Mining Hidden Knowledge About Illegal Compensation for Occupational Injury: Topic Model Approach.

Area of Science:

• Occupational Health and Safety
• Data Science
• Sociology

Background:

• Workers' compensation in South Korea is legally mandated but subject to underreporting due to employer agreements for illegal compensation (gong-sang).
• Gong-sang practices are not officially recorded, obscuring the true extent of occupational injuries and related issues.
• This underreporting hinders accurate statistical analysis and effective intervention strategies.

Purpose of the Study:

• To analyze social media data using topic modeling to explore hidden knowledge about illegal compensation (gong-sang) for occupational injuries in South Korea.
• To identify key themes and concerns related to gong-sang that are not captured in official statistics.

Main Methods:

• Collected 2210 social media documents using the keyword 'gong-sang' from 2006 to 2017.
• Applied natural language processing and a morphological analyzer for Korean language text.
• Utilized Latent Dirichlet Allocation (LDA) topic modeling with a 10-topic model and 3000 Gibbs sampling iterations.

Main Results:

• LDA classified gong-sang documents into 10 topics, with 4 key categories identified.
• Topic 1 (60.5%) revealed worker anxiety regarding illegal compensation versus legal insurance claims.
• Topics 3 & 4 (19%) indicated monetary compensation aspects of gong-sang, while Topics 5-10 highlighted vulnerable job sectors and injury types.

Conclusions:

• Topic modeling of social media data successfully uncovered salient issues surrounding gong-sang.
• Findings offer valuable insights for improving South Korea's occupational health and safety administration.
• Recommendations are provided to protect vulnerable workers from illegal gong-sang compensation practices.