Financial language models improve sentiment using knowledge graphs
FinLLM-KG: Knowledge Graph-Enhanced LLMs for Financial Sentiment Analysis in Algorithmic Trading
Computation and LanguageMachine Learning
Summary
Financial sentiment analysis helps predict whether financial news about companies is positive or negative, which can guide trading decisions. The authors developed a method called FinLLM-KG that combines large language models with financial knowledge graphs from official company filings. This allows the system to consider detailed company information and the timing of events when analyzing sentiment, making the predictions more accurate and easier to understand. The approach also makes it possible for humans to see which parts of the text or facts influenced the sentiment scores.
What this means in practice
- •For algorithmic trading teams: Improve trading strategy decisions by integrating company-level financial facts into sentiment analysis of news articles.
- •For financial compliance teams: Provide transparent sentiment scores with evidence attribution to support regulatory review and risk assessment in finance.
Authors
Giorgos Iacovides, Oliver Hannibal, Wuyang Zhou, Danilo Mandic
Abstract
Recent advances in Generative AI have substantially improved financial sentiment analysis through post-trained financial large language models (LLMs). However, existing approaches remain confined to predicting the sentiment of each article in isolation and operate as black boxes, which limits contextual understanding and hinders deployment in high-stakes financial decision-making. To this end, we introduce FinLLM-KG, a model-agnostic framework that combines causal LLMs with a financial knowledge graph, in order to inject temporally grounded, company-level knowledge from SEC 10-K filings at inference time, without updating model weights. Within FinLLM-KG, context is supplied either as a natural-language company profile or as a small set of relevance-ranked factual triples. This enables sentiment predictions to account for broader firm-level information rather than the article alone, at almost zero computational cost and for any causal LLM. In addition, attributing each article-level sentiment score to individual sentences and to the retrieved knowledge-graph facts makes the evidence behind the prediction inspectable, thereby enhancing explainability and supporting human-in-the-loop review. Simulations demonstrate that FinLLM-KG consistently improves portfolio performance over the state-of-the-art baseline.