CN-121352673-B - Tracing method and device for cross-border return reasons, electronic equipment and storage medium

CN121352673BCN 121352673 BCN121352673 BCN 121352673BCN-121352673-B

Abstract

The application relates to the technical field of tracing of cross-border return reasons, and discloses a method, a device, electronic equipment and a storage medium for tracing the cross-border return reasons, wherein the method comprises the following steps: and carrying out deep fusion on the multilingual text comment and the image evidence by adopting a joint semantic vector technology, dynamically associating front-end return data with rear-end supply chain data through a root cause map, forming a preset root cause map, and realizing tracing of return reasons of return comment information. The method has the advantages that the data island is broken, the deep causal relationship can be accurately identified, the root cause misjudgment rate is obviously reduced, the cross-border goods returning root positioning time is shortened, the accuracy of dividing the responsibility of the supply chain is improved, and therefore the accurate tracing of the cross-border goods returning cause is realized.

Inventors

WANG XUETENG
CHEN DAWEI
ZENG WEIJIA
XU LINGZI
HE ZHONGQING
XU KUNYANG
ZHAO SHAN
Xie Qiongbing

Assignees

深圳市明心数智科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251217

Claims (8)

1.A method for tracing a cross-border return cause, the method comprising: acquiring target return comment information to be traced; Inputting the target return comment information into a preset root cause map to acquire a causal relationship corresponding to the target return comment information; acquiring a plurality of legal terms related to the target return comment information to form a temporary legal database; generating a hierarchical improvement scheme based on the temporal rules database and the causal relationships; transmitting the hierarchical improvement scheme to a first designated terminal; before the step of inputting the target return comment information into a preset root cause map to obtain the causal relationship corresponding to the target return comment information, the method further comprises the following steps: acquiring a plurality of groups of reference return comment information and labels corresponding to each group of reference return comment information, wherein the labels are reference causal relations corresponding to the reference return comment information; extracting the image-text data of each piece of reference return comment information; Generating a corresponding joint semantic vector based on each image-text data; generating the preset root cause map based on each joint semantic vector and the corresponding label; the step of generating the preset root cause map based on each joint semantic vector and the corresponding label comprises the following steps: extracting production batch data and logistics sensing data in the reference causal relationship; correlating the joint semantic vector, the production batch data and the logistics sensing data to construct a multidimensional correlation matrix; constructing an initial map by taking entities in the joint semantic vector, the production batch data and the logistics sensing data as nodes and taking causal relations among the entities as edges; And processing the multi-dimensional association matrix and the initial map by using a graph neural network to quantify the influence weights among the nodes and generate the preset root cause map.
2. The method of claim 1, wherein the step of generating a corresponding joint semantic vector based on each of the teletext data comprises: extracting text semantic features of the text content in the reference return comment information by using a preset large model, and identifying visual concepts contained in the image content in the reference return comment information by using a computer visual model; Vectorizing the text semantic features to obtain text vectors, and vectorizing the visual concepts to obtain visual vectors; And fusing the text vector and the visual vector according to a preset weighting rule to generate the joint semantic vector.
3. The method for tracing a cross-border return reason according to claim 2, wherein after the step of inputting the target return comment information into a preset root cause map to obtain a causal relationship corresponding to the target return comment information, further comprises: collecting actual effect data of an improvement strategy implemented based on the causal relationship; Based on the actual effect data, fine tuning model parameters of the preset large model and the computer vision model by LoRA technology to obtain an updated preset large model and computer vision model; regenerating a joint semantic vector according to the updated preset large model and the updated computer vision model; And updating weight distribution in the preset root cause map according to the regenerated joint semantic vector.
4. The method for tracing a cross-border return cause according to claim 1, wherein the step of using a graph neural network to process the multi-dimensional correlation matrix and the initial graph to quantify the impact weight between nodes, and generating the preset root cause graph further comprises: acquiring a plurality of groups of time sequence data, wherein the time sequence data is data which occurs before the return comment information in a production or logistics link; Carrying out causal strength verification on the edges of the preset root cause graph by utilizing a plurality of groups of time sequence data; and if the result of the causal strength verification is that the verification passes, judging that the preset root cause map is qualified.
5. The method for tracing a cross-border return reason according to claim 1, wherein the step of obtaining a plurality of sets of reference return comment information and labels corresponding to each set of reference return comment information comprises: acquiring a plurality of groups of historical return comment information to form a historical return comment information resource pool; inputting each historical return comment information into a preset root cause classification model to obtain an information entropy value of each historical return comment information; taking the historical return comment information with the information entropy value larger than a preset entropy value as reference return comment information; and sending the reference return comment information to a second preset terminal to obtain a corresponding label.
6. A cross-border return cause tracing apparatus, the apparatus comprising: the target return comment information acquisition module is used for acquiring target return comment information to be traced; The causal relation acquisition module is used for inputting the target return comment information into a preset root cause map so as to acquire a causal relation corresponding to the target return comment information; A temporary regulation database forming module for acquiring a plurality of regulation terms related to the target return comment information to form a temporary regulation database; A hierarchy level enhancement scheme generation module for generating a hierarchy level enhancement scheme based on the temporal regulation database and the causal relationship; A sending module, configured to send the hierarchical improvement scheme to a first designated terminal; The system comprises a label acquisition module, a storage module and a storage module, wherein the label acquisition module is used for acquiring a plurality of groups of reference return comment information and labels corresponding to each group of reference return comment information, wherein the labels are reference causal relations corresponding to the reference return comment information; the image-text data extraction module is used for extracting image-text data of each reference return comment information; The joint semantic vector generation module is used for generating a corresponding joint semantic vector based on each image-text data; The preset root cause map generation module is used for generating the preset root cause map based on each joint semantic vector and the corresponding label; the preset root cause map generation module comprises: the extraction sub-module is used for extracting production batch data and logistics sensing data in the reference causal relationship; The association sub-module is used for associating the joint semantic vector, the production batch data and the logistics sensing data to construct a multidimensional association matrix; the construction submodule is used for constructing an initial map by taking entities in the joint semantic vector, the production batch data and the logistics sensing data as nodes and taking causal relations among the entities as edges; And the quantization sub-module is used for processing the multidimensional association matrix and the initial map by utilizing a graph neural network so as to quantize the influence weights among the nodes and generate the preset root cause map.
7. A computer readable storage medium, characterized in that a computer program is stored, which, when being executed by a processor, causes the processor to perform the steps of the method for tracing a cross-border return cause according to any one of claims 1 to 5.
8. An electronic device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method of tracing a cross-border return cause according to any one of claims 1 to 5.

Description

Tracing method and device for cross-border return reasons, electronic equipment and storage medium Technical Field The present invention relates to the field of tracing technologies for cross-border return reasons, and in particular, to a method, an apparatus, an electronic device, and a storage medium for tracing a cross-border return reason. Background Under the background of globalization e-commerce, cross-border return management faces serious challenges due to the fact that complicated supply chain links are involved, a traditional return analysis tool usually adopts an isolated data processing mode, on one hand, text comments from different markets are analyzed by relying on manual or basic semantic tools, on the other hand, analysis of broken package images and video evidences uploaded by users is independently carried out, effective mutual evidence with text information is difficult, more importantly, feedback information from the front end and supply chain data of the rear end are mutually split to form a data island, and the split processing mode leads to the fact that the error judgment rate is high on one side due to the analysis view angle. For example, merely by comments and images, a problem may be attributed to a "package defect" while in fact the root cause is the violent operation of the logistics sorting link, but this critical information is isolated from being correlated in the log of the logistics system. Disclosure of Invention Based on this, it is necessary to provide a tracing method, a tracing device, an electronic device and a storage medium for the cross-border return reason for the tracing problem of the existing cross-border return reason. A method of tracing a cross-border return reason, the method comprising: acquiring target return comment information to be traced; Inputting the target return comment information into a preset root cause map to acquire a causal relationship corresponding to the target return comment information; before the step of inputting the target return comment information into a preset root cause map to obtain the causal relationship corresponding to the target return comment information, the method further comprises the following steps: acquiring a plurality of groups of reference return comment information and labels corresponding to each group of reference return comment information, wherein the labels are reference causal relations corresponding to the reference return comment information; extracting the image-text data of each piece of reference return comment information; Generating a corresponding joint semantic vector based on each image-text data; and generating the preset root cause map based on each joint semantic vector and the corresponding label. Further, the step of generating a corresponding joint semantic vector based on each of the teletext data comprises: extracting text semantic features of the text content in the reference return comment information by using a preset large model, and identifying visual concepts contained in the image content in the reference return comment information by using a computer visual model; Vectorizing the text semantic features to obtain text vectors, and vectorizing the visual concepts to obtain visual vectors; And fusing the text vector and the visual vector according to a preset weighting rule to generate the joint semantic vector. Further, after the step of inputting the target return comment information into a preset root cause map to obtain a causal relationship corresponding to the target return comment information, the method further includes: collecting actual effect data of an improvement strategy implemented based on the causal relationship; Based on the actual effect data, fine tuning model parameters of the preset large model and the computer vision model by LoRA technology to obtain an updated preset large model and computer vision model; regenerating a joint semantic vector according to the updated preset large model and the updated computer vision model; And updating weight distribution in the preset root cause map according to the regenerated joint semantic vector. Further, after the step of inputting the target return comment information into a preset root cause map to obtain a causal relationship corresponding to the target return comment information, the method further includes: acquiring a plurality of legal terms related to the target return comment information to form a temporary legal database; generating a hierarchical improvement scheme based on the temporal rules database and the causal relationships; And sending the grading improvement scheme to a first appointed terminal. Further, the step of generating the preset root cause map based on each joint semantic vector and the corresponding label includes: extracting production batch data and logistics sensing data in the reference causal relationship; correlating the joint semantic vector, the production batch data and the logist