Impact of ByteDance crisis communication strategies on different social media users

Effective crisis communication is crucial for organizations navigating online controversies. While various crisis management methods exist, few address online contexts comprehensively. The Social-Mediated Crisis Communication (SMCC) model offers a framework for studying online crisis management, considering information source and form's influence on responses and outlining social media crisis response strategies. SMCC builds upon the Situational Crisis Communication Theory (SCCT) by incorporating internet-based media, offering insights into crisis origins, information transmission, and response strategies. Existing SMCC-based research emphasizes the impact of information source (organizational vs. third-party) on public sentiment, with official accounts generally preferred. However, it lacks analysis of platform-specific differences in public opinion and enterprise preferences. The US sanctions on TikTok in July 2020 created an existential crisis for ByteDance in the US and a trust crisis in China. ByteDance's crisis communication announcements, primarily on its Toutiao platform, sparked substantial discussions, particularly on Weibo. This study aims to compare the impacts of ByteDance's crisis communication on Toutiao and Weibo users, employing machine learning and semantic network analysis to examine the quantitative and qualitative aspects of user responses.

This study's foundation lies in the SMCC model, which categorizes communication media (social media, traditional media, offline word-of-mouth) and identifies five crisis management factors (crisis origin, type, infrastructure, message strategy, message form) influencing communication during crises. The SMCC model clarifies how crisis origin influences responsibility attribution and public acceptance of responses. Crisis types (victim, accidental, intentional) dictate appropriate communication strategies. Infrastructure refers to the organizational structure for crisis management, while message strategy encompasses denial, diminishing, and rebuilding approaches. Message form includes social media, traditional media, and word-of-mouth. Social media users are categorized as influential creators, followers, and inactives. The study acknowledges the homogeneity problem in previous social media research, emphasizing the necessity of platform-specific analysis. Existing research on Chinese social media crisis communication, primarily focusing on Weibo, often utilizes SCCT or attribution theory. Machine learning-based sentiment classification and semantic network analysis (SNA) are employed to analyze user sentiments and identify thematic patterns in comments. Lexicon-based methods are deemed insufficient for identifying sentiment targets, prompting the use of machine learning for more nuanced analysis.

The study adopts a four-step methodological framework: SMCC analysis, data collection, data processing, and sentiment analysis with SNA. SMCC analysis qualitatively examines organizational and social media aspects, identifying crisis origin, type, infrastructure, message strategy, message form, and influential social media creators. Data was collected from Toutiao (ByteDance's official account) and Weibo (top five accounts generating highest discussion volume for each of five selected ByteDance announcements). Octoparse was used for data crawling. Manual data annotation was performed by five PhD students, classifying comments as positive, neutral, or negative based on ByteDance's perspective, achieving a high level of inter-annotator agreement (0.9 using Holsti's coefficient). Machine learning (logistic regression, KNN, random forest, SVM, Bayes) and deep learning (LSTM, BERT, ERNIE) models were trained and evaluated to identify the best-performing sentiment classifier. Hyperparameter tuning was done using grid search. ERNIE, a pre-trained model, showed the best performance (F1 = 82.4%). SNA, using the TF-IDF algorithm and Gephi software, was performed on the sentiment-classified data to identify thematic patterns in user comments. The Louvain algorithm was used for clustering and the identification of clustering topics required the subjective interpretation of the context by domain experts in SNA.

SMCC analysis revealed that the ByteDance crisis stemmed from external factors (US sanctions), classifying ByteDance as a victim. ByteDance used a centralized communication infrastructure on Toutiao, employing denial and diminishing strategies. Influential Weibo users included organizations, media outlets, journalists, and individual users, based on retweet counts. Sentiment analysis showed that ERNIE outperformed other models in sentiment classification. Following the fourth crisis communication, Toutiao users displayed significantly increased positive sentiment (49.16% to 86.15%), while Weibo users showed a sharp decrease (72.31% to 24.34%), with a concurrent rise in negative sentiment. SNA revealed distinct thematic patterns in positive, neutral, and negative comments on both platforms. Toutiao users largely supported ByteDance's actions, while Weibo users expressed skepticism and criticism, often linking patriotism to corporate behavior. Weibo users displayed a tendency to anthropomorphize ByteDance's actions, perceiving the company as cowardly.

The contrasting sentiments on Toutiao and Weibo highlight the limitations of focusing solely on one platform. The findings support the idea that information source and access channel significantly influence public perception during a crisis. Toutiao users, directly exposed to ByteDance's official communication, showed more positive sentiment, while Weibo users, receiving information from third-party accounts, exhibited greater negativity and criticism. This extends existing research by demonstrating the interplay between crisis origin, information source, and platform-specific user characteristics in shaping public opinion. The study also reveals a tendency for social media users to anthropomorphize corporate entities, interpreting crisis communication as reflections of the company's character.

This study contributes to the field of crisis communication by highlighting the importance of a multi-platform approach, emphasizing the need to consider platform-specific characteristics and user attributes. The findings suggest that companies should not overlook influential platforms in their crisis communication strategies. Further research could explore the nuances of platform-specific user attributes and their influence on crisis communication effectiveness across different cultural contexts. Additionally, more research on the anthropomorphic perception of companies by social media users could inform the development of more effective crisis communication strategies.

The study focuses on a single case (ByteDance's response to TikTok sanctions), limiting the generalizability of findings. The use of the TF-IDF algorithm, while simple and efficient, may not fully capture the complexity of language and context in user comments. The manual annotation process, while striving for objectivity, may still contain inherent biases. Further, while the study identified influential users, it did not delve into the network structure and interactions among those users. Future research could explore these aspects further.