現今企業為了吸引顧客到店消費，時常推出許多行銷活動，因此在以數位媒體為主的針對性行銷中，行銷前如何挑選正確行銷對象，可說是行銷者最重要的工作。
過去企業主要使用響應模型（Response Model）預測顧客對特定行銷活動的回應機率，並針對回應機率高的顧客做行銷，但根據研究發現，使用響應模型可能選擇到不論有無做行銷皆會做出回應的顧客，造成企業花費額外成本。針對此問題，增益模型(Uplift model)可用於預測顧客接受行銷前後的回應機率增幅，透過模型估計行銷活動對於每一位顧客的「邊際效用」，據以選擇行銷對象，提高行銷績效。
由於過去文獻大多是將增益模型與使用「隨機方式」挑選顧客做比較，而現今企業在選擇行銷對象時，大多是使用響應模型的預測結果進行選擇。因此，本篇研究透過分析國外電商公司與台灣A公司的資料，利用邏輯斯迴歸建立響應模型，以及透過雙模型方法(Two-model approach)、將實驗變數轉換為虛擬變數方法(Treatment dummy approach)、轉換標籤法(Transformation approach)三種不同方法建立增益模型，比較增益模型挑選行銷對象的表現是否優於響應模型。結果發現模型表現在不同資料集中，並不一致，並無特定模型在所有資料集表現都最佳；因此面對不同資料集與不同行銷方案，會有不同挑選行銷對象的策略，但從研究中發現在大部分資料集，增益模型表現較優於響應模型，這一點與相關文獻的結論是吻合的。
此外，過去增益模型的研究多著重發展更好的建立模型方法，較缺乏實務應用，因此本研究以假設模擬方式將「行銷成本」、「毛利率」與「顧客因行銷可能帶來的價值」，結合用模型所預測的結果，從行銷實務的角度，模擬當企業以「整體投資報酬增額最大」策略挑選行銷對象，是否會有不同決策。透過研究發現當加入其他考慮因素與改變策略目標時，其挑選行銷對象會有所改變，預期的行銷成本效益也會有所改善；相對於以往以模型演算方法為主的技術性研究，本研究聚焦在增益模型的商業應用實務，為如何使用增益模型改善行銷績效提供一個具體的參考做法。

In today's business landscape, to attract customers to physical stores, companies often launch various marketing campaigns. Therefore, in targeted marketing that primarily relies on digital media, selecting the right marketing targets before conducting the marketing campaign can be considered the most crucial task for marketers.
In the past, businesses mainly used response models to predict the probability of customers’ response to specific marketing instruments, and they used to select the targets with high expected response probabilities. However, research has found that using response models may select customers who would respond regardless of whether treatment is applied, resulting in wasted marketing costs. In addressing this issue, Uplift modeling can be used to predict the probability of a customer's response before and after marketing intervention. By estimating the "marginal utility" of marketing activities for each individual customer, the model helps in selecting the right marketing targets and improving marketing performance.
While most of the previous literature comparing the uplift model with randomly selecting customers, modern businesses now primarily use response models for target selection. To cope with the actual business practice, this study analyzes data from a foreign e-commerce company and Company A in Taiwan. We utilize logistic regression to establish a response model and employs three different methods, namely the Two-model approach, Treatment dummy approach, and Transformation approach, to build uplift models. The purpose is to examine which targeting methods would maximize the number of responses. Our study reveals that the model performance varies across different datasets, and there is no specific model that performs best across all datasets. Therefore, different strategies are needed for selecting marketing targets depending on the dataset and marketing plan. However, the study revealed that uplift models performed better than response models in most datasets, as was suggested in the relevant literature.
Furthermore, previous study has primarily focused on developing better model-building methods, often lacking practical application. Therefore, this study combines hypothetical simulations with factors such as "marketing costs," "gross profit margin," and "marketing instrument’s marginal effect on customer value", along with the predicted results of the models. The aim is to align with the decision-making process of businesses in selecting marketing targets in practical settings. Additionally, the study examines whether there are different decisions when companies select marketing targets based on the strategy of maximizing overall return on investment. The results of our simulation indicate that when relevant factors and strategy objectives are considered, the selection of marketing target schemes will vary. Unlike previous technically oriented study centered on model algorithm advancements, this study focuses on the practical business application of uplift modeling, I provide examples with which, companies can incorporate the predicted results of the models into their own marketing considerations and goals, choose the appropriate model for selecting marketing targets, and make informed decisions.

論文審定書 i
誌謝 ii
中文摘要 iii
Abstract iv
目錄 vi
圖次 viii
表次 x
第一章 緒論 1
第一節 研究背景 1
第二節 研究動機 2
第三節 研究目的 2
第四節 研究流程 3
第二章 文獻探討 5
第一節 響應模型 5
第二節 增益模型 5
第三節 建立增益模型方法介紹 7
第四節 模型評估方式 10
第三章 研究方法 14
第一節 研究架構 14
第二節 資料來源與介紹 14
第三節 資料處理 17
第四章 研究結果 19
第一節 探索性資料分析 19
第二節 利用邏輯斯迴歸模型觀看自變數對應變數之影響 40
第三節 增益模型與響應模型衡量結果 45
第五章 增益模型延伸應用-用假設模擬選擇行銷對象 53
第一節 考慮其他因素下-用假設模擬挑選整體響應人數增額最多作法 53
第二節 考慮其他因素下-用假設模擬挑選整體投資報酬增額最高作法 56
第六章 研究結論與建議 60
第一節 研究結論 60
第二節 研究貢獻 61
第三節 研究限制與未來研究建議 61
參考文獻 62

一、英文文獻
Ascarza, E. (2018). Retention futility: Targeting high-risk customers might be ineffective. Journal of Marketing Research, 55(1), 80-98.
Belbahri, M., Murua, A., Gandouet, O., & Nia, V. P. (2019). Uplift Regression: The R Package tools4uplift. arXiv preprint arXiv:1901.10867.
Cunningham, S. (2021). Causal inference: The mixtape. Yale university press.
Devriendt, F., Moldovan, D., & Verbeke, W. (2018). A literature survey and experimental evaluation of the state-of-the-art in uplift modeling: A stepping stone toward the development of prescriptive analytics. Big data, 6(1), 13-41.
Gubela, R., Bequé, A., Lessmann, S., & Gebert, F. (2019). Conversion uplift in e-commerce: A systematic benchmark of modeling strategies. International Journal of Information Technology & Decision Making, 18(03), 747-791.
Kane, K., Lo, V. S., & Zheng, J. (2014). Mining for the truly responsive customers and prospects using true-lift modeling: Comparison of new and existing methods. Journal of Marketing Analytics, 2, 218-238.
Lai, L. Y. T. (2006). Influential marketing: a new direct marketing strategy addressing the existence of voluntary buyers.
Lorenz, M. O. (1905). Methods of measuring the concentration of wealth. Publications of the American statistical association, 9(70), 209-219.
Lo, V. S. (2002). The true lift model: a novel data mining approach to response modeling in database marketing. ACM SIGKDD Explorations Newsletter, 4(2), 78-86.
Radcliffe, N., & Surry, P. (1999). Differential response analysis: Modeling true responses by isolating the effect of a single action. Credit Scoring and Credit Control IV.
Radcliffe, N. (2007). Using control groups to target on predicted lift: Building and assessing uplift model. Direct Marketing Analytics Journal, 14-21.
Rößler, J., & Schoder, D. (2022). Bridging the Gap: A Systematic Benchmarking of Uplift Modeling and Heterogeneous Treatment Effects Methods. Journal of Interactive Marketing, 57(4), 629-650.
Rzepakowski, P., & Jaroszewicz, S. (2012). Uplift modeling in direct marketing. Journal of Telecommunications and Information Technology, 43-50.

二、中文文獻
廖承哲. (2019). 增量模型於不均衡實驗組之表現.

三、資料集來源
Hillstrom, K. (2008, March 21). Kevin Hillstrom: MineThatData. Blogger. https://blog.minethatdata.com/2008/03/minethatdata-e-mail-analytics-and-data.html