Commodity sorting neural network model training method and device and electronic equipment

Abstract

The embodiment of the invention discloses a commodity sorting neural network model training method, belongs to the technical field of computers, and is helpful for improving the model training efficiency. The method comprises the following steps of: migrating partial model parameters of a pre-trained merchant sorting neural network model to the commodity sorting neural network model; initializingparameters of partial hidden layers in a first network of a commodity sorting neural network model according to the partial model parameters, and randomly initializing parameters of the hidden layersexcept the partial hidden layers in the commodity sorting neural network model; and respectively obtaining merchant dimension features and commodity dimension features of each commodity training sample, constructing sample data of the corresponding commodity training sample, and training the commodity sorting neural network model.

Description

technical field [0001] The present application relates to the field of computer technology, and in particular to a product ranking neural network model training method, device, electronic equipment and computer-readable storage medium. Background technique [0002] In the product sorting scenario, the usual practice is to collect product samples, and then extract the product features of the collected product samples based on the preset product feature dimensions, and then train the product sorting neural network model based on the extracted product features to output product information. Index scores used for product ranking, such as the probability of placing an order. To train the commodity sorting neural network model in this way, a large number of training samples need to be obtained. Due to the sparse characteristics of commodity samples, the model is not easy to converge during the model training process, and the model parameters need to be learned and adjusted repeat...

AI Technical Summary

Problems solved by technology

Due to the sparse characteristics of commodity samples, the model is not easy to converge during the model training process, and it is necessary to learn and adjust the model parameters repeatedly, and the training efficiency is relatively low.

Moreover, the accuracy of the product ranking index scores output by the neural network model trained by this method is directly affected by the number of product samples. In the case of sparse product samples, the product ranking indicators output by the trained neural network model Scoring accuracy will decrease

Images