Diagnosing autism spectrum disorders using a double deep Q-Network framework based on social media footprints

Social media is increasingly used in many contexts within the healthcare sector. The improved prevalence of Internet use via computers or mobile devices presents an opportunity for social media to serve as a tool for the rapid and direct distribution of essential health information. Autism spectrum disorders (ASD) are a comprehensive neurodevelopmental syndrome with enduring effects. Twitter has become a platform for the ASD community, offering substantial assistance to its members by disseminating information on their beliefs and perspectives via language and emotional expression. Adults with ASD have considerable social and emotional challenges, while also demonstrating abilities and interests in screen-based technologies.

The novelty of this research lies in its use in the context of Twitter to analyze and identify ASD. This research used Twitter as the primary data source to examine the behavioral traits and immediate emotional expressions of persons with ASD. We applied Convolutional Neural Networks with Long Short-Term Memory (CNN-LSTM), LSTM, and Double Deep Q-network (DDQN-Inspired) using a standardized dataset including 172 tweets from the ASD class and 158 tweets from the non-ASD class. The dataset was processed to exclude lowercase text and special characters, followed by a tokenization approach to convert the text into integer word sequences. The encoding was used to transform the classes into binary labels. Following preprocessing, the proposed framework was implemented to identify ASD.

The findings of the DDQN-inspired model demonstrate a high precision of 87% compared to the proposed model. This finding demonstrates the potential of the proposed approach for identifying ASD based on social media content.

Ultimately, the proposed system was compared against the existing system that used the same dataset. The proposed approach is based on variations in the text of social media interactions, which can assist physicians and clinicians in performing symptom studies within digital footprint environments.