A/B testing method and device aiming at recommendation strategy

Abstract

The invention discloses an A/B testing method and device aiming at a recommendation strategy. A global sole identification number is allocated for each recommendation request, recommendation strategyinformation which includes each recommendation strategy name and the global sole identification number corresponding to each recommendation request and is called in response to each recommendation request is recorded, and pingback information returned by a client side is collected and comprises the global sole identification number of the recommendation request corresponding to the pingback information, according to the global sole identification number, corresponding relations of the recommendation strategy names and the pingback information are determined, and based on data of the multiple determined corresponding relations, recommendation effect analysis is conducted. According to the A/B testing method and device aiming at the recommendation strategy, the global sole identification number is adopted to mark the data generated in one request process and is utilized to associate relevant data, the effect analysis is conducted based on the associated data, collection of the strategy names in the whole path is avoided, and any service strategy in a multi-layer service structure can be subjected to independent effect analysis.

Description

technical field [0001] The present invention relates to the field of recommendation services, and more specifically relates to an A / B testing method and device for recommendation strategies. Background technique [0002] The recommendation service can recommend and display product information for users according to the user's hobbies or search intentions according to a certain recommendation strategy, which is convenient for users to view and browse. In order to improve the recommendation effect, in the selection of the recommendation strategy, it is necessary to use the online A / B test results as a benchmark to judge the recommendation effect of the recommendation strategy. When using a recommendation strategy, when the front-end displays and clicks on the recommended content calculated by the recommendation strategy, the recommendation service system will receive a pingback (automatic reference notification), return the recommendation strategy name, and then count the retu...

AI Technical Summary

Problems solved by technology

However, in actual situations, the service invocation process and path of the recommendation request may be very complicated, w

Images