Sorting learning method based on dual-cooperation generative adversarial network

Abstract

The invention discloses a sorting learning method based on a dual-cooperation generative adversarial network. The generative adversarial network comprises three roles: a tutor, a generator and a discriminator. The method mainly comprises the following steps of: S1, constructing a data set required by sorting learning; S2, creating a tutor network and a generator network; S3, establishing a discriminator network; S4, setting data details and parameters; S5, iteratively training a T-G network model; S6, iteratively training a G-D network model; and S7, returning a multi-index validity test result. According to the invention, the generative adversarial network for sorting learning is improved, and on the basis of a previous knowledge system, through interactive help and supervision between the tutor and the generator, the difference between the generator and the discriminator is reduced, the efficiency and effectiveness of the generator are improved, the risk of over-fitting is greatly weakened in the training process, and the performance of sorting learning is improved.

Description

Technical field [0001] This method involves the fields of information retrieval and machine learning, and is particularly closely related to the generative confrontation network in ranking learning. Background technique [0002] As the intersection of information retrieval and machine learning, ranking learning has been widely used in different ranking tasks based on supervised machine learning methods. The ranking learning model takes the ranking-oriented loss function as the optimization goal. The final training model can effectively adapt to the ranking scheme and provide a list of relevant documents for the query. Therefore, the construction of the loss function for ranking learning is very important. According to the type of loss function, the ranking learning model is usually divided into three categories, namely point, pair and list. These three types of models also correspond to three level loss functions, and their construction methods also correspond to three differe...

AI Technical Summary

Problems solved by technology

[0007] The present invention is a dual collaborative generation confrontation network for sorting learning. Its purpose is to solve the limitations of the GAN network, that is, to alleviate the problem of insufficient training, and to introduce the NDCG index into the previous knowledge system, which greatly weakens the overfitting Degree

Images