2026 年 10 月 8 日 CRETA Seminar - Selecting Optimal Influencers for Marketing Campaigns: Bayesian Additive Regression Tree Approach with Structural Priors

訊息標題：

2026 年 10 月 8 日 CRETA Seminar - Selecting Optimal Influencers for Marketing Campaigns: Bayesian Additive Regression Tree Approach with Structural Priors

簡介摘要：

國立臺灣大學計量理論與應用研究中心 (CRETA) ，國立臺灣大學經濟學系及臺灣經濟計量學會 (TES) 將於 2026 年 10 月 8 日舉辦 CRETA Seminar。相關資訊如下：

【CRETA Seminar】

時間：2026 年 10 月 8 日 (週四) 下午 1:30~3:00

地點：國立臺灣大學管理學院二號館 3 樓 304 教室

講者：Prof. Mandy Hu (https://www.bschool.cuhk.edu.hk/staff/hu-mandy-mantian/)

演講主題：Selecting Optimal Influencers for Marketing Campaigns: Bayesian Additive Regression Tree Approach with Structural Priors

講題摘要：Managing influencer marketing is a critical yet challenging resource-allocation problem because firms typically observe only the collective impact of simultaneous influencer promotions. Optimizing influencer selection requires predicting outcomes from high-dimensional, nonlinear response models with complex interactions. However, firms face a “large p, small n” problem: the pool of potential influencers (p) is vast, but historical campaign data (n) are severely limited. To resolve this, we propose a novel Bayesian Additive Regression Trees with Structural Priors (BARTwSP) approach. BARTwSP leverages tree-based machine learning to capture complex nonlinearities, utilizing a Bayesian framework with depth priors to prevent overfitting and structural priors to guide variable-splitting in high-dimensional spaces. We demonstrate that Large Lan-guage Models (LLMs) can generate this structural prior knowledge. In Monte Carlo simulations, BARTwSP reduces Mean Squared Error (MSE) by an average of 42.09% over benchmarks, while our ensemble approach yields an additional 8.31% reduction. Applied to data from a leading video game developer, our model achieves a 37.67% average MSE improvement for daily active users (DAU). Finally, using our endogeneity-corrected estimated response function to optimize selection under budget constraints, we find that transitioning to an intensive concentration strategy prunes 71% of redundant influencers, boosting DAU by 28% without increasing budgets.

備註事項：