Beyond Sentiment: Examining the Role of Moral Foundations in User Engagement with News on Twitter

TL;DR

This study investigates how moral foundations and affective language in Twitter news tweets relate to user engagement and discourse dynamics. It integrates sentiment analysis (EmoLex) and moral lexicons (MoralStrength) with Non-negative Matrix Factorization (NMF) to derive latent moral-emotional profiles, supported by label-propagation-based macro-area categorization. Regression analyses reveal that Surprise, Trust, and Harm robustly explain engagement and discussion length, with tweet-level content generally more predictive of engagement than linked articles, highlighting topic-dependent moral-emotional patterns. The findings inform how moral-emotional cues shape public discourse on social media and underscore engagement-priming risks in contemporary media ecosystems.

Abstract

This study uses sentiment analysis and the Moral Foundations Theory (MFT) to characterise news content in social media and examine its association with user engagement. We employ Natural Language Processing to quantify the moral and affective linguistic markers. At the same time, we automatically define thematic macro areas of news from major U.S. news outlets and their Twitter followers (Jan 2020 - Mar 2021). By applying Non-Negative Matrix Factorisation to the obtained linguistic features we extract clusters of similar moral and affective profiles, and we identify the emotional and moral characteristics that mostly explain user engagement via regression modelling. We observe that Surprise, Trust, and Harm are crucial elements explaining user engagement and discussion length and that Twitter content from news media outlets has more explanatory power than their linked articles. We contribute with actionable findings evidencing the potential impact of employing specific moral and affective nuances in public and journalistic discourse in today's communication landscape. In particular, our results emphasise the need to balance engagement strategies with potential priming risks in our evolving media landscape.

Figures (3)

• Figure 1: Time evolution of the daily ratio of news outlets' tweets expressing a given moral (top) or emotion (bottom) on a moving window of $7$ days.
• Figure 2: Standardized regression coefficients measuring the effect size of features extracted from the news outlets' tweets (red: only emotional features, blue: only moral features, yellow: the components of our NMF; each colour group constitutes a different experimental scheme) on different Twitter engagement metrics (replies, quotes, likes and retweets) and the sentiments found through the Twitter conversation. All the models are controlled for the number of followers of the news outlet and the tweet macro area(s) as confounders. Inside each feature group, the values are sorted according to the highest coefficient on reply count, and we only report those weights whose p-value is $\leq 0.05$.
• Figure 3: Standardized regression coefficients measuring the effect size of emotional, moral, or NMF features on different Twitter engagement metrics, for subsets of the news outlets' tweets corresponding to specific macro areas. In this experiment we follow the regression setup described above on each subset of the dataset corresponding to a specific macro area. The horizontal bars separate different regression models. All the models are controlled for the number of followers of the news outlet, and weights whose p-value is $\geq 0.05$ are omitted and shown with a slash.