Method and device for rights push

Abstract

The embodiments of this specification provide a method and device for pushing rights and interests, which can be generally used in various business scenarios where rights and interests are to be pushed. This method combines the ideas of meta-learning and reinforcement learning, takes different business scenarios as the tasks of meta-learning, combines the advantage design of reinforcement learning for long-term profit maximization decision-making, and obtains a machine learning solution for the optimal strategy of multi-scenario dynamic decision-making, so that Activity modeling for short cycles and new business scenarios becomes possible. When running online, the business coding network adopts an online strategy, requiring only a small amount of business data to quickly adapt to new business scenarios, obtain the code representation carrying business scenario information, and make push strategy decisions based on it, further improving the push strategy and business relevance of the scene. The network parameters in the agent can also be updated with an offline strategy, which improves the usability of the model.

Description

technical field [0001] One or more embodiments of this specification relate to the field of computer technologies, and in particular, to a method and apparatus for pushing rights. Background technique [0002] With the development of Internet technology, integration has gradually become the development trend of various network platforms. The integration of the platform can link various related business parties together, and can realize business interoperability. For example, a payment platform can link multiple payment channels and access multiple payment scenarios, such as shopping scenarios, water, electricity and gas payment scenarios, phone bill recharge scenarios, and takeaway scenarios. These scenarios can be jumped to the current payment platform through the relevant platform link, or can be entered through the applet or micro-application embedded in the payment platform. The payment channel may be, for example, credit card payment, savings card payment, online loan...

AI Technical Summary

Problems solved by technology

The unified model is suitable for scenarios with rich historical data, and the effect is poor for scenarios with insufficient or no data, which may not meet the business requirements, and the training process may be complicated, and it is necessary to continuously collect data to retrain the model

Modeling for different scenarios usually requires maintaining multiple models, for example, the number of models is equal to the number of scenarios, the cost of training and maintenance is high, and when the scene data is small, the model cannot be trained or the trained model is not accurate enough

Images