Source-linked AI summary
Breaking the News: First Impressions Matter on Online News
Julio Reis, Fabrıcio Benevenuto, Pedro O. S. Vaz de Melo, Raquel Prates, Haewoon Kwak, Jisun An
TL;DR
Online news organizations need effective ways to attract attention in a competitive environment, but the strategies that make news attractive remain insufficiently understood. The paper analyzes 69,907 news items from four major media corporations using headline sentiment, popularity estimates, and comment sentiment. It finds that extreme headline sentiment is associated with greater popularity, while comments are generally negative regardless of headline sentiment.
Problem
Online news sites compete for limited user attention, but the strategies companies use to make news more attractive remain insufficiently understood.
Method
The study analyzes 69,907 news items from four major media corporations using sentiment classification, Bit.ly click counts as a popularity estimate, and comment-sentiment analysis.
Results
Extremely negative and positive headlines tend to attract more popularity, while comments tend to be negative independently of headline sentiment.
Takeaways & Limitations
Headline sentiment is relevant to online-news popularity, while comment dynamics do not simply mirror the sentiment of the associated headline.
Takeaways & Limitations
Popularity inferred from Bit.ly clicks may reflect sharing through social networks or emails where users changed how the news was presented, including its sentiment.
Abstract
from arXiv · showhide
A growing number of people are changing the way they consume news, replacing the traditional physical newspapers and magazines by their virtual online versions or/and weblogs. The interactivity and immediacy present in online news are changing the way news are being produced and exposed by media corporations. News websites have to create effective strategies to catch people's attention and attract their clicks. In this paper we investigate possible strategies used by online news corporations in the design of their news headlines. We analyze the content of 69,907 headlines produced by four major global media corporations during a minimum of eight consecutive months in 2014. In order to discover strategies that could be used to attract clicks, we extracted features from the text of the news headlines related to the sentiment polarity of the headline. We discovered that the sentiment of the headline is strongly related to the popularity of the news and also with the dynamics of the posted comments on that particular news.
Introduction
The paper examines how online news organizations use headline sentiment to attract attention and clicks in a competitive environment. It analyzes 69,907 news items from four major media corporations and relates headline sentiment to popularity and comments.
- Online news sites compete for users’ limited attention, making effective strategies for attracting clicks important.
- Headlines are designed to quickly draw attention and may influence how readers perceive and remember news content.
- The study asks whether popular newspapers favor negative, positive, or neutral news and whether extreme sentiment attracts more clicks.
- It also examines whether the sentiment of users’ comments reflects the polarity of associated headlines.
- 69,907 news items from The New York Times, BBC, Reuters, and Dailymail were analyzed over at least eight consecutive months in 2014.
- The analysis reports predominantly negative headlines, greater popularity for extremely negative and positive headlines, and generally negative comments across headline polarities.
Literature Review
The literature review situates the study across research on online news, reading habits, coverage, propagation, and sentiment analysis. Prior work links headlines to interpretation, mood to news selection, and sentiment to dissemination and discourse quality.
- News headlines: Prior experiments found that misleading headlines affect readers’ memory, inferential reasoning, and behavioral intentions.
- Reading Habits: Mood-management research reports that prior emotional environments influence which news people choose to read.
- News Coverage and Popularity: Large-scale studies have examined how news features and comments relate to global coverage and popularity.
- News Propagation: Twitter-based research identified evidence that negative news spreads faster and examined temporal dynamics of information propagation.
- Sentiment analysis of news: Sentiment-analysis applications include stock trading, polarity-based news selection, and using comment sentiment as an indicator of discourse quality.
Methodology
The study characterizes online news by collecting articles from four major sources and inferring sentiment, popularity, and topical category. It uses sentiment-method comparison, Bit.ly clicks, URL or archive categories, and validation against Twitter sharing text, while acknowledging sampling and popularity-measurement biases.
- Data collection: The dataset combines BBC, Dailymail, The New York Times, and Reuters news collected across 2013–2014, with Reuters comments available for analysis.BBC, Dailymail, and The New York Times were monitored from March to November 2014; Reuters Top News was collected retroactively from January 2013 to September 2014.
- Sentiment analysis: Sentiment analysis was selected by comparing eight methods against manually labeled headlines and comments.Three volunteers labeled 100 headlines and 100 comments, and SentiStrength performed best, with 74% accuracy for headlines and 79% for comments.
- Inferring popularity: Popularity was approximated by Bit.ly clicks collected 20 days after news collection, rather than by direct URL-click totals.The measure relies on Bit.ly’s persistent association between shortened and original URLs and on the concentration of clicks soon after release.
- Categorizing news: Topical categories were inferred from URL metadata for three sources and from Reuters archive categories, with 9% of Reuters items discarded when categories could not be identified.World news was the most representative category across all four sources.
- Limitations: The study recognizes that Bit.ly clicks may reflect sharing contexts rather than headline-driven interest and may favor news that spreads quickly through social networks.The authors released the dataset so these limitations can be assessed with alternative popularity measures.
- Validation: A Twitter validation sample contained 5,182 tweets for 100 crawled URLs, and polarity agreement with headlines exceeded 72% for every class.The sample tested whether shared Twitter text preserved the headlines’ sentiment polarity; 49% of tweets used the exact headline text.
Headlines Polarity
Across the four news sources, negative headlines are most common, with sentiment varying by topic and remaining broadly stable over time. Category-level patterns show short-lived spikes, while positive headlines never dominate.
- Overall polarity: Negative headlines are the majority across all four sources, followed by neutral and positive headlines.For Dailymail, 65% of headlines are negative.
- Topic differences: Health and World headlines are predominantly negative across all four news sources.World is the most popular category in the analyzed sources.
- Topic differences: Sentiment distributions differ across categories, but negative headlines generally exceed positive headlines.Science and Technology tends to be more neutral except for Dailymail; only a few unpopular categories, such as Dining and Travel, are more positive.
- Temporal patterns: Negative headlines remain consistently more numerous over the analyzed months, with only slight fluctuations and no aggregate sentiment spikes.Day-of-week analysis also showed little change in sentiment distribution.
- Temporal patterns: Category-level sentiment spikes occur briefly and may reflect events or hot topics, but their distributions differ substantially across sources.Positive sentiment never reaches the highest percentage in any category or period; the highest percentage is negative or neutral.
Sentiment Score versus Popularity
The paper examines whether headline sentiment relates to news popularity by comparing normalized mean popularity across sentiment scores. Extreme sentiment is associated with the greatest popularity, especially for BBC and Dailymail.
- Popularity relationship: Popularity is highest for headlines with extreme sentiment scores across all four news sources.Popularity values were normalized for comparison, and extreme sentiment strengths were grouped at the tails.
- Popularity relationship: For BBC and Dailymail, both extreme negative and extreme positive headlines are associated with the most popular articles.The authors suggest that strongly negative or strongly positive news tend to attract Internet users.
- Interpretation: The findings suggest that extreme headline sentiment is associated with greater success, whereas neutral headlines are less likely to be successful.The paper relates this pattern to observations about successful meme propagation in social networks.
Characterizing the Sentiment of Comments
Comments were predominantly negative and were associated with negative or extreme headline sentiment, while negative commenting remained widespread across headline sentiments and news categories.
- The comment analysis used Reuters articles because Reuters was the only source with comments available.
- Negative headlines were more likely to receive comments, and comment volume increased as headline sentiment moved farther from neutral.Neutral headlines triggered fewer user interactions than more extreme headlines.
- 75% of comments had negative sentiment, compared with 11% neutral and 14% positive comments.
- More than 70% of comments were negative even when headlines expressed extremely positive sentiment.Negative comments were significantly more common than other comment sentiments across headline sentiment scores.
- Negative comments exceeded neutral and positive comments in every category, with World and Health attracting the most and Sports and Science and Technology the least.Even Science and Technology news received approximately 60% negative comments.
Discussion
The discussion links headline sentiment to article popularity and examines how headlines may differ from content and shape readers’ perceptions, while noting pervasive negativity in comments.
- Discussion: A non-negligible amount of news had different sentiment in its headline and content, although the comparison is limited because content is longer and can contain mixed sentiment.
- Discussion: The predominance of negative headlines is especially relevant because World news is both the most common category and among the most negative.
- Implications for System Design: Neutral news were usually less popular than extremely positive or negative news, making headline sentiment a potentially useful feature for popularity-prediction systems.
- Implications for System Design: Sentiment scores inferred from headlines may support popularity prediction, alongside other headline patterns such as “Top N” enumerations.
- Why are Comments So Negative?: Negative comments were ubiquitous across headline sentiments and news categories, a pattern the authors relate to possible negative-comment snowballing.
Conclusion
The paper characterizes online news using headlines as a central artifact and a large cross-source dataset, contributing findings, methods, and a reusable dataset for future research.
- 69,907 news items from four major global media corporations were collected over at least eight consecutive months in 2014.
- The methodology used sentiment analysis to study headline properties and a URL-shortening service to infer news popularity.
- The findings contribute to research on online-news characterization and prediction, online-news-system design, and broader societal understanding.
- The authors made the constructed dataset available to facilitate further research.