Title: Machine-Learning-Based News Sentiment Index (NSI) of Korea

Author: Beomseok Seo(BOK), Younghwan Lee(BOK), Hyungbae Cho(BOK)

 We develop the Korean news sentiment index(NSI) that measures the economic sentiment of Korea by computing it daily from the news texts scrapped from the internet. We use a set of natural language processing techniques and develop a state-of the-art transformer-neural-network-based sentiment classifier particularly designed for computing NSI of Korea. The proposed model handles large news samples effectively and computes NSI efficiently. NSI is more frequently and immediately compiled than official indices based on monthly surveys, and hence, helps to identify changes in economic sentiments before the official statistics are released. Also, NSI provides explanations for why the economic sentiments fluctuate via its keyword analysis and sector indices. NSI is designed to be compiled automatically. We assess the validity and utility of NSI from multiple perspectives. The assessments support our findings that NSI is useful as a leading index and informative to find inflection points in economic sentiments.