Examining CSR communication on social media during a victim crisis: a machine learning based text analytics approach

Purpose This study examines how top global brands changed their corporate social responsibility (CSR) communication on social media during a victim crisis, and how their CSR communication on social media influenced consumer sentiment. Design/methodology/approach Using 18,502 firms’ Facebook posts and their most relevant consumer comments from pre-pandemic and during-pandemic timeframes, this study integrates machine learning techniques (BERTopic) with human-based qualitative analysis to analyze CSR posts. It also measures the polarity and magnitude of consumer sentiment with Google Natural Language AI. We tested seven hypotheses using Hierarchical Linear Modeling (HLM). Findings The machine learning-based topic modeling analysis showed that firms increased CSR communications intensity on social media and they more intentionally chose different CSR communication strategies for different topics on social media during the victim crisis. The hypothesis testing results show proactive, accommodative and interactive strategies have a significant impact on consumer sentiment polarity and magnitude, and these effects are moderated by the level of interactivity and industry type. Originality/value (1) This study takes a dynamic view to examine the firms’ CSR communication on social media during a victim crisis. It used machine learning-based text analytics and found many interesting results on how firms changed their CSR communication topics and strategies on social media during the crisis. (2) It measures both consumer sentiment polarity and sentiment magnitude to conduct sentiment analysis. The results indicate that the CSR communication strategies have different impacts on the two sentiment components. (3) It integrates machine learning techniques with human-based qualitative analysis. It shows how researchers can gain the benefits of both approaches.