Summary
Detecting suicide risk from social media posts is challenging because data are limited and decisions must be careful. The authors tested many common techniques used in language models to see which really help in this sensitive task. They found that most popular tricks don’t improve results unless they are chosen carefully based on how the task works. By designing models that use task-specific knowledge, such as linking posts to clinical definitions and evidence, the authors built a system that predicts suicide risk and related factors more accurately. This approach ranked third in a large competition and shows the value of tailoring methods to the problem.
suicide risk assessmentnatural language processingclass imbalanceensemble modelingclinical risk factorsevidence taggingtask-conditioned technique selectionmodel calibrationmacro-F1 score
Authors
Shlok Shelat, Shrey Salvi, Souvik Roy, Manas Gaur, Amit Sheth
Abstract
Assessing suicide risk from social media text is a small-data, high-stakes setting requiring not only severity prediction but also supporting evidence and clinically relevant risk and protective factors. Yet common NLP techniques, including model scaling, synthetic data, loss reweighting, ensembling, and threshold tuning, are often applied without testing whether their gains hold up under severe class imbalance, coupled outputs, and limited author-level data. We study 1,635 clinician-annotated posts and audit 31 pre-specified techniques from 7 methodological families through roughly 300 controlled experiments on author-disjoint partitions. We found no prior audit of this playbook in this regime. The findings guide a task-grounded system for three outputs: 4-level suicide risk, evidence spans, and 24 clinical risk and protective factors. Only 5 of 31 comparisons produced reliable gains. We reformulate factor prediction as entailment between each post and its codebook definitions, using an architecturally diverse ensemble with class-balanced training and score rescaling. Risk predictions condition a 7-model evidence tagger ensemble; evidence restricts symbolic risk rules; and a difficult risk class is routed separately. The factor predictor remains independent because risk evidence provides no additional factor signal. We also correct a mismatch between validation scores used for threshold fitting and test-time ensemble scores through deployment-consistent calibration, yielding the largest improvement to the factor system. The final system achieves 0.8203 for risk, 0.7953 for evidence, and 0.7045 macro-F1 for factors, with a 0.7781 composite, ranking third among 53 teams. We call the underlying principle task-conditioned technique selection: retain techniques only when task-specific knowledge, structure, or empirical evidence justifies them.