Feature classification methods on measuring user engagement in social media campaigns

Krishnamurthy, Soujanya, Misopoulos, Fotios ORCID: 0000-0002-0046-8055, Leung, Eric KH ORCID: 0000-0003-2058-0287 and Antonopoulou, Katerina (2026) Feature classification methods on measuring user engagement in social media campaigns IEEE Transactions on Engineering Management, 73 (99). pp. 1-16. ISSN 0018-9391, 1558-0040

Text Feature_classification_methods_on_measuring_user_engagement_in_social_media_campaigns.pdf - Author Accepted Manuscript Available under License Creative Commons Attribution. Download (598kB) | Preview

Abstract

One-third of food produced for human consumption is wasted, with households in high-income countries responsible for nearly half of this total. This study examines user engagement with three major U.K. food waste reduction campaigns on X (formerly Twitter), analyzing more than 50 000 tweets collected over seven years. Using sentiment analysis, topic modeling (TM), and feature-importance classification (random forest, support vector machines, Naïve Bayes, and gradient boosting), the study identifies the content attributes that are most strongly associated with higher engagement. Practical advice (e.g., food storage tips), community mobilization, and food sharing consistently generate the highest levels of interaction, while the strategic use of hashtags, mentions, and influencer endorsements amplifies engagement. This article makes three theoretical contributions. First, it advances agenda-setting theory by demonstrating how long-term digital campaigns curate content to elevate food waste as a salient public issue, with distinct strategies (aligning messages with organizational vision, offering actionable tips, and promoting food sharing) driving engagement. Second, it extends framing theory by showing how platform-native tools associate with interaction, with tweets using either hashtags or mentions (but not both) achieving greater resonance, suggesting that simplified framing may be more effective in crowded digital spaces. Third, it offers a methodological contribution by integrating feature importance analysis with sentiment and TM, moving beyond descriptive metrics to predictive insights into which tweet attributes are most strongly associated with higher engagement. For practitioners, the study provides a cost-effective framework for dynamically adapting campaign content, equipping engineering managers and policy makers to design more impactful, scalable, and sustainable digital interventions that promote public engagement and behavioural change.

Item Type:	Article
Uncontrolled Keywords:	Social networking (online), Food waste, Analytical models, Predictive models, Sentiment analysis, Organizations, Sustainable development, Media, Monitoring, Web sites, Classification methods, feature importance, food waste, sentiment analysis (SA), social media campaigns, topic modeling (TM)
Divisions:	Faculty of Humanities & Social Sciences Faculty of Humanities & Social Sciences > School of Management Faculty of Humanities & Social Sciences > Faculty of Humanities & Social Sci (All T&R Staff) Faculty of Humanities & Social Sciences > School of Management > School of Management (T&R Staff) Faculty of Humanities & Social Sciences > School of Management > Operations and Supply Chain Management
Depositing User:	Symplectic Admin
Date Deposited:	16 Feb 2026 08:32
Last Modified:	09 Apr 2026 15:27
DOI:	10.1109/tem.2026.3664701
Related Websites:	Author Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3197029
Disclaimer:	The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.