H2HTalk : Evaluating Large Language Models as Emotional Companion
Boyang Wang, Yalun Wu, Hongcheng Guo†, Zhoujun Li† · 2025
Read H2HTalk : Evaluating Large Language Models as Emotional Companion by Boyang Wang, Yalun Wu, Hongcheng Guo†, Zhoujun Li† (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-03543.txt.
Description
Boyang Wang and co-authors introduce H2HTalk in a 2025 paper on judging chatbots that act as emotional companions. More people turn to large language models for comfort and support, yet tests of how safe and caring these systems are have lagged behind the models. The benchmark looks at both personality growth and warm, understanding replies.
The test set holds more than four thousand crafted scenes, spanning chats, memory of past talks, and planning tasks that mirror real support talks. A safety module based on attachment theory aims to keep exchanges healthy. The team ran fifty models through the same protocol. Long-term planning and memory stood out as weak points, mostly when user needs were unstated or shifted mid-chat. All materials are released for others to use.
Catalog record
| CREATOR | Boyang Wang, Yalun Wu, Hongcheng Guo†, Zhoujun Li† |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-03543