Uplift modeling with value-driven evaluation metrics

Gubela, Robin M.; Lessmann, Stefan

doi:10.1016/j.dss.2021.113648

Cited by 14 publications

(5 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Which instances are to be prioritized when only a limited number of positive treatments are available? As highlighted by Gubela and Lessmann (2021), when ranking instances using the estimated ITE and ignoring value considerations, a suboptimal ranking would be obtained. As a solution, Gubela and Lessmann (2021) propose three measures for ranking instances that combine the ITE estimate and the estimated value of an instance using various aggregation functions, with the objective of optimizing the resulting total value.…”

Section: Analysis Of the Cost-sensitive Causal Classification Boundarymentioning

confidence: 99%

“…As highlighted by Gubela and Lessmann (2021), when ranking instances using the estimated ITE and ignoring value considerations, a suboptimal ranking would be obtained. As a solution, Gubela and Lessmann (2021) propose three measures for ranking instances that combine the ITE estimate and the estimated value of an instance using various aggregation functions, with the objective of optimizing the resulting total value. An experimental evaluation on marketing campaign data indicates that a substantial improvement over ranking with ITE can be obtained in terms of profit by using the proposed measures to rank and select customers to target.…”

Section: Analysis Of the Cost-sensitive Causal Classification Boundarymentioning

confidence: 99%

“…An experimental evaluation on marketing campaign data indicates that a substantial improvement over ranking with ITE can be obtained in terms of profit by using the proposed measures to rank and select customers to target. Whereas the ranking measures presented in Gubela and Lessmann (2021) are flexible and facilitate the use of a broad range of aggregation functions to combine the ITE and expected value of an instance, we aim to derive a measure that ranks instances so as to maximize the expected causal profit at any threshold, as determined by the number of positive treatments that can be applied.…”

Section: Analysis Of the Cost-sensitive Causal Classification Boundarymentioning

confidence: 99%

See 2 more Smart Citations

To do or not to do? Cost-sensitive causal classification with individual treatment effect estimates

Verbeke¹,

Olaya²,

Guerry³

et al. 2023

European Journal of Operational Research

View full text Add to dashboard Cite

Section: Analysis Of the Cost-sensitive Causal Classification Boundarymentioning

confidence: 99%

Section: Analysis Of the Cost-sensitive Causal Classification Boundarymentioning

confidence: 99%

Section: Analysis Of the Cost-sensitive Causal Classification Boundarymentioning

confidence: 99%

See 1 more Smart Citation

To do or not to do? Cost-sensitive causal classification with individual treatment effect estimates

Verbeke¹,

Olaya²,

Guerry³

et al. 2023

European Journal of Operational Research

View full text Add to dashboard Cite

“…(2) The second distinction -and one of the study's key findings -relates to customer ratings. In the few articles in which the customer rankings are, the rankings are usually based on arbitrary, authority, evaluation measures, or environmental weights (see [4][5][6]). In this study, six new innovative ranking formulas are presented.…”

Section: Contributionmentioning

confidence: 99%

Scheduling of transportation fleet based on the customer’s priority in a hub location problem

et al. 2023

View full text Add to dashboard Cite

This study proposes six novel strategies on the customer's priority while addressing the conventional hub location issue. Each strategy assigns a value to every customer based on distance and demand parameters, in which customers are prioritized based on this value. Then, the vehicle fleet is scheduled according to the customer's priority. A new mixed-integer linear programming model is presented and applied for each strategy in a new hub location-scheduling problem solved by three approaches. Then, by using the CAB dataset, extensive experiments are designed to evaluate each strategy. The strategies are evaluated with statistical and nonstatistical analyses and ranked accordingly. In each case, a comparison of the non-priority strategy with the best customer's prioritization strategy shows that the non-priority strategy has an adverse effect on the delivery time (i.e., 129.7%, 171.68%, and 161.33% than the best strategy in the case of near, medium, and far nodes, respectively). In addition to the above tests, other tests are conducted to evaluate the optimum number of vehicles for different conditions. The results show that as the distance between customers and hubs increases, reducing the number of vehicles while increasing their capacity is preferable. Also, each strategy requires using a certain number of vehicles.

show abstract

“…Uplift modeling refers to the set of techniques used to estimate the effect of an action on a user's outcome. This technology can be applied to various fields, such as economics [1,2], medicine [3,4] and sociology [5,6]. For example, an e-commerce company is preparing to send promotional coupons (i.e., action or treatment) to some users to attract them to purchase more products (i.e., outcome).…”

Section: Introductionmentioning

confidence: 99%

Graph Neural Network with Two Uplift Estimators for Label-Scarcity Individual Uplift Modeling

Zhu¹,

Wang²,

Zhang³

et al. 2023

Proceedings of the ACM Web Conference 2023

View full text Add to dashboard Cite

Uplift modeling aims to measure the incremental effect, which we call uplift, of a strategy or action on the users from randomized experiments or observational data. Most existing uplift methods only use individual data, which are usually not informative enough to capture the unobserved and complex hidden factors regarding the uplift. Furthermore, uplift modeling scenario usually has scarce labeled data, especially for the treatment group, which also poses a great challenge for model training. Considering that the neighbors' features and the social relationships are very informative to characterize a user's uplift, we propose a graph neural network-based framework with two uplift estimators, called GNUM, to learn from the social graph for uplift estimation. Specifically, we design the first estimator based on a class-transformed target. The estimator is general for all types of outcomes, and is able to comprehensively model the treatment and control group data together to approach the uplift. When the outcome is discrete, we further design the other uplift estimator based on our defined partial labels, which is able to utilize more labeled data from both the treatment and control groups, to further alleviate the label scarcity problem. Comprehensive experiments on a public dataset and two industrial datasets show a superior performance of our proposed framework over state-of-the-art methods under various evaluation metrics. The proposed algorithms have been deployed online to serve real-world uplift estimation scenarios. CCS CONCEPTS• Mathematics of computing → Graph algorithms.

show abstract

Uplift modeling with value-driven evaluation metrics

Cited by 14 publications

References 22 publications

To do or not to do? Cost-sensitive causal classification with individual treatment effect estimates

To do or not to do? Cost-sensitive causal classification with individual treatment effect estimates

Scheduling of transportation fleet based on the customer’s priority in a hub location problem

Graph Neural Network with Two Uplift Estimators for Label-Scarcity Individual Uplift Modeling

Contact Info

Product

Resources

About