The Bank of Korea Announces 'News Sentiment Index'... Analyzing News Sentences with Machine Learning

[Asia Economy Reporter Kim Eun-byeol] The Bank of Korea will begin trial publication next week of the "News Sentiment Index (NSI)," which analyzes and indexes sentences used in economic news reports. This decision comes from the need for rapidly released economic indicators amid the COVID-19 economic crisis.

On the 1st, the Bank of Korea stated, "Starting next week, the NSI will be trial published on the Bank of Korea Economic Statistics System (ECOS), and once the Statistics Korea pilot statistics system is established, we plan to proceed with the approval process." During the trial publication period, they plan to examine the existence of feedback phenomena where the index is influenced again after being published and appearing in economic news. The NSI is expected to be released once a week.

The NSI classifies economic news data at the sentence level into positive, negative, and neutral sentiments and indexes the differences between these sentences. An index above 100 indicates more positive sentences, while below 100 indicates more negative sentences. The daily index released is created using news articles from the previous seven days. The analyzed news articles target those from 2005 onward, including articles from about 50 media outlets in recent times.

Notably, unlike typical sentiment indices that analyze only news headlines, the NSI randomly samples sentences and uses machine learning to analyze the meaning contained in those sentences. For example, a sentence like "The government is pessimistic about domestic demand recovery" would be classified as negative, while "The government set the gross domestic product (GDP) forecast at 2.1%" would be classified as neutral. Humans first classify the characteristics of sentences, and then machines learn and independently judge sentences to create the index.

The Bank of Korea explained that the NSI leads major economic sentiment indices and real economy indicators by one to two months and shows a high correlation. The monthly News Sentiment Index showed a correlation coefficient above 0.7 with the Consumer Confidence Sentiment Index (CCSI) and major CSI items, as well as a high correlation with the Business Survey Index (BSI). It led the Composite Leading Index by one month with a correlation of 0.67 and showed a correlation of 0.60 with the GDP growth rate (real seasonally adjusted quarter-on-quarter growth rate).

Park Yang-su, Director of the Economic Statistics Bureau at the Bank of Korea, said, "There have been many calls for timely data amid the COVID-19 situation," adding, "Because economic conditions are changing rapidly, it is necessary to respond to economic policies with quickly released data." He also added, "The Bank of Korea is conducting work analyzing card consumption and barcode recognition using big data."

This content was produced with the assistance of AI translation services.