Evaluation of LLMs-based Hidden States as Author Representations for Psychological Human-Centered NLP Tasks
Nikita Soni, Pranav Chitale, Khushboo Singh · 2025
Read Evaluation of LLMs-based Hidden States as Author Representations for Psychological Human-Centered NLP Tasks by Nikita Soni, Pranav Chitale, Khushboo Singh (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-2503-00124.txt.
Description
Nikita Soni and co-authors asked in 2025 how best to turn a writer's text into numbers that capture the writer's state of mind. Language models store inner states as they read, and those states can serve as a portrait of the author. Yet teams differ on which part of the model to draw from.
The team tested many ways to build portraits of single texts and of whole users, across tasks such as mood, empathy, and distress. Averaging the inner states of all words in a text worked best in most cases. A stored portrait of the user rarely won on its own, but adding it made other portraits stronger. The results guide teams building tools that read people, not just text.
Catalog record
| CREATOR | Nikita Soni, Pranav Chitale, Khushboo Singh |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-00124