Know Your Users via Image Analytics Before Developing Posts: Data-Driven Optimization Framework to Enhance Social Media Engagement

• 成果类型: Article; Early Access
• 署名作者: Majumdar, Mayukh; Kumar, Subodha; Sriskandarajah, Chelliah
• 署名单位: University of San Diego; Pennsylvania Commonwealth System of Higher Education (PCSHE); Temple University; Texas A&M University System; Texas A&M University College Station; Mays Business School
• 刊物名称: MANAGEMENT SCIENCE
• ISSN/ISSBN: 0025-1909
• DOI: 10.1287/mnsc.2023.01838
• 发表日期: 2026
• 关键词: data-driven optimization; user engagement; Social media; discrete optimization; image analytics; Deep learning

摘要

Social media platforms are popular for advertisers to promote products to their audience via posts. Developing these posts with the appropriate number of image features is essential as these features enhance informativeness and visual complexity, impacting how users engage with posts. However, the existing literature offers a limited investigation into this topic, primarily examining low-level imagery information while overlooking high-level image features, cross-platform differences arising from different user expectations, and the tradeoffs firms must make in tailoring their strategies. To address this critical gap, we develop an optimization framework for analyzing and publishing social media image posts for different platforms within a firm's limited budget. This optimization framework is grounded in empirical modeling of the association between image features and social media user engagement, with primary and secondary features identified using deep learning algorithms. We focus on secondary features because they add visual complexity and informativeness, are controllable by designers, and are costly to extract, underscoring the need to assess their value for engagement. We find a nonlinear association between secondary features and engagement that varies across the two platforms. The optimization framework, which models the nonlinear relationship and is tested on realistic scenarios, considers and compares our approach against commonly used strategies that allocate budgets solely based on the user bases of platforms, those that ignore secondary features, and those that do not use duplicate features that can enhance informational richness and context. We present key implications for firms seeking to maximize user engagement on social media platforms.