Research on Fusion of Deep Learning Models Based on Microservice Architecture

Due to the differences in development languages and development frameworks, various difficulties are often faced in the process of integrating microservice architecture and deep learning models. This paper proposes a DND-P (Deep Model in Nacos Discovery Protocol) protocol to realize the rapid integration of deep learning models and microservice architectures. First, this paper selects three representative deep learning models DCN (Deep & Cross Network), DeepFM and xDeepFM in the field of CTR (Click Through Rate) as the research objects, and randomly samples 600,000 of them from the Criteo public data set. The model is trained with 10,000 sample sets to obtain a relatively stable model version. Second, use the Django framework to integrate the trained model as a web application. Finally, the DND-P protocol is directly referenced in Django to implement service registration in the microservice architecture, and the CTR is estimated at the business layer through OpenFeign. Through experiments, it is concluded that the integration of microservice architecture and deep learning model with DND-P protocol can break the barriers between the two, so as to quickly realize the integration of business applications.