Strategy execution method and device based on multilayer diversion experiment frame

Abstract

The invention discloses a strategy execution method and device based on a multilayer diversion experiment frame. The method includes steps of 1, distributing a part of flow for strategy combination; 2, reading strategy nodes in strategy links according to sequence; 3, judging whether strategy nodes meet the constraint conditions or not and executing a step 4, or returning to step 2 to read the next strategy node; 4, judging the strategy type, executing a step 4 for distributing strategies to flow labels, and executing a step 6 for experiment strategies; 5, distributing the flow labels for flows, returning to the step 2 to read the next strategy node; 6, realizing the strategy logics and returning to the step 2 to read the next strategy node. The strategies for multiple experiments to be carried out are distributed to multiple experiment layers according to relevancy of the strategies, orthorhombic flow labels are distributed to different experiment layers through algorithm, and multiplexing of flow is realized. Strategy configuration can be performed by the strategy link, so that configuration is provided with high readability and maintainability.

Description

technical field [0001] The invention relates to the field of data mining and analysis, in particular to a strategy execution method and device based on a multi-layer shunting experiment framework. Background technique [0002] After most recommendation systems or new functions and strategies are developed, there will be a set of default parameter configurations. Before going online, a set of mechanisms should be used to conduct experimental tests and evaluations on the parameters, and find out the best parameter configuration for adjustment. At present, the most commonly used evaluation method is A-B test. The experimental test is completed by assigning independent traffic to the experiment. By isolating the traffic, this method can ensure the independence of the experiment and avoid the mutual influence between the experiments. However, the experiment in the A-B test evaluation method is that the traffic of a single layer cannot be reused. As the number of experiments incr...

AI Technical Summary

Problems solved by technology

However, the experiment in the A-B test evaluation method is that the traffic of a single layer cannot be reused. As the number of experiments increases, the traffic that can be allocated to each group of experiments will decrease, because the total traffic is limited, and the exclusivity of the traffic makes the traffic unreusable.

Images