Method and device for training model, equipment and storage medium

Abstract

The invention discloses a method and device for training a model. The specific implementation scheme comprises the steps: obtaining a training sample set, wherein a training sample in the training sample set comprises the sub-data of each data set in all data sets of one batch, the click rate of each object, and the click rate order; by utilizing a machine learning algorithm, taking each sub-data of each data set in all the data sets as input data, taking the click rate and click rate sequence of each object corresponding to the input data set as expected output data, and training to obtain a click rate prediction model, wherein a loss function of the click rate prediction model is constructed by using a first loss function and a second loss function, the first loss function is used for calculating the accuracy of the click rate of each object, and the second loss function is used for calculating the sorting capability of the click rate sorting. According to the scheme, the problem that an existing click rate prediction model cannot perceive the actual sorting capability of the model in the training process is solved.

Description

technical field [0001] The embodiments of the present application relate to the field of computer technology, specifically to the field of information processing technology, and especially to a method and device for training a model. Background technique [0002] The e-commerce system solves the problem of information overload by integrating a personalized recommendation system, helping users quickly find the products they are interested in, and improving the efficiency of all parties including users and platforms. The Click-Through-Rate (CTR) prediction model is a very critical part of the personalized recommendation system. When the CTR prediction model serves online, it predicts and sorts the click-through rate of all candidate products for each specific request. , the recommendation system sorts according to the predicted value of the click-through rate, and selects several products with the highest click-through rate to display to the user. Academia and the industry us...

AI Technical Summary

Problems solved by technology

Academia and the industry usually use the AUC (Area Under Curve) indicator to measure the sorting ability of the CTR prediction model. However, the model uses the cross-entropy loss function or the hinge loss function as a supervisory indicator during the training process, and does not directly introduce the AUC (area under curve). the curve) as a supervisory index, therefore, the existing click-through rate prediction model, under the premise of ensuring the number of training logs and the accuracy of the predicted value, leads to a difference between the training supervisory index and the actual effect measurement index after the model training, which cannot Perceived Model Actual Ranking Capabilities During Training

Images