A time-aware transformer based model for suicide ideation detection on social media

Social media platforms have become a critical space for mental health discourse, making automated suicide ideation detection increasingly important for early intervention. We present a time-aware transformer-based model that leverages temporal patterns in user behavior to improve suicide ideation detection on social media. Our approach incorporates both the content of posts and the temporal dynamics of posting behavior, recognizing that changes in posting frequency and timing can be indicative of mental health states. The model uses a transformer architecture enhanced with temporal encoding mechanisms to capture long-range dependencies in user posting sequences. We evaluate our approach on real-world social media datasets and demonstrate significant improvements over baseline methods that do not consider temporal information. Our time-aware model achieves better precision and recall in identifying users at risk, potentially enabling more timely and effective mental health interventions.