Evaluating Text Creativity across Diverse Domains: A Dataset and Large Language Model Evaluator

TL;DR

This paper introduces CreataSet, a large dataset for evaluating textual creativity, and CrEval, an LLM-based evaluator that aligns well with human judgments, improving creativity assessment across diverse domains.

Contribution

It presents a novel pairwise-comparison framework and a large-scale dataset, CreataSet, to train a more accurate LLM-based creativity evaluator, CrEval.

Findings

CrEval outperforms existing methods in aligning with human judgments.

Training on both human and synthetic data enhances evaluator robustness.

CrEval effectively boosts LLMs' creative output.

Abstract

Creativity evaluation remains a challenging frontier for large language models (LLMs). Current evaluations heavily rely on inefficient and costly human judgments, hindering progress in enhancing machine creativity. While automated methods exist, ranging from psychological testing to heuristic- or prompting-based approaches, they often lack generalizability or alignment with human judgment. To address these issues, we propose a novel pairwise-comparison framework for assessing textual creativity that leverages shared contextual instructions to improve evaluation consistency. We introduce CreataSet, a large-scale dataset with 100K+ human-level and 1M+ synthetic creative instruction-response pairs spanning diverse open-domain tasks. Through training on CreataSet, we develop an LLM-based evaluator named CrEval. CrEval demonstrates remarkable superiority over existing methods in alignment with human judgments. Experimental results underscore the indispensable significance of integrating both human and synthetic data to train highly robust evaluators, and showcase the practical utility of CrEval in boosting the creativity of LLMs.