CN-121982369-A - Trade product classification method and system

CN121982369ACN 121982369 ACN121982369 ACN 121982369ACN-121982369-A

Abstract

The invention discloses a trade product classification method and system, which are characterized by acquiring multi-mode data of a target product, wherein the multi-mode data comprises text data and image data, screening product keywords from the text data, constructing text features, performing first-level classification on the product keywords to obtain a first classification result, selecting a target neural network from pretrained convolutional neural networks aiming at different product types according to the first classification result, adopting the target neural network to extract image features from the image data, performing feature fusion on the text features and the image features to obtain multi-mode features, and performing second-level classification on the target product according to the multi-mode features to obtain a second classification result. By adopting the embodiment of the invention, the advantages of different data can be fully utilized, the classification error can be effectively reduced, and the practicability and reliability can be improved.

Inventors

CHENG PENGFEI
CHEN DIANKUN
ZHANG YUE
GUO XIAOQING
WANG YIWEI
ZHANG XIAOMING
WANG JIAJUN
FU JINYUAN
Pu Lianhua
BAO YUE
HUANG YIFEI
CUI JINYIN
LU SIYUAN

Assignees

广州市城市规划勘测设计研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20251225

Claims (10)

1. A method of classifying a trade product, comprising: the method comprises the steps of obtaining multi-modal data of a target product, wherein the multi-modal data comprises text data and image data; Screening product keywords from the text data, constructing text features, and performing first-level classification on the product keywords to obtain a first classification result; Selecting a target neural network from the pre-trained convolutional neural networks aiming at different product types according to the first classification result, and extracting image features from the image data by adopting the target neural network; carrying out feature fusion on the text features and the image features to obtain multi-modal features; And carrying out second-level classification on the target product according to the multi-mode characteristics to obtain a second classification result.
2. A method of classifying a trade product according to claim 1, wherein said obtaining multi-modal data for the target product comprises: Acquiring customs declaration data of a target product; obtaining product data of a target product on an electronic commerce platform by adopting a crawler algorithm; Classifying the customs declaration data and the product data into text data and image data to obtain multi-mode data.
3. The method for classifying a trade product according to claim 1, wherein said screening product keywords from said text data, constructing text features, and performing a first hierarchical classification on said product keywords to obtain a first classification result comprises: Dividing the text data into a plurality of ideographic units by adopting a word segmentation technology; Extracting product keywords from the ideographic units according to the parts of speech of the ideographic units, and constructing text features; Establishing a keyword mapping table according to chapter levels or mesh levels of trade product classification rules; and classifying the product keywords according to the keyword mapping table to obtain a first classification result.
4. A method of classifying a trade product according to claim 3, wherein said extracting product keywords from said ideographic units based on the parts of speech of said ideographic units, constructing text features, comprises: Extracting product keywords from the ideographic units according to the parts of speech of the ideographic units; dividing the product keywords into coarse-granularity keywords and fine-granularity keywords; Identifying hierarchical relationships between the coarse-grained keywords and the fine-grained keywords; and constructing coarse-granularity text features, fine-granularity text features and associated text features according to the hierarchical relationship, and taking the coarse-granularity text features, the fine-granularity text features and the associated text features as text features.
5. The method for classifying a trade product according to claim 1, wherein said selecting a target neural network from a pretrained convolutional neural network for different product classes based on said first classification result, and extracting image features from said image data using said target neural network comprises: Selecting a target neural network from the pre-trained convolutional neural networks aiming at different product types according to the first classification result; extracting texture features, shape features and color features of the image data by adopting the target neural network to serve as initial image features; and adopting a principal component analysis method to perform dimension reduction treatment on the image characteristics to obtain the image characteristics.
6. The method of claim 1, wherein prior to said feature fusing said text features and said image features to obtain multi-modal features, said method of classifying a trade product further comprises: And performing OCR (optical character recognition) on the image data to obtain second text data, and updating the text characteristics according to the second text data.
7. A method of classifying a trade product according to claim 1 or 6, wherein said feature fusing said text features and said image features to obtain multi-modal features comprises: Carrying out association strengthening on the text features and the image features by adopting an attention mechanism to obtain association features; And extracting multi-level attention features from the associated features according to preset constraint level conditions to obtain multi-mode features.
8. A method of classifying a trade product according to claim 1, wherein said classifying said target product according to said multi-modal characteristics at a second level to obtain a second classification result comprises: Obtaining granularity requirements of trade product classification rules corresponding to the target product according to the entrance and exit data of the target product; According to the granularity requirement, adjusting parameters of a pre-trained multi-layer perceptron; And inputting the multi-mode characteristics into the multi-layer perceptron, and performing second-level classification on the target product to obtain a second classification result.
9. The method of claim 8, wherein the pre-trained multi-layer perceptron comprises an input layer, a plurality of hidden layers, and an output layer, the input layer comprising a texture mask and a functional mask.
10. A trade product classification system, comprising: the system comprises a data acquisition module, a target product acquisition module and a display module, wherein the data acquisition module is used for acquiring multi-modal data of the target product, and the multi-modal data comprises text data and image data; The text feature extraction module is used for screening product keywords from the text data, constructing text features, and carrying out first-level classification on the product keywords to obtain a first classification result; The image feature extraction module is used for selecting a target neural network from the pretrained convolutional neural networks aiming at different product types according to the first classification result, and extracting image features from the image data by adopting the target neural network; The feature fusion module is used for carrying out feature fusion on the text features and the image features to obtain multi-modal features; And the product classification module is used for carrying out second-level classification on the target product according to the multi-mode characteristics to obtain a second classification result.

Description

Trade product classification method and system Technical Field The invention relates to the technical field of data processing, in particular to a trade product classification method and system. Background When trade activities are increasingly frequent, the method has important significance in accurately classifying and identifying massive import and export products. The declaration person is generally required to conduct class declaration on the products at the entrance, but the declaration person is possibly not matched with the actual properties of the products due to unfamiliar commodity properties and HS coding rules, and in addition, the fine granularity requirements and specific classification standards of the codes in different countries are different. Therefore, approval classification of the declared code is still required. The traditional trade product classification mainly depends on manual operation according to the current classification standard, so that the method is low in efficiency, classification errors are easy to cause due to human factors, and the limitation of manual classification is more remarkable especially when the method faces complex and diverse product information, continuously updated product types and a large amount of product data. The prior art is difficult to meet the actual requirements of rapid and accurate classification and identification of products in trade. Disclosure of Invention The invention aims to provide a trade product classification method and a trade product classification system, which can be suitable for trade scenes with complicated types and various characteristics, and effectively reduce classification errors and improve practicability and reliability by fully utilizing different data advantages. The embodiment of the invention provides a trade product classification method, which comprises the following steps: the method comprises the steps of obtaining multi-modal data of a target product, wherein the multi-modal data comprises text data and image data; Screening product keywords from the text data, constructing text features, and performing first-level classification on the product keywords to obtain a first classification result; Selecting a target neural network from the pre-trained convolutional neural networks aiming at different product types according to the first classification result, and extracting image features from the image data by adopting the target neural network; carrying out feature fusion on the text features and the image features to obtain multi-modal features; And carrying out second-level classification on the target product according to the multi-mode characteristics to obtain a second classification result. As an improvement of the above solution, the obtaining multi-modal data of the target product includes: Acquiring customs declaration data of a target product; obtaining product data of a target product on an electronic commerce platform by adopting a crawler algorithm; Classifying the customs declaration data and the product data into text data and image data to obtain multi-mode data. As an improvement of the above solution, the screening product keywords from the text data, constructing text features, and performing a first hierarchical classification on the product keywords to obtain a first classification result, including: Dividing the text data into a plurality of ideographic units by adopting a word segmentation technology; Extracting product keywords from the ideographic units according to the parts of speech of the ideographic units, and constructing text features; Establishing a keyword mapping table according to chapter levels or mesh levels of trade product classification rules; and classifying the product keywords according to the keyword mapping table to obtain a first classification result. As an improvement of the above solution, selecting a target neural network from the pretrained convolutional neural networks for different products according to the first classification result, and extracting image features from the image data by using the target neural network includes: Selecting a target neural network from the pre-trained convolutional neural networks aiming at different product types according to the first classification result; extracting texture features, shape features and color features of the image data by adopting the target neural network to serve as initial image features; and adopting a principal component analysis method to perform dimension reduction treatment on the image characteristics to obtain the image characteristics. As an improvement of the above solution, the extracting product keywords from the ideographic units according to the parts of speech of the ideographic units, and constructing text features includes: Extracting product keywords from the ideographic units according to the parts of speech of the ideographic units; dividing the product keywords into coarse-granularity