Meta-learner-based training data generation method and heterogeneous response difference estimation method for causal effects

Abstract

The invention discloses a method for generating training data based on a meta-learner and a method for estimating the difference between causal effects and heterogeneous responses. Firstly, two basic learners are learned by using the original training data obtained from a random field test, and then the two The base learner conducts cross-test on the original training data to generate training data; then uses the generated training data to learn a task sub-learner for the treatment group and the control group respectively, and then two task sub-learners constitute the final task learner ; Through the final task learner can realize the estimation of user causal effect heterogeneous response difference. Based on the meta-learner, the present invention proposes a framework for estimating the heterogeneous response differences of causal effects, which can be used in conjunction with any basic learner model to estimate any type of causal effect response differences; for example, it can be used to guide pricing Discounting means, design of effective advertising strategies, and design of product dimensions and packaging solutions, etc.

Description

technical field [0001] The invention belongs to the combined application of the field of machine learning (Machine Learning) technology and the field of randomized field experiments (Randomized Field Experiments), and relates to the difference estimation of causal effect heterogeneous responses (Heterogeneous Response) based on meta-learners. [0002] The difference in heterogeneous responses refers to the difference in a specific response between a user who is affected by a certain behavior and who is not affected by a certain behavior. Sometimes the impact is defined as treatment (Treatment), and the unaffected is defined as control (Control), so the present invention is to estimate the difference of a certain behavior of a certain user between the treatment group (TreatmentGroup) and the control group (ControlGroup) , is said to be heterogeneous because a specific user can only be in one group, that is, either in the treatment group or in the control group, and it is imposs...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to aim at the causal inference problem existing in training data samples in traditional machine learning methods, to provide a general training data generation method, so that the generated training data can meet the learning requirements of the heterogeneous response difference estimation method, so that it can be used A Machine Learning Approach Directly Models the Relationship Between Heterogeneous Reaction Responses and User Characteristics

Images