External Financial Knowledge Fusion for Accurate Financial Sentiment Analysis

Financial Sentiment Analysis (FSA) seeks to deduce the sentiment contained in financial texts (FT) like news articles and commentaries, enabling investors to form more precise assessments of market trends. For both fintech firms and investors, accurately identifying whether the sentiment in financial texts is positive or negative holds great significance for their investment choices. Nevertheless, existing FSA models are solely centered on the text itself and encounter a key challenge: Neglect of external knowledge. That is, they fail to take into account external information related to financial texts, such as background context and sentiment captions. This oversight leads to subpar performance when dealing with scenarios where the sentiment expressed is ambiguous. To tackle this problem, we put forward a new FSA approach based on the integration of external financial knowledge, named ExtFSA. It consists of two main parts: (1) We utilize advanced large language models (LLMs) to generate background knowledge and sentiment captions associated with financial texts, thereby building external knowledge representations that facilitate sentiment comprehension. (2) We integrate these two types of external knowledge with the original text, ultimately enhancing the accuracy of sentiment classification. Experimental findings on the benchmark dataset FiQA demonstrate that ExtFSA outperforms baseline models by a significant margin, achieving more interpretable and precise financial sentiment analysis.