How to Write High-quality News on Social Network? Predicting News Quality by Mining Writing Style

Yuting Yang, Juan Cao, Mingyan Lu, Jintao Li, Chia-Wen LinarXiv 2019 · OpenReview

Abstract

As news consumption shifts to social media, understanding what makes content "viral" is crucial for journalists and platforms alike. This research investigates the linguistic "DNA" of high-quality news by analyzing writing style across eight quantifiable dimensions. Using a dataset of 300,000 Weibo posts, the authors developed the Social Network News Quality Assessment Model (SNQAM), which predicts news quality with 99.6% accuracy. The findings suggest that emotional resonance ("Sensation") and rhetorical flair ("Interestingness") are key drivers of engagement, independent of a publisher's fame or the inherent newsworthiness of an event.

The Challenge of Social News Quality

A common frustration for digital journalists is the engagement gap: two accounts may report the exact same event, but one post goes viral while the other is largely ignored. Traditionally, popularity metrics like likes and shares are used as proxies for quality, but these are heavily biased by the publisher's existing follower count and the nature of the event itself.

Comparison of two posts on the same topic with vastly different engagement levels.

This study seeks to isolate writing style—the specific linguistic choices made by the author—to determine its impact on success. By comparing "Very Good" accounts (high average engagement) against "Typical" ones, the researchers identified patterns that transcend simple newsworthiness. As shown in the comparison above, stylistic choices like the use of color, emojis, and urgent language can lead to a massive disparity in reposts and comments for the same underlying news topic.

Methodology: The Eight Facets of Style

The researchers proposed the SNQAM framework, which evaluates news based on 53 linguistic features grouped into eight distinct facets:

  1. Readability: Sentence complexity and the use of professional jargon.
  2. Logic: Use of coherence markers, conjunctions, and forward references.
  3. Credibility: Inclusion of numbers, official speech, and specific entities (time, place, object).
  4. Formality: Distribution of parts of speech, such as the ratio of nouns to pronouns.
  5. Interactivity: Engagement tactics like questions and the use of first/second-person pronouns.
  6. Interestingness: Use of rhetoric, idioms, and descriptive adjectives.
  7. Sensation: Emotional intensity measured via sentiment scores and degree adverbs.
  8. Integrity: Structural elements including hashtags, links, and multimedia.

Key Insights: The Power of Sensation

The most striking finding was the consistent gap in "Sensation" and "Interestingness" between high-performing and average accounts. High-quality news isn't just factual; it is emotionally resonant.

Comparison of Sensation scores between 'Very Good' and 'Typical' news accounts over time.

As illustrated in the chart above, "Very Good" accounts consistently maintain significantly higher Sensation scores than "Typical" accounts across different periods. The researchers also noted a phenomenon called "Quality Drift" in the history of accounts like Xinhua Net. When the account transitioned to using more emojis, exclamation marks, and evocative language, their average engagement metrics saw a significant upward shift, proving that viral writing follows specific, optimizable linguistic patterns.

Results and Practical Applications

The SNQAM model achieved a 0.606 Spearman Rank Correlation (SRC) with real-world popularity metrics, proving a strong link between calculated style scores and actual performance. Beyond simple classification, the model provides actionable feedback. Unlike "black box" deep learning systems, SNQAM can generate a diagnostic "radar chart," telling a writer exactly where their post falls short—whether it needs more logical flow, better credibility markers, or increased interactivity to spark discussion.

Limitations and Future Directions

While the study provides a robust framework, it is currently limited to the Chinese language and the specific character constraints of the Weibo platform. Linguistic features like specific Chinese particles may not translate directly to English-centric platforms like X or Threads. Future work aims to adapt these features for broader contexts and explore the use of AI to automatically rewrite dry factual reports into high-engagement social media posts following these established quality guidelines.

Summary updated

Abstract

Rapid development of Internet technologies promotes traditional newspapers to report news on social networks. However, people on social networks may have different needs which naturally arises the question: whether can we analyze the influence of writing style on news quality automatically and assist writers in improving news quality? It's challenging due to writing style and 'quality' are hard to measure. First, we use 'popularity' as the measure of 'quality'. It is natural on social networks but brings new problems: popularity are also influenced by event and publisher. So we design two methods to alleviate their influence. Then, we proposed eight types of linguistic features (53 features in all) according eight writing guidelines and analyze their relationship with news quality. The experimental results show these linguistic features influence greatly on news quality. Based on it, we design a news quality assessment model on social network (SNQAM). SNQAM performs excellently on predicting quality, presenting interpretable quality score and giving accessible suggestions on how to improve it according to writing guidelines we referred to.
BibTeX
@misc{yang2019how,
  author = {Yuting Yang and Juan Cao and Mingyan Lu and Jintao Li and Chia-Wen Lin},
  title = {{How to Write High-quality News on Social Network? Predicting News Quality by Mining Writing Style}},
  year = {2019},
  eprint = {1902.00750},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1902.00750},
}
Figure