TL;DR
This paper introduces a novel leader-follower opinion evolution model to understand how public sentiment shifts online following major maritime incidents, using the Suez Canal blockage as a case study. By combining hybrid sentiment analysis with a Hegselmann-Krause-based model, it identifies opinion leaders and tracks how their influence shapes collective opinion dynamics.
What Problem Does This Paper Address?
The Suez Canal blockage (SCB) event, a significant disruption to global trade, garnered immense public attention. While previous research has explored the economic and environmental impacts of such events, there's a gap in understanding how public opinion forms, evolves, and is influenced online, especially in the context of social media's rapid information dissemination. This paper aims to bridge this gap by developing a model that can capture public sentiment, identify key influencers (opinion leaders), and predict the trajectory of public opinion following such incidents, with the goal of better managing and guiding public discourse.
Methods
The authors propose a three-stage framework:
- Data Collection and Preprocessing: Tweets related to the SCB were collected from Twitter in both Chinese and English. This data underwent preprocessing, including removing unwanted characters, segmenting Chinese text using Jieba, lowercasing, punctuation removal, lemmatization, and stop-word removal.
- Sentiment Analysis: A hybrid sentiment analysis algorithm combining SnowNLP (for Chinese) and TextBlob (for English) was developed. This algorithm classifies online individuals into three sentiment categories: positive, neutral, and negative. The sentiment scores were then standardized to a [0,1] range to represent initial opinions.
- Leader-Follower Opinion Evolution Model: A new Hegselmann-Krause (HK)-based model was developed.
- Opinion Leader Identification: Individuals were identified as opinion leaders based on their influence, measured by the number of replies, forwards, and likes on their tweets. Leaders were further categorized into positive, neutral, and negative based on their sentiment. Followers were defined as individuals not meeting the leader influence threshold.
- Opinion Interaction and Updating: The model simulates opinion dynamics where individuals interact within a "bounded confidence" threshold. Interactions between leaders, followers, and between leaders and followers are modeled. The update rules are based on the HK model, where individuals adjust their opinions by averaging those within their confidence set, with weights influenced by the similarity of opinions and the leader-follower relationship. The process continues until a stable state is reached.
Datasets
The primary dataset used in this study consists of tweets related to the Suez Canal blockage, collected from Twitter. The data was gathered between March 21 and October 1, 2021, and included both Chinese and English tweets. The authors filtered out repetitive and official media tweets to focus on personal opinions. After preprocessing, the dataset comprised 170 opinion data points (27 Chinese, 143 English). No new benchmark datasets were introduced; existing social media data was utilized.
Key Insights and Ideas
- The Power of Opinion Leaders: The study highlights that opinion leaders, identified by their influence (engagement metrics), play a crucial role in shaping public discourse. Their opinions are not only disseminated but also influence the speed and direction of opinion convergence.
- Sentiment as Initial Opinion: The sentiment analysis of individual tweets serves as a direct proxy for their initial opinion state in the evolution model, bridging the gap between raw text and quantifiable opinion.
- Bounded Confidence is Key: The Hegselmann-Krause model's core concept of "bounded confidence" is essential. It explains how individuals only engage with and are influenced by those whose opinions are close to their own, preventing extreme polarization and facilitating consensus-building within certain groups.
- Leader-Follower Dynamics Matter: Differentiating between leaders and followers, and further categorizing leaders by sentiment, provides a more nuanced understanding of influence. Followers are shown to be more susceptible to influence, while leaders drive the discourse.
- Convergence to Neutrality: Under certain conditions, particularly with increasing follower confidence levels, the model suggests that public opinion tends to converge towards a neutral stance, indicating a potential for moderation in online discussions.
Key Results
- The study successfully implemented a hybrid sentiment analysis to classify online opinions into positive, neutral, and negative categories.
- An opinion leader identification mechanism was established, categorizing leaders into positive (3), neutral (16), and negative (2) groups, with 149 followers.
- The leader-follower opinion evolution model demonstrated that opinion dynamics can reach a stable state.
- Simulations showed that increasing the confidence level of followers generally leads to faster convergence towards a neutral opinion.
- Conversely, increasing the confidence level of leaders also accelerates opinion convergence, suggesting that higher trust within groups promotes faster consensus.
Limitations and Future Work
The authors acknowledge several limitations:
- Homogeneous Confidence Levels: The model assumes that all individuals within the same subgroup (leaders or followers) share the same confidence level. In reality, individual trust thresholds vary based on psychological and physical characteristics.
- Limited Data Scope: While Twitter was used, the study focused on a specific event, and the number of opinion data points was limited.
Future work could address these limitations by:
- Incorporating more diverse and dynamic confidence levels for individuals.
- Exploring psychological factors that influence trust and opinion formation.
- Extending the model to other types of maritime incidents or social events.
- Investigating the impact of misinformation and its role in opinion evolution.
Summary updated