Developing a supervised learning-based social media business sentiment index.

Area of Science:

• Computational Social Science
• Big Data Analytics
• Sentiment Analysis

Background:

• The digital era generates vast amounts of big data, with social media contributing significantly.
• Understanding public sentiment expressed on social media is crucial for economic analysis.
• Sentiment analysis offers a method to gauge attitudes and emotions from online social network services.

Purpose of the Study:

• To investigate the relationship between public sentiment expressed on social media and actual economic performance.
• To evaluate the effectiveness of sentiment analysis techniques in predicting economic indicators.

Main Methods:

• Trained sentiment classifiers using diverse social media datasets and feature sets (feature vector, sequence vector, combined).
• Assessed the performance of six machine learning classifiers: MaxEnt-L1, C4.5 decision tree, SVM-kernel, Ada-boost, Naïve Bayes, and MaxEnt.
• Collected economic-related keywords from social media and employed vector auto-regression analysis to confirm findings.

Main Results:

• A statistically significant relationship was found between public sentiment and economic performance.
• Negative sentiments, specifically 'depression' and 'unemployment,' were shown to influence the Korea Composite Stock Price Index (KOSPI).
• Extracted keywords like 'price,' 'year-end-tax,' and 'budget deficit' were identified as drivers of exchange rate fluctuations.

Conclusions:

• Sentiment analysis of social media data provides valuable insights into economic dynamics.
• Public opinion, as captured through sentiment analysis, can serve as a leading indicator for economic trends.
• The study confirms the predictive power of social media sentiment on key economic indicators like stock market performance and exchange rates.