Leveraging Large Language Models for Predictive Analysis of Human Misery
Bishanka Seal, Rahul Seetharaman, Aman Bansal, Abhilash Nandy
2025-08-20
Summary
This research explores how well Large Language Models (LLMs) can guess how miserable someone feels based on written descriptions of real-life situations, scoring this misery on a scale from 0 to 100. They tested different ways of asking the LLMs for these predictions and came up with a fun, game-show-like method to test the LLMs' abilities more deeply, including how they learn from mistakes.
What's the problem?
The main challenge is figuring out if LLMs can accurately understand and numerically represent human-perceived misery from text, which is a complex emotional concept. Existing methods for testing these models are often static and don't show how they might improve with feedback.
What's the solution?
The researchers experimented with different ways to prompt the LLMs, such as giving them no examples (zero-shot) or giving them a few examples to learn from (few-shot), finding that providing examples improved results. They also created a unique "Misery Game Show" which presents tasks in a more dynamic and interactive way, like comparing misery levels or adjusting predictions based on given feedback.
Why it matters?
This work is important because it shows how LLMs can be used to understand and potentially even help with complex human emotions expressed in text. The innovative gamified approach offers a better way to test and improve these models for tasks involving emotional intelligence and dynamic reasoning, moving beyond simple prediction.
Abstract
This study investigates the use of Large Language Models (LLMs) for predicting human-perceived misery scores from natural language descriptions of real-world scenarios. The task is framed as a regression problem, where the model assigns a scalar value from 0 to 100 to each input statement. We evaluate multiple prompting strategies, including zero-shot, fixed-context few-shot, and retrieval-based prompting using BERT sentence embeddings. Few-shot approaches consistently outperform zero-shot baselines, underscoring the value of contextual examples in affective prediction. To move beyond static evaluation, we introduce the "Misery Game Show", a novel gamified framework inspired by a television format. It tests LLMs through structured rounds involving ordinal comparison, binary classification, scalar estimation, and feedback-driven reasoning. This setup enables us to assess not only predictive accuracy but also the model's ability to adapt based on corrective feedback. The gamified evaluation highlights the broader potential of LLMs in dynamic emotional reasoning tasks beyond standard regression. Code and data link: https://github.com/abhi1nandy2/Misery_Data_Exps_GitHub