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

Yuting Yang, Juan Cao+3 more|arXiv (Cornell University)|Feb 2, 2019

Topic Modeling13 references4 citations

TL;DR

This paper proposes a social network news quality assessment model (SNQAM) that predicts news quality by analyzing eight linguistic features derived from established writing guidelines. By weighting these features based on their correlation with popularity (used as a proxy for quality), the model achieves high interpretability, providing targeted suggestions for improving news writing style on social media platforms.

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.

Motivation & Objective

• To investigate the impact of writing style on news quality in social media contexts.
• To address the challenge of measuring news quality without human judgment by using popularity as a proxy.
• To develop a model that not only predicts quality but also provides actionable, interpretable suggestions for improvement.
• To isolate the influence of writing style from confounding factors like event virality and publisher reach.
• To create a framework for automated, guideline-based quality assessment of social media news content.

Proposed method

• Define eight linguistic feature types based on established news writing guidelines: Readability, Credibility, Interestingness, Sensation, etc.
• Extract 53 specific linguistic features from news posts, including punctuation use, pronoun frequency, and structural elements.
• Use Inter-User and Intra-User analysis to disentangle the effects of event and publisher from writing style on popularity.
• Calculate feature importance (I_i) via binary classification (VERY GOOD vs. TYPICAL) within each user (Intra-User) and correlation (C_i) across users (Inter-User).
• Compute feature weights as W_i = I_i * C_i and aggregate them into a quality score using the formula: News_Quality_Score = Σ(W_i * F_i).
• Generate targeted feedback by analyzing the contribution of each of the eight writing facets to the overall score.

Experimental results

Research questions

• RQ1 To what extent do specific writing style features correlate with news popularity on social networks?
• RQ2 How can the influence of external factors like event type and publisher reach be mitigated when assessing writing style impact?
• RQ3 Can a weighted combination of linguistic features predict news quality with high interpretability and practical utility?
• RQ4 Which writing guidelines (e.g., readability, credibility, sensation) have the strongest impact on news popularity?
• RQ5 Can the model provide actionable, guideline-based suggestions for improving news quality?

Key findings

• The SNQAM model achieves a Spearman rank correlation (SRC) of 0.606 between predicted quality scores and actual news popularity.
• The top 20 features correlated with quality are highly consistent across users, with 10 features appearing in the top 20 for all users.
• Features related to Readability, Credibility, Interestingness, and Sensation show the strongest and most consistent correlation with news quality.
• A notable shift in writing style—increased use of exclamation marks, faces, and engaging descriptions—coincided with a sustained increase in quality for Xinhua Net after period 50.
• The model successfully differentiates between high- and low-quality posts, assigning a score of 0.770 to Post1 and 0.207 to Post2, with clear diagnostic insights into the underperforming facets.