Index-based large-scale automatic AB experiment shunting method

Abstract

An index-based large-scale automatic AB experiment shunting method comprises the following steps: setting the iteration number of AB experiments according to the number of item, calculating each iteration time length according to the iteration number, setting a key index, starting the AB experiments, calculating the index height of each experiment version after each iteration time is ended, reserving the highest AB experiment version, selecting a plurality of items from the items which are not executed for supplementation, carrying out AB experiment iteration again until the items are taken up, and finally obtaining the optimal item to serve as the result of the whole AB experiment. The problem that in the traditional AB experiment process, the AB experiment flow proportion cannot be automatically adjusted is solved, manual staring of all iterations is replaced by automatic calculation of the initial iteration number and the iteration duration, automatic execution of the AB experiment is carried out through automatic selection and rejection item of each iteration based on the index result, the labor cost is reduced, and the execution efficiency of the AB experiment is improved.

Description

technical field [0001] The invention relates to the technical field of computer data processing and application, in particular to a large-scale automatic AB experiment shunting method based on indicators. Background technique [0002] AB experiment is to make two (A / B) or more (A / B / N) versions in the product application or page or process, in the same time dimension, divide the visitors with the same or similar components from the total user traffic Groups to randomly access these different experimental versions, run different strategies or display different interfaces and processes, set up different experimental schemes for different experimental versions, collect user experience data and business data of each group, observe and analyze The effect of experimental indicators is driven by data to promote product iteration, verify algorithm effect, obtain business output, etc., obtain experimental conclusions, and finally analyze and evaluate the optimal version, which is offi...

AI Technical Summary

Problems solved by technology

[0004] In the existing traffic distribution method, when there are thousands of items (items), multiple rounds of manual AB experiment iterations will be carried out, and the AB experiment will be manually monitored on and off the line, and finally the optimal item will be obtained as the final result. Launched an AB experiment, but this diversion method uses manual monitoring, which requires a lot of manual intervention, poor diversion effect, and high labor costs

Images