A Sequence Recommendation Method for Knowledge Distillation Based on Land Movement Distance

Abstract

The invention discloses a sequence recommendation method for knowledge distillation based on land moving distance. The method includes: constructing a teacher model and a corresponding student model; training the teacher model with a set loss function as the target to obtain a pre-training teacher model; The parameters of the pre-trained teacher model are added to the student model for collaborative training, and only the parameters of the student model are optimized to distill the knowledge of the pre-trained teacher model into the student model. learn the many-to-many mapping relationship between the middle hidden layer of the pre-trained teacher model and the middle hidden layer of the student model; use the trained student model to take the user's historical browsing sequence as input to provide the user with a sequence recommendation service. The student model obtained by the invention can significantly reduce the parameter scale without losing precision, thereby providing users with fast and accurate recommendation services.

Description

technical field [0001] The present invention relates to the technical field of sequence recommendation, and more particularly, to a sequence recommendation method for knowledge distillation based on land moving distance. Background technique [0002] Recommendation system is a very prosperous field in recent years, and it has attracted much attention because of its broad application scenarios and huge commercial value. Simulated sales staff help customers complete the purchase process, and personalized recommendation is to recommend interesting information and products to users according to the user's interest characteristics and purchase behavior. Sequence recommender system is an important branch of recommender system. Its purpose is to accurately recommend users by analyzing the historical browsing sequence of users. It has always been a hot research issue concerned by academia and industry. [0003] At present, the sequence recommendation system has achieved rapid devel...

AI Technical Summary

Problems solved by technology

[0009] The shortcomings of the existing technology are mainly that when performing recommendation services, the number of model parameters is large, and the inference time is long, which makes it difficult to meet the needs in the real world

NextItNet needs to stack a large number of empty convolution residual blocks to achieve better results, resulting in a huge amount of model parameters, and for each input user history browsing sequence, a complete model is required to complete the output prediction, which will train well. It is difficult to deploy the model in actual application, the calculation cost is large, and it takes a long time to infer, so it is difficult to meet the actual needs of users

Images