Evidential Reasoning Approach for Predicting Popularity of Instagram Posts

Machine learning predictive models are increasingly used to analyse social media data. However, they often fall short in transparency and interpretability. This paper shows how Maximum likelihood evidential reasoning (MAKER), a method rooted in evidential reasoning, can address these issues by predicting post popularity based on the number of likes. To illustrate this approach, Instagram posts from Harvard University and the University of Oxford are analysed. MAKER’s performance is compared with decision tree (DT), support vector machine (SVM), and k-nearest neighbours (KNN) algorithms. A total of 289 posts from @harvard and 507 from @oxford_uni, posted during 2022, are collected, using the number of likes as the measure of popularity. Two models are developed for each university: one focusing on textual features, including emojis, sentiment, hashtags, mentions, and season, and another on visual elements like image type, time of day, and dominant colour. MAKER outperforms the other algorithms in several key metrics. For instance, MAKER achieves the highest precision in the textual model for both universities, reaching 0.8579 for Harvard and 0.8608 for Oxford, followed by KNN, which achieves 0.7245 for Harvard and 0.7287 for Oxford. Moreover, MAKER’s interpretability means that it provides actionable insights. For Harvard, popular posts most often feature vibrant colours and scenic landscapes. For Oxford, the use of emojis is associated with higher popularity in the textual model. The implication is that MAKER can help users make informed decisions based on its insights and improve content strategies by revealing which features most influence engagement.