracting emotional signals in various dimensions using sentiment analysis AI. The resulting data includes emotions in eight dimensions, trends in suicide, and sentiment scores. These signals are then used to train machine learning models for predicting future self-harm incidents. The research tested FAST on two scenarios: self-harm resulting in injuries and self-harm leading to death, achieving a prediction accuracy improvement of 36-43% compared to traditional methods.
Assoc. Prof. Dr. Suppawong Tuarob also addressed challenges faced during the research, such as the difficulty in training AI to extract sentiment from Thai text due to a lack of Thai-language training data. “We initially attempted to use a translation program but found it ineffective due to informal language used on online platforms. The team then adopted a language-agnostic model that processes text in multiple languages, using a common language for translation. This improved AI efficiency and addressed the lack of Thai training data.”
In conclusion, Assoc. Prof. Dr. Suppawong Tuarob emphasized the interdisciplinary nature of the research, merging AI and social sciences. He suggested that increased utilization of such data could benefit both academia and national development. The research not only contributes to mental health discussions but also opens avenues for predicting various economic indicators, including GDP, labor market, or economic expectations. The approach could be applied beyond mental health, providing valuable insights into societal thoughts and behaviors.
For more details and to access the published work, please visit: https://ieeexplore.ieee.org/document/10162177
Follow Assoc. Prof. Dr. Suppawong Tuarob ‘s research at: