Predicting Social Media Engagement from Emotional and Temporal Features
By: Yunwoo Kim, Junhyuk Hwang
Potential Business Impact:
Predicts song likes and comments from feelings.
We present a machine learning approach for predicting social media engagement (comments and likes) from emotional and temporal features. The dataset contains 600 songs with annotations for valence, arousal, and related sentiment metrics. A multi target regression model based on HistGradientBoostingRegressor is trained on log transformed engagement ratios to address skewed targets. Performance is evaluated with both a custom order of magnitude accuracy and standard regression metrics, including the coefficient of determination (R^2). Results show that emotional and temporal metadata, together with existing view counts, predict future engagement effectively. The model attains R^2 = 0.98 for likes but only R^2 = 0.41 for comments. This gap indicates that likes are largely driven by readily captured affective and exposure signals, whereas comments depend on additional factors not represented in the current feature set.
Similar Papers
Early Multimodal Prediction of Cross-Lingual Meme Virality on Reddit: A Time-Window Analysis
Artificial Intelligence
Predicts if memes will go viral very fast.
Sentiment Analysis of Social Media Data for Predicting Consumer Behavior Trends Using Machine Learning
Human-Computer Interaction
Predicts what people will want to buy next.
Spatiotemporal EEG-Based Emotion Recognition Using SAM Ratings from Serious Games with Hybrid Deep Learning
Machine Learning (CS)
Reads your feelings from brain waves.