Distributed type A/B test method, system and video recommendation system for personalized recommendation system

Abstract

The invention discloses a distributed type A / B test method for a personalized recommendation system. The method comprises the steps of performing module or intersystem distribution on data, and adding module or system identification on the data; performing inter-flow re-distribution on the data in the module or the system, and adding corresponding flow identification on the data during re-distribution; and collecting user behavior point data by a client end, integrating with the data having the module or system identification and the flow identification, and uploading to a server. The distribution strategy disclosed by the invention supports multi-dimension and multi-method distribution realization; the data flow is subjected to multi-dimension and multi-method feedback data tracking, so that multi-dimension and multi-method distinguished statistics can be carried out conveniently; flow information is recorded in the distribution process, namely, a dimension use method in the flow is recorded at the same time, namely corresponding data tracking; and therefore, each detailed processing leaves identification on the data, so that high subsequent handlability is achieved, and analysis requirements of different levels can be satisfied.

Description

technical field [0001] The invention relates to the technical field of video playback, in particular to a distributed A / B testing method and system for a personalized recommendation system and a video recommendation system. Background technique [0002] Chinese patent 105913145 A discloses a data-driven A / B test method. The A / B test design adopted is based on data-driven and modularized tests. In various combinations, various modules need to be constructed, and the test system The amount of development and system construction are relatively complicated. With the increase of test cases, the cost of system maintenance also gradually increases, that is, the test support for multiple combinations is not good. At the same time, the data statistics design corresponding to the feedback is not flexible enough, and the coding statistics can only be carried out for the unit of the module, that is, the overall situation of the version combination is statistically and effect tested, and...

AI Technical Summary

Problems solved by technology

[0002] Chinese patent 105913145 A discloses a data-driven A / B test method. The A / B test design adopted is based on data-driven and modularized tests. In various combinations, various modules need to be constructed, and the test system The amount of development and system construction are relatively complicated. With the increase of test cases, the cost of system maintenance also gradually increases, that is, the test support for multiple combinations is not good.

At the same time, the data statistics design corresponding to the feedback is not flexible enough, and the coding statistics can only be carried out for the unit of the module, that is, the overall situation of the version combination is statistically and effect tested, and it is difficult to test and compare the internal details, that is, it is difficult to find out the reasons for the combination of versions The method in that specific process leads to a decline in its measurement indicators. Inquiring about the details of the problem requires A / B testing again, which virtually increases the workload due to the unreasonable design.

Images