This study employs machine learning to analyze news article sentiment, developing a stock market sentiment index (SSI) based on this analysis. By examining the textual data from news articles, which constitute unstructured data, we aimed to capture the prevailing sentiments among market participants across the financial market. Specifically, this study utilizes The BERT model to decipher the psychological sentiment embedded in the articles through its contextualized understanding of the tone and language patterns. The variables tested included the risk aversion estimate, calculated using the VKOSPI and Bekaert’s method for assessing risk aversion. The empirical analysis involving the SSI, VKOSPI, and risk aversion reveals a significant negative impact of SSI on VKOSPI and risk aversion. We further find that the news sentiment index (NSI) and SSI simultaneously exhibit a converging trend.