Bank public opinion risk control method and system based on knowledge distillation for model compression

Abstract

The invention discloses a bank public opinion risk control method and system for model compression based on knowledge distillation, including: a distributed message queue module, a model compression entity identification module, and a model compression public opinion classification module; the output terminal of the distributed message queue module is connected to the The input end of the model compression entity recognition module is connected, and the output end of the model compression entity recognition module is connected with the input end of the model compression public opinion classification module. In the present invention, the industry-leading pre-training model is compressed by the technology of knowledge distillation, the structure of the neural network model is simplified, and the performance and effect of the model can be guaranteed when the model parameters are few, and the model prediction can be improved. The accuracy rate can also reduce the prediction time. Through the deployment of distributed message queue and distributed cache, the response time of model prediction can be reduced to meet the high real-time requirements of banks.

Description

technical field [0001] The invention relates to the technical field of bank public opinion risk control systems, in particular to a bank public opinion risk control method and system for model compression based on knowledge distillation. Background technique [0002] At present, banks have higher and higher requirements for the accuracy and real-time of public opinion risk control. The traditional method is to purchase public opinion data from third-party data providers, but there are the following pain points: the public opinion risk early warning is not accurate enough, and the public opinion risk early warning is not timely enough. Increased labor costs for banks: Due to the inaccuracy of public opinion risk early warning, it requires a lot of manual review costs. Although the industry's popular pre-training models have high accuracy, the model parameters are too large, the hardware cost is high, and the model prediction time is long , it is difficult to implement the pro...

AI Technical Summary

Problems solved by technology

[0002] At present, banks have higher and higher requirements for the accuracy and real-time performance of public opinion risk control. The traditional method is to purchase public opinion data from third-party data providers, but there are the following pain points: the public opinion risk warning is not accurate enough, and the public opinion risk warning is not timely enough

Increased bank labor costs: Due to the inaccuracy of public opinion risk warnings, large manual review costs are required. Although the pre-training models popular in the industry have high accuracy, the model parameters are too large, the hardware cost is high, and the model prediction time is long. , it is difficult to implement the project

Images