Methods and apparatus for constructing correction tree models and using them to evaluate intervention effects.

Abstract

This specification provides a method and apparatus for constructing a correction tree model and using the model to evaluate the intervention effect. The method for constructing the tree model includes: obtaining a sample set, including indicator values ​​of user behavior for each user in two groups of users over m time periods, where the m time periods include m-1 preceding time periods and a target time period; one group of users is subjected to an intervention behavior during the target time period. A regression tree is constructed based on the splitting loss of nodes, where the splitting loss includes the confusion entropy of each of the two child nodes obtained from the split; the confusion entropy of any node is the sum of the cross-losses of each user falling into that node in the preceding time periods, where the cross-loss characterizes the difference between the user's historical indicator value in the preceding time periods and the mean indicator value of users from different groups falling into that node during the same time period. Thus, based on the regression tree, a correction tree model is obtained, which is used to determine the intervention effect of the intervention behavior applied to the user.

Images