Exploration of the relationship between SDGs and CSR reports with text mining techniques for stock exchange companies over Taiwan

Authors:
Preprint
Publisher website
Google Scholar
Abstract
This research collects corporate social responsibility (CSR) reports from stock exchange companies in Taiwan and employs text mining technologies to analyze the relationship and document similarity between CSR reports from various industries and the Sustainable Development Goals (SDGs). The methods used include natural language processing (NLP), TF-IDF weighting, principal component analysis (PCA), and document similarity assessment. The study applies sub-items of selected SDG terms to screen feature words, build the TF-IDF matrix, analyze the CSR report contents using PCA, and utilize cosine similarity to compare the similarity between CSR reports and SDG sub-items. A total of 225 feature words were identified based on SDG sub-items, with the top 60 feature words (26.7%) accounting for 77.9% of the total TF-IDF weights, aligning with the Pareto principle. Analyzing 370 CSR reports from selected stock exchange companies (0050 ETF), unique and representative feature words and explained variations were identified. Each rotated principal component allowed the identification of corresponding SDG sub-items through specific feature words. The high diversity of feature words resulted in low and unique explained variance for each rotated principal component. Document similarity comparisons between CSR reports and SDG sub-items revealed confidence levels indicating the degree of alignment between CSR reports and SDG sub-items. For the natural language segmentation process and automatic document classification of CSR reports, the assistance of domain experts is recommended to ensure accurate and consistent segmentation and classification results.

This publication has 53 references indexed in Scilit: