Commodity classification method based on multi-modal deep neural network model

Abstract

The invention discloses a commodity classification method based on a multi-modal deep neural network model. The method comprises the steps: firstly converting text information into a word vector whichcan better reflect the relation between words, and carrying out the feature extraction of commodity text description through employing a text classification model TextCNN; in order to prevented over-fitting, using image augmentation operations such as random image flipping and random image brightness change on the images, and then inputting the processed result into a commodity image classification model ResNet101 to perform feature extraction on the commodity images; flattening the feature vectors extracted from the two models through a flatten function, directly connecting the feature vectors of the two modal data on the feature dimension, and finally sending the feature vectors to a classifier to classify the commodities. According to the commodity classification method, the limitationof traditional single-mode data on commodity classification is prevented, the text data and the picture data are combined, and the classification performance and the classification accuracy are better than those of a model using the single-mode data.

Description

technical field [0001] The invention relates to a commodity classification method, in particular to a commodity classification method based on a multimodal deep neural network model, and belongs to the technical field of commodity classification. Background technique [0002] With the vigorous development of e-commerce platforms, the category system construction method based on big data mining will gradually replace the method based on manual construction due to its advantages of speed, efficiency, automation, and low cost. In modern times, Internet technology is developing at a high speed every day, and e-commerce is becoming more and more important in people's lives. Online shopping has become the first choice for many people to buy goods. With the increase in the number of commodities, finding a commodity classification method with fast classification speed and high classification accuracy is not only an urgent need for users to choose the commodities they need from a lar...

AI Technical Summary

Problems solved by technology

This method solves the problem that data fusion methods may have difficulties in directly combining different modal data in specific classification tasks and the problem that decision fusion methods cannot learn the correlation between different modalities well.

Images