Categorizing Philanthropic CSR Activities Through Hybrid Approach of Fuzzy Clustering and Topic Modeling

This research paper presents a new approach for systematically categorizing philanthropic corporate social responsibility (PCSR) initiatives using document clustering and topic modeling techniques. The primary aim is to efficiently organize and analyze the substantive content within corporate reports. The study leverages a five-year comprehensive dataset, sourced from the annual reports of 19 esteemed CSR-award recipient companies in Malaysia. These reports undergo rigorous transformation into a structured format and followed by meticulous text pre-processing to ensure data integrity. In the pursuit of precision, this study identifies the most suitable Latent Dirichlet Allocation (LDA) topic modeling technique, subsequently integrating it with document clustering using K-Means and Fuzzy C-Means (FCM) clustering. The findings reveal that FCM outperforms K-Means by delineating thirteen distinct clusters, each representing various dimensions of PCSR activities undertaken by the distinguished CSR-award companies in Malaysia. This superiority of FCM is particularly evident in its capability to handle datasets characterized by noise and overlap issues. Additionally, this research enriches comprehension by compiling ten keywords for each cluster, facilitating a more comprehensive and nuanced insights of PCSR topics.