SuiChat-CN: Benchmark for Suicide Risk Assessment in Chinese Group Chats
SuiChat-CN has been launched by researchers as a benchmark for evaluating suicide risk in Chinese group chats. This initiative targets the relatively neglected domain of instant messaging platforms such as Telegram, characterized by brief, fragmented, and multi-user interactions. The benchmark was developed by gathering public data from Telegram group chats, identifying key signal words, and enhancing context in both directions. Risk levels for users were marked using a methodology that combines expert validation with LLM support. SuiChat-CN comprises 13,312 contextual segments from 1,406 users, encompassing a total of 258,228 messages. This research underscores the necessity for suicide prevention measures within group chat environments, contrasting with existing studies that primarily examine platforms like Twitter and Weibo.
Key facts
- SuiChat-CN is a Chinese group-chat benchmark for contextual suicide risk assessment.
- It uses public Telegram group-chat data.
- The benchmark contains 13,312 contextual segments from 1,406 users.
- It covers 258,228 messages.
- Annotation was done with an expert-validated, LLM-assisted paradigm.
- Existing computational studies focus on post-based platforms like Twitter and Weibo.
- Group chats pose challenges due to short, fragmented, multi-party messages.
- Suicide causes approximately 720,000 deaths each year globally.
Entities
Institutions
- arXiv