Online Hostility Towards UK Politicians: A Deep Dive

In recent years, social media has become a popular platform for politicians in the UK to interact with the public. They use sites like X (formerly known as Twitter) to engage with constituents, answering questions and receiving feedback. However, this openness can lead to unwanted attention. Politicians often face a wave of hostile comments directed at both their professional roles and personal Identities, making social media a double-edged sword.

This Hostility can harm politicians and the public's trust in government. Some comments are so severe that they can incite real-world violence. Therefore, understanding and addressing this issue is crucial for maintaining healthy Political discourse.

#The Dataset

To tackle the problem of hostility in online comments aimed at politicians, researchers have created a dataset containing 3,320 Tweets collected over a two-year period. These tweets were carefully reviewed and labeled for their degree of hostility towards UK Members of Parliament (MPs). Also, the dataset includes details about the identity characteristics of the targets, such as race, gender, and religion.

This dataset is not just a collection of random tweets. It aims to spotlight the unique language and issues that arise in political discussions in the UK, which can be quite different from those in other countries. For example, certain issues like Brexit are especially relevant in the UK, and this dataset reflects that.

#Why Is This Important?

The need for this type of dataset arises from the specific language used in political hostility. Existing models for detecting general hostility often fall short when applied to political contexts. They miss out on the nuances of language and public sentiment related to political issues, making it essential to have a more focused approach.

Without this focused effort, public trust in political institutions could continue to erode. Thus, creating and analyzing this dataset not only helps in classifying hostile tweets but also opens doors for future research in understanding online abuse in a political context.

#Previous Research

Before this dataset was established, previous studies had looked into hostility towards politicians, but often in a general sense. Many of these studies focused on specific incidents or trends rather than providing a comprehensive analysis of the language and identity issues at play.

Research highlighted that female politicians and those from minority backgrounds tend to face more hostility than their counterparts. Instruments like sentiment analysis have been utilized to gauge negative sentiment online, but they aren't always effective in the political realm.

Existing Datasets have often lacked labels to identify the specific nature of hostility. Some datasets focused only on one type of abuse, such as Islamophobia, while others included a broader range of hate speech but didn’t pay attention to identity characteristics.

#Methodology

#Data Collection

The researchers used X's Streaming API to gather tweets related to MPs over two years. They tracked both the original tweets from the MPs and the subsequent replies and retweets. This extensive approach resulted in over 30 million tweets. However, because this number was overwhelming, the researchers had to sample a smaller, manageable subset for detailed analysis.

#Sampling Process

To ensure diversity, the researchers chose tweets from 18 MPs representing different identities and political parties. They balanced the sample to include both minority and majority identity groups. The sampling also focused on various time periods to capture different contexts and events.

In total, 3,330 tweets were collected for manual labeling. Tweets were categorized based on hostility, allowing the researchers to create a clearer picture of the landscape of online abuse directed at MPs.

#Annotation Process

Researchers formulated guidelines to help the annotators classify the tweets effectively. A series of training sessions ensured that everyone involved understood the definitions and criteria for identifying hostility accurately. The annotators worked in teams and were encouraged to consult external resources when they encountered unfamiliar language.

Three different annotators labeled each tweet, providing a certain level of reliability to the dataset. This multiple annotation process helped minimize errors and ensured that the labels were as accurate as possible.

#Analyzing the Tweets

#Linguistic Patterns

To understand the language used in hostile tweets, researchers conducted a linguistic analysis. They found that hostile tweets often contained negative terms and phrases aimed at discrediting politicians. Words like "liar," "corrupt," and "evil" were notably common among hostile comments.

On the flip side, non-hostile tweets tended to feature positive phrases. Instead of insults, these tweets often expressed gratitude or constructive feedback, using language that adhered to social norms.

#Topic Analysis

The researchers also explored the topics associated with both hostile and non-hostile tweets. They identified that many tweets related to current political events, like Brexit or the handling of healthcare during the pandemic. This connection between current events and online hostility underscores how certain issues can escalate public anger toward politicians.

#Hostility Identification

Researchers used the dataset to train models for detecting hostility in tweets. This involved two main tasks: first, identifying whether a tweet was hostile or not, and second, categorizing the type of hostility based on identity characteristics like race, gender, or religion.

Multiple models were tested to see which performed best in identifying both binary hostility (hostile vs. non-hostile) and multi-class hostility types.

#Findings

#Results of Hostility Detection

When analyzing the performance of models, researchers found that certain models, like RoBERTa-Hate, performed particularly well in detecting hostility, achieving a high macro F1 score. It became clear that models trained on the dataset using confidence scores yielded better results than those trained on previous datasets.

#Trends in Hostility

A notable trend found in the data is that politicians from specific identity backgrounds, such as women and those from minority races or religions, often received a higher volume of hostility. This highlights the intersection of various identities, where the combination of race, gender, and religion can amplify the amount of abuse faced by politicians.

#Importance of Context

The research also demonstrated that the context in which a tweet was sent played a significant role in determining the language used. Hostility often peaked around significant political events, revealing the close relationship between social commentary and politics.

#Conclusion

The creation of this dataset is a step toward better understanding and identifying online hostility aimed at UK politicians. It highlights the need for specialized tools to effectively tackle this issue in a political context.

By focusing on the language and identity characteristics involved in hostile comments, researchers can glean vital insights that pave the way for future research aimed at reducing online abuse.

As social media continues to evolve, so too must our approaches to engaging with the public and addressing the hostile sentiments that can emerge from it.

Now, if only MPs could come equipped with a thick skin, a sense of humor, and maybe a digital shield, they might just survive the maelstrom of online comments!