CN-121998651-A - Knowledge base-based cross-border logistics AI customer service system processing method

CN121998651ACN 121998651 ACN121998651 ACN 121998651ACN-121998651-A

Abstract

The invention relates to a processing method of a cross-border logistics AI customer service system based on a knowledge base, which comprises the following steps of acquiring and explaining a user session, acquiring historical session information through user information, judging whether the user session and the historical session information contain picture information, processing pictures or characters in the pictures, compiling a prompt word to complement semantic information in the historical session information into the user session by using a large model, classifying the user session based on a multi-classification model, wherein the types of the user session comprise boring type, statement fact type, unclear problem type, manual intervention type, exchange rate problem type, real-time query database type and combined knowledge base type, and calling corresponding models to answer user problems according to different user session classification types. The invention realizes the intelligent, automatic, high-reliability and low-cost enterprise-level customer service.

Inventors

ZENG LINGWEI
WANG WEIJIAN
HU WEI
Cai Daqin
LI PENG
Gao Minzhi
LI BOHAO
LIU HANYANG
Xiong Mingdao
ZENG HONGXING
ZHENG JIE
XIA QIANG

Assignees

深圳市西邮智仓科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251224

Claims (10)

1. The processing method of the cross-border logistics AI customer service system based on the knowledge base is characterized by comprising the following steps of: s100, acquiring and explaining a user session, and acquiring historical session information through user information; S200, judging whether the user session and the history session information contain picture information or not, and processing the pictures or the characters in the pictures; S300, compiling a prompt word, and completing semantic information in historical session information into a user session by using a large model; s400, classifying user sessions based on a multi-classification model, wherein the types of the user sessions comprise boring type, statement fact type, problem unclear type, manual intervention type, exchange rate problem type, real-time query database type and combined knowledge base type; S500, calling a corresponding model to answer the user questions according to different user session classification types.
2. The method for processing a knowledge base-based cross-border logistics AI customer service system as set forth in claim 1, wherein in step S200, If the user session and the history session information do not contain the picture information, correcting wrongly written characters in the user problem by writing a prompt word and utilizing a large model; And if the user session and the history session information contain picture information, extracting text information in the picture by using a large model through writing a prompt word.
3. The method for processing a knowledge base-based cross-border logistics AI customer service system as set forth in claim 1, wherein in step S400, The classification model classifies an input text by utilizing a neural network architecture, firstly, needs vectorizing an input user session, secondly, connects vectorizing results with a series of neural network hierarchical structures, and calculates the probability between a text vector and each category, wherein the classification category corresponding to the text is the classification category corresponding to the maximum probability.
4. The method for processing a knowledge base-based cross-border logistics AI customer service system as set forth in claim 1, wherein in step S500, When the user session is of the chatting type, the user questions are answered by the large model through writing chatting type prompt words; when the user session is Chen Shushi real types, the user questions are answered by the large model by writing statement fact prompt words; When the user session is the question unclear, the user questions are answered by the large model by writing the question unclear prompting words; When the user session is of the type requiring manual intervention, the user question type is of the type requiring manual intervention, the transfer manual service information is answered and the manual service is converted.
5. The method for processing a knowledge base-based cross-border logistics AI customer service system as set forth in claim 1, wherein, When the user session is an exchange rate question type, firstly writing an exchange rate information extraction prompt word, extracting conversion amount, inquiry time and inquiry currency information of the user in relation to the exchange rate question by utilizing a large model, calling an exchange rate inquiry API according to the extracted inquiry time and inquiry currency, inquiring a corresponding exchange rate, and answering the user question by utilizing the large model by writing the exchange rate conversion prompt word according to the extracted conversion amount and the corresponding exchange rate.
6. The method for processing a knowledge base-based cross-border logistics AI customer service system as set forth in claim 1, wherein, When the user session is a real-time query database type, firstly, extracting a single number in a user problem through a regular expression, extracting a query target in the user problem through writing a query target prompt word, combining the matched single number and a corresponding query target into an SQL query statement, querying corresponding real-time data in a service database, and sleeving the queried real-time data into a reply template for answer reply.
7. The method for processing a knowledge base-based cross-border logistics AI customer service system as set forth in claim 1, wherein, When a user session is a combined knowledge base type, judging whether a problem of the user belongs to a dynamic workflow through a semantic similarity model, wherein the semantic similarity model is used for calculating semantic similarity degree between different texts, judging whether at least two texts are synonymous or not, vectorizing the at least two texts to obtain text vectors, and carrying out cosine similarity calculation on the two text vectors to obtain cosine similarity, wherein the higher the cosine similarity value is, the more similar the two texts are, and the dynamic workflow is a condition that a plurality of steps are needed to carry out joint processing when one problem of the user session is replied; If the user session belongs to the dynamic workflow, a first step flow in the dynamic workflow is obtained, the first step flow in the dynamic action flow is executed, if the first step flow is inquiring user information, the inquired information is directly replied, if the first step flow is inquiring a database or combining a knowledge base, corresponding flow operation is carried out, an operation result is stored, the next step flow in the dynamic workflow is obtained, and the result is output until all the flow steps in the dynamic workflow are processed; If the user session does not belong to a dynamic workflow, extracting keywords in the user problem by training a named entity recognition model, wherein the named entity recognition model is an end-to-end neural network model, vectorizing a text, connecting a multi-layer neural network layer structure, calculating a starting position and an ending position of the keywords in the text, combining all words from the starting position to the ending position into keywords, extracting an intention result in the user problem by writing intention recognition prompt words, and judging whether answering the user session needs to be combined with a knowledge base or not by using a semantic similarity model.
8. The method for processing a knowledge base-based cross-border logistics AI customer service system as set forth in claim 7, wherein, If the user session needs to combine the knowledge graph knowledge base, calculating the text similarity between the key words in the user session and the nodes and edges in the knowledge graph through cosine similarity, filtering the nodes and edges with the text similarity smaller than a first preset value, further performing correlation filtering through a reordering model of a large model through the nodes and edges filtered through cosine similarity, wherein the reordering model is used for simultaneously performing text correlation judgment on one text and a plurality of texts through the large model, sequencing the texts from high to low according to the correlation, taking the top N nodes and edges with the highest correlation, inquiring the nodes and edges connected in the links in the knowledge graph through the top N nodes and edges with the highest correlation, judging the knowledge graph triplet data connected in the inquired links with the user problem through writing knowledge graph knowledge filtering prompting words, judging whether the triplet data can answer the user problem, filtering the triplet data irrelevant to the problem, generating prompting words through writing the knowledge graph answers, and answering the user session through the large model; If the user session needs to be combined with the document knowledge base, calculating the text similarity between the keywords in the user session and the keywords in the document knowledge base through cosine similarity, filtering out the document knowledge base keywords with the text similarity smaller than a second preset value, further carrying out correlation filtering on the keywords of the document knowledge base filtered through cosine similarity through a reordering model of a large model, acquiring the first M keywords with highest correlation, searching in a document block by the first M keywords with highest correlation, finding all candidate knowledge base documents containing the keywords, carrying out semantic screening on the candidate knowledge base documents through a reordering model of the large model, acquiring the first K candidate knowledge base documents with highest correlation, screening out candidate knowledge base documents strongly correlated with the user problem through writing document correlation prompting words by using the large model, generating prompting words by writing document knowledge base answers, and answering the user session by using the large model.
9. The method for processing a cross-border logistics AI customer service system based on a knowledge base according to claim 8, wherein the knowledge base is constructed by the following steps: collecting related domain document data; Writing a prompt word, extracting entities in the document by using a large model, and aligning the semantics of the entities to obtain aligned entities; writing a prompt word, extracting the relation in all documents by using a large model, and aligning the relation semantically to obtain an aligned relation; vectorizing the aligned entity and the aligned relation, and storing the vectorization result into a vector database; and combining all the entities and the relations into knowledge-graph triple data, and storing the knowledge-graph triple data into a knowledge-graph knowledge base.
10. The method for processing a knowledge base-based cross-border logistics AI customer service system of claim 8, wherein the document knowledge base is constructed by: collecting related field document data; Cutting and sorting the characters of the document by using the large model; Writing a prompt word, and extracting key words from the document after the block arrangement by using a large model; vectorizing the document and the corresponding keywords after the block arrangement, and storing vectorization results into a vector knowledge base; and storing all the documents subjected to the block cutting and sorting and the corresponding keywords into a document knowledge base.

Description

Knowledge base-based cross-border logistics AI customer service system processing method Technical Field The invention belongs to the field of customer service system processing, and particularly relates to a cross-border logistics AI customer service system and method based on a knowledge base. Background With the rapid development of cross-border electronic commerce and international logistics business, enterprises need to deal with a large number of customer consultations involving complex problems such as freight, clearance, storage, exchange rate, return, dispatch and the like every day. The intelligent customer service system gradually replaces the traditional manual customer service, and becomes a core tool for enterprises to carry out customer service, product consultation and after-sales support. The traditional customer service mode has the problems of high labor cost, limited service time, slow response speed, difficult standardization of service quality and the like, and the intelligent customer service system can work continuously for 7x24 hours, so that the operation cost is greatly reduced, and the service efficiency is improved. The existing intelligent customer service system mainly goes through the following two development stages: 1) Primary system based on rule and keyword matching Such systems rely on a pre-set rule base and keyword base. When a user enters a query, the system retrieves the preset answer through pattern matching or simple logic rules. The method has the advantages of clear rule and strong controllability, and has the defects of high maintenance cost and incapability of processing continuous multi-round interaction requiring context information. 2) Deep learning and Natural Language Understanding (NLU) based generative system In recent years, the generation AI customer service based on a Large Language Model (LLM) has become the mainstream. The system can be used for deeply understanding the intention of the user, generating smooth and natural replies and remarkably improving the intelligent level. The present inventors have found that this solution still has the following limitations: The fact accuracy is difficult to guarantee ("phantom" problem) that the model may generate information that appears reasonable but in fact erroneous or inconsistent with business data, which is fatal in rigorous customer service scenarios (e.g., financial, medical, logistical tracking). Business processes are weak in processing power, and generative models are good at "conversations" but not "offices". For complex business processes (such as refund, order modification and fault repair) requiring database query, external API call, execution condition judgment and completion of multi-step operation, the pure generation model lacks reliable and controllable execution logic, which easily causes process disorder or operation failure. The controllability and the interpretability are poor, the decision process of the system is a 'black box', and when errors occur, the problems are difficult to track and position. 3) Summary of the prior art and bottleneck In summary, the technical bottleneck in the current intelligent customer service field is how to organically combine the powerful language understanding capability of the generated AI with the accuracy, controllability and business process execution force of the conventional rule system. That is, the system needs to have the following capabilities: Accurate intention recognition and multi-round dialogue management, namely accurately understanding a user target and supplementing missing key information through actively clarifying questions. The system is reliably integrated with an external system, can be connected with and inquire a service database in a seamless way, can be connected with an external API interface in a seamless way, and can acquire real-time and accurate data. The complex condition judgment and dynamic workflow arrangement can carry out condition judgment according to the query result, the business rule and the user context, and dynamically select and execute the next operation to form a complete task processing closed loop. At present, no technical solution disclosed is found to systematically solve all the problems. Therefore, a new architecture and method of intelligent customer service system are needed to overcome the defects of the prior art, and to realize the enterprise-level customer service with real intelligence, automation, high reliability and low cost. Disclosure of Invention The invention provides a processing method of a cross-border logistics AI customer service system based on a knowledge base, which aims at solving at least one of the technical problems existing in the prior art. The technical scheme of the invention relates to a processing method of a cross-border logistics AI customer service system based on a knowledge base, which comprises the following steps: s100, acquiring and explaining a user sessio