Abstract:
In the continuously changing world of social media, Instagram has taken a prominent place by becoming one of the most popular social media platforms. Instagram has not only the biggest organic reach but also the highest organic engagement rate. Above all, understanding and predicting what makes posts go viral is an uneasy yet significant challenge. This study focuses on Instagram and presents a fresh approach to discovering the key factors contributing to post virality, specifically on image posts. In this study, a metric named ‘Virality Rate’ is defined to predict the likelihood of going viral. It is calculated by dividing the sum of number of likes and comments by the number of followers. There were studies on Instagram post popularity prediction based on various features and datasets. But with a focus on public image-based posts from influencers worldwide, this research delves into sentiment analysis, image processing for technical features and content, hashtag assessment, user history and user features to forecast the potential virality of a post. This research trained and compared several regression models to predict the Viral Rate and employed Faster R-CNN and OpenCV to detect objects and help extract essential technical details. Through rigorous model training and evaluation, our results highlight the Random Forest Regression model as the most effective predictor. It boasts an impressive Mean Absolute Percentage Error (MAPE) of 0.15, which implies an accuracy of 85% and a notable R-squared (R2) value of 0.924 which is significant compared to previous studies. It was found that the User History Features, sentiment score, technical features and posting time have a high impact on Virality Rate. In conclusion, this research aims to advance social media analytics by offering actionable insights for content creators, influencers, marketers and regular users.
