Bank intelligent outbound dialogue system building method based on transverse federal learning

Abstract

The invention discloses a bank intelligent outbound dialogue system building method based on transverse federal learning. The method comprises the following steps: building a consciousness identification model; each bank terminal trains a consciousness identification model of the bank terminal; each bank end uploads model information obtained after training to an aggregation server; the aggregation server averages the received model information uploaded by different bank terminals to obtain average information and then sends the average information to each bank terminal; each bank terminal updates the own consciousness identification model according to the received average information; and repeating the iteration process for a plurality of times and then stopping training. According to the bank intelligent outbound dialogue system building method based on transverse federal learning, each bank end can train the consciousness identification model of the bank end by integrating customer data of other banks on the basis of keeping privacy of the customer data of the bank end, so that the training precision of the consciousness identification model of each bank can be enhanced, and the training efficiency of the consciousness identification model of each bank is improved. And the identification efficiency of the intelligent outbound call is improved.

Description

technical field [0001] The invention relates to a method for building a bank intelligent outbound dialogue system based on horizontal federated learning. Background technique [0002] Although big data is used as the "fuel" for the vigorous development of artificial intelligence applications, in the actual application process, the data we face is often small-scale and fragmented. Especially for banking systems, due to the limitations of banking scenarios, traditional bank-oriented intelligent dialogue systems often encounter various difficulties such as "data islands" and "privacy protection". There are mainly the following disadvantages. [0003] 1. The accumulation and growth of the bank's external data basically tends to zero, and it is impossible to obtain user data from the external network. The improvement of its own model is small, and it is impossible to form a generalized model. [0004] 2. The generalization ability of the model is limited. Basically, if you chan...

AI Technical Summary

Problems solved by technology

[0003] 1. The accumulation and growth of the bank's external data basically tends to zero, and it is impossible to obtain user data from the external network. The improvement of its own model is small, and it is impossible to form a generalized model

[0004] 2. The generalization ability of the model is limited. Basically, if you change a bank, you need to rebuild the model again

[0005] 3. The workload of model optimization is huge. With the increase of customers, model optimization needs to be customized and optimized for customer scenarios, which lacks universality.

[0006] 4. The knowledge of inter-bank model training cannot be shared, resulting in waste at the level of data and computing power

Images