Understanding Negative Word of Mouth during Crisis: Public Responses to Corporate Apologies using Topic Modeling, Semantic Network Analysis, and Sentiment Analysis
Sejung Park, Jin-A ChoiCorporate crisis is an unexpected occurrence which poses a reputational threat to the organization. The recent trend of commenting on or sharing posts about crises prompts social media users to form behavioral intentions such as spreading negative word of mouth as well as boycotting the organization. This study aims to assess the effectiveness of two different crisis communication strategies by analyzing public comments using semantic network analysis, topic modeling, and sentiment analysis. The study examines a case in which United Airlines sparked public outrage by forcibly removing a passenger from an overbooked flight on April 9, 2017. This incident quickly went viral and received much criticism on social media. United attempted to repair their damaged reputation on social media by offering apologies. The study examined public perception and sentiment of the public’s reaction to two types of corporate apologies. Through the lens of situational crisis communication theory (SSCT), a content analysis of response messages United Airline posted on Facebook after the crisis was conducted to analyze which crisis response strategies were implemented. We performed topic modeling based on LDA with ConText to classify the thematic trends in public responses. Semantic network analysis was conducted to compare the framing patterns in word-of-mouth communication of the stakeholders in response to crisis strategies and public perceptions toward the brand. Core-and-periphery structures in the semantic networks were also identified. In addition, we examined the intensity of positive and negative emotions expressed in the public comments to United Airlines’ post crisis communication. Both network and textual data were collected using the Facebook API. The results of the content analysis indicated that United Airlines employed a full apology and a partial apology to manage the overbooking crisis on Facebook over time. This study analyzed a total of 73,592 replies created by 67,518 users to a partial apology strategy United employed on April 10, 2017 and a total of 41,343 replies generated by 30,478 users to a full apology strategy on April 11, 2017. The analysis of core and periphery indicated that the words representing anti-purchase intentions were core members in the semantic network of the partial apology strategy. Similarly, boycott-related words and the terms expressing concerns about the victim were central members in the network of the full apology strategy. The results of topic modeling and semantic network analysis revealed that public responses to both apology strategies were primarily negatively associated with physical violence against the passenger, concern for the victim, and hypocritical apology. The results indicate that compared to a full apology strategy, stakeholders generated larger and loosely connected word-of-mouth network in response to the partial apology and the comments were more negative. These results may be attributed to the late adoption of the matched strategy (full apology) suggested by the SSCT model. Methodological triangulation for textual data analysis enabled us to offer evidence-based lessons from United on how to strategically manage stakeholders’ outcry to repair corporations’ damaged reputation after a crisis.