Data and Code for "How Does Social Media Content Relate to User Engagement? A Benchmark-Relative Content Efficiency Perspective"

This repository provides the dataset and accompanying code for a study of benchmark-relative content efficiency in social media. The study examines likes, retweets, comments, and bookmarks separately. Benchmark-relative content efficiency denotes a post’s engagement output relative to a model-predicted conditional benchmark incorporating observed exposure, account differences, and temporal conditions. Benchmark residuals define high- and low-efficiency groups, which are predicted using content-only, context-only, and combined feature sets. The descriptive sample comprises 13,779 image-and-text posts published in 2024 by 69 brand accounts on X (formerly Twitter). The analytical sample includes 13,718 posts with positive recorded views. Cumulative engagement and view metrics were collected between 10 and 20 September 2025, and publication timestamps are recorded. The dataset contains post-level engagement counts, views, extracted textual and visual features, and account and temporal characteristics. This version updates the terminology and documentation to reflect the benchmark-relative content efficiency framework and includes code for data preparation, feature extraction, benchmark estimation, prediction, sensitivity analyses, and SHAP interpretation. Account-held-out and temporal validation assess predictive transfer. Dependency information and execution instructions are included to support examination and replication of the analyses.