Method and device for realizing regional AB experiment management based on algorithm

Abstract

The invention discloses a method and a device for realizing regional AB experiment management based on an algorithm. The method comprises the following steps: creating an experiment, receiving initialexperiment configuration, and generating an experiment ID; when each language access SDK starts an application, pulling the configuration of all experiment IDs used by the application configuration through an API and storing the configuration in a file system, and pulling new experiment configuration from the API periodically; performing experimental calculation according to the user access information, returning to the access application to use the scheme A or the scheme B currently, and recording and storing the scheme used by the user ID; sending the recorded information to a data analysisplatform through a log collection system, and displaying an analysis result and experimental index data through a visualization platform. After the functions are online, developers do not need to intervene in operation, one application supports parallel operation of a plurality of experiments, the plurality of applications can share the same experiment, and services are not affected by faults ofan API system.

Description

technical field [0001] The invention relates to the technical field of online AB test experiments, in particular to a method and device for managing regional AB experiments based on algorithms. Background technique [0002] At present, the general AB experiment method is to use the function of the traffic statistics platform to make two pages with different URLs, and to make statistical summary of the effect according to different URLs. The problem with this solution is that it is impossible to control which page is displayed to the user. The flexibility is not strong, and the AB test can only be done based on the entire page; after the AB test is completed, the function needs to be re-adjusted and launched again or the page jumps. When there are too many pages that need to be jumped on an application, the management will be very troublesome. Another solution is to do different processing at the load balancing layer. A and B use different application versions to release. The...

AI Technical Summary

Problems solved by technology

[0002] At present, the general AB experiment method is to use the function of the traffic statistics platform to make two pages with different URLs, and to make statistical summary of the effect according to different URLs. The problem with this solution is that it is impossible to control which page is displayed to the user. The flexibility is not strong, and the AB test can only be done based on the entire page; after the AB test is completed, the function needs to be re-adjusted and launched again or the page jumps. When there are too many pages that need to be jumped on an application, the management will be very troublesome

Another solution is to do different processing at the load balancing layer. A and B use different application versions to release. The problem with this solution is that two versions need to be run, and O&M needs to perform online change operations to adjust and configure different traffic configurations. In contrast, the final release also requires online operation and maintenance changes, making the cost of AB experiments high

Images