On the Diversity of Analogy Making in Large Language Models

2026-08-04

AI Summary

This study addresses the pervasive limitation of insufficient output diversity in large language models (LLMs) when generating analogies, which manifests as constrained cross-domain connections and hinders their potential for scientific innovation. The work presents the first systematic evaluation of analogy generation diversity across ten state-of-the-art LLMs, integrating quantitative diversity metrics, causal information flow analysis, and cross-model comparative experiments. It reveals that model-sensitive regions critically influence the trade-off between diversity and output quality. The findings demonstrate that mainstream models exhibit high homogeneity in target domains, and existing approaches to enhancing diversity often compromise output fidelity. These insights provide both theoretical grounding and empirical evidence to guide future improvements in analogy generation capabilities of LLMs.

Abstract

Large Language Models (LLMs) have demonstrated remarkable potential for analogy making, a core cognitive capability that drives novelty and creativity. While prior research has extensively investigated the applications and underlying mechanisms of LLM-based analogy making, its output diversity remains largely unexplored, despite being essential for broadening cross-domain connections and fostering scientific innovation. In this work, we present a comprehensive evaluation of analogy diversity across ten state-of-the-art open- and closed-source LLMs. Our findings highlight a concerning issue of domain homogeneity, a prevalent tendency for LLMs to generate analogies from a narrow set of target domains, limiting both inter-query and intra-model diversity. Furthermore, our analysis reveals a fundamental trade-off in existing LLM diversity-enhancement methods: increasing output diversity often comes at the expense of output quality. Finally, our causal analysis of LLM information flow reveals substantial differences in the model-sensitive regions governing analogy diversity across LLMs, suggesting a potential mechanism for the observed diversity-quality trade-off. To our knowledge, this is among the first studies to systematically investigate output diversity in LLM-based analogy making.

Problem

• analogy making
• output diversity
• domain homogeneity
• large language models
• diversity-quality trade-off

Innovation

• analogy diversity
• large language models
• domain homogeneity
• diversity-quality trade-off
• causal analysis