Search Engines in the use of Financial Sentiment Analysis

Abstract

Financial market prediction often rely on historical and numerical data, but recent advancements in large Language models encourage the use of alternative datasets like fnancial news text. However, this methodology often faces limitations due to the scarcity of extensive datasets that combine both quantitative and qualitative sentiment analyses. To address this gap, we used the Bing Search API to build a dataset comprising over 100.000 financial news articlesfrom more than 90 websites. Our work aims to illuminate the process of Building a data set using search engines, demonstrating that the use of keywords to collect ”custom” data from the vast Internet is an effective alternative for data collection. We evaluated the dataset using a sentiment index, which we later compared with the S&P 500 stock index. We concluded that while news sentiment may not immediately reflect price variations, it can effectively indicate broader market trends.