[Paper Review] Characterizing the spread of exaggerated news content over social media

Jasabanta Patro, Sabyasachee Baruah+4 more|arXiv (Cornell University)|Nov 19, 2018

Misinformation and Its Impacts11 references4 citations

TL;DR

This paper investigates how exaggerated health news spreads on Twitter, finding that late-arriving tweets about exaggerated content show stronger use of opinion and realization words (e.g., 'feel', 'realize'), while having fewer negative or death-related terms. Using linguistic features and user behavior, the authors classify users who frequently share exaggerated content with an F1-score of 0.83, revealing distinct behavioral and linguistic patterns in such users.

ABSTRACT

In this paper, we consider a dataset comprising press releases about health research from different universities in the UK along with a corresponding set of news articles. First, we do an exploratory analysis to understand how the basic information published in the scientific journals get exaggerated as they are reported in these press releases or news articles. This initial analysis shows that some news agencies exaggerate almost 60\% of the articles they publish in the health domain; more than 50\% of the press releases from certain universities are exaggerated; articles in topics like lifestyle and childhood are heavily exaggerated. Motivated by the above observation we set the central objective of this paper to investigate how exaggerated news spreads over an online social network like Twitter. The LIWC analysis points to a remarkable observation these late tweets are essentially laden in words from opinion and realize categories which indicates that, given sufficient time, the wisdom of the crowd is actually able to tell apart the exaggerated news. As a second step we study the characteristics of the users who never or rarely post exaggerated news content and compare them with those who post exaggerated news content more frequently. We observe that the latter class of users have less retweets or mentions per tweet, have significantly more number of followers, use more slang words, less hyperbolic words and less word contractions. We also observe that the LIWC categories like bio, health, body and negative emotion are more pronounced in the tweets posted by the users in the latter class. As a final step we use these observations as features and automatically classify the two groups achieving an F1 score of 0.83.

Motivation & Objective

• To understand how exaggerated health news from scientific press releases and news articles spreads over social media platforms like Twitter.
• To identify linguistic and behavioral differences between users who rarely or never share exaggerated content versus those who frequently do.
• To develop a predictive model that distinguishes users based on their tendency to share exaggerated news using linguistic and network features.
• To examine the temporal dynamics of exaggerated news sharing, particularly the evolution of language use in late-arriving tweets.

Proposed method

• The study uses a dataset of 462 annotated press releases and 668 news articles from UK universities, with labels indicating exaggeration based on changes in causal claims, advice explicitness, and sample identity.
• Linguistic Inquiry and Word Count (LIWC) analysis is applied to tweets to extract features related to emotion, cognition, and social processes, including categories like 'assent', 'feel', 'negative emotion', and 'bio'.
• User behavior is analyzed by categorizing users based on their frequency of sharing exaggerated content: never (users_NEX), rarely (users_EX1), twice (users_EX2), or three or more times (users_EX3).
• Features such as retweet/mention counts, slang, hyperbolic words, contractions, and tweet length are extracted from user timelines to compare posting behaviors.
• A supervised classification model is trained using these features, with stratified 10-fold cross-validation and class weighting to handle class imbalance.
• Random Forest and XGBoost classifiers achieve the highest performance, with an F1-score of 0.83 in distinguishing users who rarely or never share exaggerated content from those who do so frequently.

Experimental results

Research questions

• RQ1 How does the linguistic content of tweets about exaggerated health news evolve over time, particularly in late-arriving tweets?
• RQ2 What behavioral and linguistic differences distinguish users who frequently share exaggerated health news from those who do not?
• RQ3 Can user posting patterns and linguistic features be used to predict whether a user is likely to share exaggerated news content?
• RQ4 How do LIWC categories such as 'assent', 'anxiety', 'religion', and 'negative emotion' differ between tweets sharing exaggerated versus non-exaggerated content?

Key findings

• Late-arriving tweets sharing exaggerated news show a stronger presence of words in LIWC categories like 'assent', 'feel', 'religion', and 'anxiety', indicating increased opinion and emotional realization.
• These late tweets exhibit a weaker presence of words related to 'death', 'sexual', 'sad', and 'negative emotion', suggesting reduced focus on distressing or fatalistic language.
• Users who frequently share exaggerated content have significantly more followers, use more slang, fewer hyperbolic words, and fewer word contractions compared to those who rarely share such content.
• Linguistic analysis reveals that users posting exaggerated content more often use more words from LIWC categories 'bio', 'health', 'body', and 'negative emotion' in their tweets.
• The classifier achieves an F1-score of 0.83 in distinguishing users who never or rarely share exaggerated content from those who do so frequently, with Random Forest and XGBoost models performing best.
• The study demonstrates that over time, the 'wisdom of the crowd' may help identify exaggerated content through linguistic cues, particularly increased use of opinion and realization-related language.