Machine learning-based proactive social-sensor service for mental health monitoring using twitter data

Abstract

The social media platforms are considered an ecosystem of social sensors where each social media platform user is treated as a social sensor cloud. To overcome the limitations of large-scale mental health surveillance using traditional health administration systems. Although the existing approaches in the literature provide an online detection of mental illness, these are challenging to apply in early detection. Focusing on the Twitter platform, a generic framework is designed in this paper to support proactive mental health monitoring. Detailed data cleaning and pre-processing of the tweets are offered using regular expressions based on observed patterns to ensure accurate results in sentiment analysis. A machine learning mechanism, especially LSTM, is applied for the early detection of at-risk social sensors based on custom event definitions to overcome the limitations of traditional classifiers. The performance of the proposed mechanism has been experimented with, and it outperforms the existing approaches in terms of accurate prediction.