FinEval-KR: A Financial Domain Evaluation Framework for Large Language Models' Knowledge and Reasoning

TL;DR

This paper introduces FinEval-KR, a framework for separately evaluating large language models' financial knowledge and reasoning abilities, along with a new dataset, revealing key factors affecting reasoning accuracy.

Contribution

The paper presents a novel evaluation framework that decouples knowledge and reasoning in LLMs, introduces a cognitive score based on Bloom's taxonomy, and provides a new Chinese financial reasoning dataset.

Findings

LLM reasoning and higher-order cognitive abilities are crucial for accuracy.

Top models still struggle with knowledge application.

Specialized financial LLMs lag behind general models.

Abstract

Large Language Models (LLMs) demonstrate significant potential but face challenges in complex financial reasoning tasks requiring both domain knowledge and sophisticated reasoning. Current evaluation benchmarks often fall short by not decoupling these capabilities indicators from single task performance and lack root cause analysis for task failure. To address this, we introduce FinEval-KR, a novel evaluation framework for decoupling and quantifying LLMs' knowledge and reasoning abilities independently, proposing distinct knowledge score and reasoning score metrics. Inspired by cognitive science, we further propose a cognitive score based on Bloom's taxonomy to analyze capabilities in reasoning tasks across different cognitive levels. We also release a new open-source Chinese financial reasoning dataset covering 22 subfields to support reproducible research and further advancements in financial reasoning. Our experimental results reveal that LLM reasoning ability and higher-order cognitive ability are the core factors influencing reasoning accuracy. We also specifically find that even top models still face a bottleneck with knowledge application. Furthermore, our analysis shows that specialized financial LLMs generally lag behind the top general large models across multiple metrics.