An agent-based cross-border customer service method and system

Abstract

This invention relates to the technical field of cross-border customer service, and discloses a cross-border customer service method and system based on intelligent agents. The method includes: standardizing customer service inquiries to obtain the target language, the region involved in the inquiry, and the standard text of the customer service inquiry; semantically optimizing the standard text of the customer service inquiry using a cross-border multilingual ambiguity resolution algorithm; identifying the inquiry intent of the semantically optimized text of the customer service inquiry using an intent recognition algorithm combined with a cross-border customer service scenario knowledge base; calculating the matching degree between the inquiry intent recognition features and business nodes using a graph neural network to generate a business solution corresponding to the inquiry intent recognition features; and generating customer service content corresponding to the business solution by combining cross-border customer service service specifications. This invention achieves accurate multilingual recognition, semantic ambiguity resolution, intent matching, compliance verification, and standardized content generation in cross-border customer service, significantly improving the understanding accuracy, response speed, and compliance reliability of the customer service system.

Description

Technical Field

[0001] This invention relates to the field of cross-border customer service, and more particularly to natural language processing, specifically a cross-border customer service method and system based on intelligent agents. Background Technology

[0002] With the rapid development of cross-border e-commerce, international retail platforms, and globalized digital services, cross-border customer service has become a crucial business link connecting user inquiries, platform rules, order fulfillment, and after-sales processing. Compared to traditional local customer service, cross-border customer service faces a more complex business environment: user inquiries come in diverse languages, with expressions varying from colloquial to mixed and regionally specific; the order fulfillment chain is longer, typically involving multiple stages such as the ordering platform, merchants, overseas warehouses, international trunk logistics, clearance points, local delivery, and tax settlement; simultaneously, significant legal differences exist between different countries and regions regarding consumer protection, return periods, tax liability, refusal liability, and refund conditions. This necessitates that cross-border customer service systems cannot rely solely on a single knowledge base for question-and-answer processing. Instead, they need to understand user semantics while combining the current order status, applicable legal conditions, time zone boundaries, logistics nodes, and platform execution rules to generate processing opinions or service commitments that can be genuinely implemented in subsequent business processes.

[0003] In recent years, research on intelligent customer service has mainly focused on multilingual intent recognition, cross-language retrieval, knowledge graph question answering, rule-driven dialogue management, and enhanced retrieval generation based on large language models. These studies have significantly improved the information acquisition, dialogue generation, and knowledge retrieval capabilities of customer service systems in multilingual environments, enabling the system to more naturally understand the inquiries of users in different languages ​​and extract relevant answers from policy documents, help centers, logistics rules, or product descriptions. However, in real-world cross-border customer service applications, user questions are often not simply knowledge inquiries, but rather commitment-based inquiries with clear business consequences, such as how long it will take for a refund to arrive, who bears the taxes after a refusal, when a replacement will be triggered, and whether the goods are still within the return period after signing for them. These types of questions can superficially be answered with knowledge text, but their correct handling depends on dynamic factors such as whether the order has cleared customs, whether the signing time crosses time zones, whether taxes have been paid, whether platform review has been triggered, whether the warehouse can still intercept the goods, and whether the responsible party has been transferred according to the rules. In other words, customer service responses must not only be semantically reasonable but also functionally sound.

[0004] In existing technologies, some solutions have attempted to achieve automated customer service question answering using natural language processing (NLP) technology. For example, patent CN115905480B proposes an NLP-based virtual customer service automated question answering method. This method collects user question text data, preprocesses and classifies it to build a question database; further, it uses the bag-of-words model and TF-IDF model to generate a weight matrix, calculates the TF-IDF vector for new input questions, and matches historical questions using cosine similarity to output the standard answer corresponding to the most similar question. This method has achieved automated question answering functionality in a shared R&D platform, effectively reducing the burden on human customer service and improving question response efficiency. However, it lacks unified verification of real-time execution status, applicable legal domain, effective time point, responsibility boundary, and subsequent execution links, leading to inconsistencies between customer service responses and the actual capabilities of the backend. Moreover, in the context of rapid expansion across multiple businesses and languages, systems based on fixed features and rules struggle to flexibly adapt to new business scenarios and language requirements.

[0005] To address this issue, this invention proposes a cross-border customer service method based on intelligent agents, which helps improve the reliability of customer service responses on cross-border platforms, reduce after-sales costs caused by erroneous promises, reduce the pressure of manual intervention in complex order scenarios, and enhance users' trust in cross-border transaction services. Summary of the Invention

[0006] This invention proposes a cross-border customer service method and system based on intelligent agents. Step S1 utilizes IP region matching, rapid cross-border multilingual recognition, and multi-dimensional ambiguity resolution algorithms, combined with word vector representation, business coding feature construction, and semantic sense score calculation, to solve the technical problems of low multilingual recognition accuracy, difficulty in eliminating polysemous word ambiguity, semantic comprehension deviation caused by colloquial / hybridized consultation text, and difficulty in effectively associating order information with consultation text in cross-border customer service consultations. Step S2, through temporal pooling processing and global semantic vector extraction, combined with cross-border customer service scenario knowledge base matching, intent confidence calculation, and joint parameter optimization, solves the problems of insufficient focus on core semantics and noise in short texts in traditional intent recognition. The technical issues of high interference, low intent matching accuracy, and difficulty in filtering weakly related intents are addressed in step S3. This is achieved by constructing a graph association matrix between intent and business nodes, calculating graph convolutional matching correlation, and combining multi-dimensional constraint scores of orders, weighted differential weights of business nodes, and quantification of execution scores. This solves the technical problems of inaccurate matching between consultation intent and business nodes, and customer service responses deviating from the actual order status, leading to unexecutable solutions. Step S4 combines compliance verification with the laws and regulations of the consultation region and platform rules, and generates standardized customer service content based on cross-border customer service service specifications. This solves the technical problems of difficulty in ensuring the compliance of cross-border customer service responses, poor adaptability of different legal jurisdictions, non-standard customer service content, and the risk of violations in customer service responses.

[0007] To achieve the above objectives, the present invention provides a cross-border customer service method based on intelligent agents, comprising the following steps: S1: Obtain customer service consultation requests from cross-border users, standardize the customer service consultation requests to obtain the target language of customer service, the region involved in the consultation, and the standard text of customer service consultation, and use a cross-border multilingual ambiguity resolution algorithm to perform semantic optimization processing on the standard text of customer service consultation to obtain semantically optimized text of customer service consultation. S2: An intent recognition algorithm combining a cross-border customer service scenario knowledge base is used to identify the consultation intent of the semantically optimized text of the customer service consultation, and to generate consultation intent recognition features; S3: Based on the optimized text of the customer service consultation semantics and the consultation intent recognition features, use a graph neural network to calculate the matching degree between the consultation intent recognition features and the business nodes, and generate a business solution corresponding to the consultation intent recognition features; S4: Conduct compliance verification of the business solution based on the laws and regulations of the regions involved in the consultation and the platform rules, and generate customer service content corresponding to the business solution in conjunction with cross-border customer service service specifications.

[0008] As a further improvement of the present invention: Furthermore, in step S1, the customer service inquiry request from the cross-border user is obtained, and the customer service inquiry request is standardized, including: S11: The customer service inquiry request from the cross-border user includes the cross-border user's IP address and the customer service inquiry text. The cross-border user's IP address is matched by region, and the matched region is used as the region involved in the inquiry. S12: Use a cross-border multilingual fast recognition algorithm to identify the language of the customer service consultation text, and use the identified language as the target language for customer service. S13: Perform text preprocessing on the customer service consultation text to obtain the word sequence corresponding to the customer service consultation text, which serves as the standard text for customer service consultation. The text preprocessing includes word segmentation, stop word removal, business terminology standardization, and order association. S14: The target language of customer service, the region involved in the consultation, and the standard text of customer service consultation are used as the standardized processing result of the customer service consultation request.

[0009] Furthermore, step S1, based on the target language of customer service, employs a cross-border multilingual ambiguity resolution algorithm to perform semantic optimization processing on the standard customer service consultation text, and also includes: S15: Perform word vector representation on the standard customer service consultation text to obtain the word vector sequence of the word group sequence in the standard customer service consultation text; S16: Traverse the word vector sequence to obtain word vectors with multiple semantic meanings in the target language of customer service, and generate business coding features of the word vectors in different semantic meanings; S17: Extract the business context features from the word vector sequence, and calculate the semantic meaning score between the business encoding features and the business context features of different semantic meanings; Specifically, word vectors corresponding to the associated order information are extracted from the word vector sequence, and business encoding features of the extracted word vectors are generated as business context features in the word vector sequence; The formula for calculating the semantic meaning score between the business coding features and business context features of different semantic meanings is as follows: ; ; in, Business coding features representing semantic meaning e Business context features Semantic meaning scores between them , These represent the business coding features respectively. The code includes region adaptation code, logistics status adaptation code, payment attribute code, and time attribute code. , These represent the business context features in turn. The code includes region adaptation code, logistics status adaptation code, payment attribute code, and time attribute code. A function for calculating feature similarity. Indicates business coding characteristics Business context features The mismatch penalty parameter between them Represents the penalty function, for ,like If they are not equal, then =1, otherwise =0, All represent penalty weights. Indicates the penalty coefficient; S18: Select the semantic sense with the highest semantic sense score as the true semantic sense of the word vector, and represent the true semantic sense with word vectors. Replace the word vectors with multiple semantic senses in step S16 with the word vector representation of the true semantic sense to obtain the semantically optimized word vector sequence of the customer service consultation standard text. S19: The semantically optimized word vector sequence, the target language of customer service, the regions involved in the consultation, and the standard text of customer service consultation are used as the semantically optimized text of customer service consultation.

[0010] Furthermore, in step S2, an intent recognition algorithm combining a cross-border customer service scenario knowledge base is used to identify the consultation intent of the semantically optimized customer service consultation text, including: S21: Extract the semantically optimized word vector sequence from the semantically optimized text of the customer service inquiry; S22: Perform temporal pooling on the semantically optimized word vector sequence to obtain the global semantic vector of the consultation text; S23: Calculate the similarity and intent confidence between the global semantic vector of the consultation text and the standard features of the scenario intent corresponding to the standard intent category in the cross-border customer service scenario knowledge base, wherein the formulas for calculating the similarity and intent confidence are as follows: ; ; in, Represents the global semantic vector of the consultation text. With the dth standard intent category in the cross-border customer service scenario knowledge base Corresponding scene intent standard features Similarity between them Let L represent the L2 norm, and D represent the number of standard intent categories in the cross-border customer service scenario knowledge base. Represents the global semantic vector of the consultation text. With the dth standard intent category in the cross-border customer service scenario knowledge base Confidence of intent between them Indicates the selection of a set The maximum value in, where N represents the sequence length of the semantically optimized word vector sequence. Represents a logarithmic function; S24: Retain standard intent categories with intent confidence levels higher than a preset confidence threshold, and use the retained standard intent categories and their corresponding intent confidence levels as features for identifying consultation intent.

[0011] Furthermore, step S3, which utilizes a graph neural network to calculate the matching degree between the consultation intent recognition features and the business nodes, includes: S31: The business nodes include logistics query nodes, logistics modification nodes, payment nodes, and after-sales nodes, and a graph association matrix between the standard intent category and the business nodes in the consultation intent recognition features is constructed. S32: Combining the graph association matrix, a graph neural network is used to perform graph convolution calculation on the intent confidence corresponding to the standard intent category in the consultation intent recognition features to obtain the matching association degree between the standard intent category and the business node; S33: For any standard intent category in the consultation intent recognition features, calculate the multi-dimensional constraint score of the standard intent category, select the business node with the highest matching correlation, and calculate the execution score of the standard intent category based on the fixed business weight of the business node and the matching correlation between the standard intent category and the business node. S34: Based on the execution score, generate the execution status of the standard intent category, wherein the execution status of the standard intent category includes full execution, partial execution, and failure to execute.

[0012] Furthermore, step S3, which generates a business solution corresponding to the consultation intent identification features based on the execution status of the standard intent category, also includes: S35: Based on the execution status of the standard intent category and the order association processing result in the standard customer service consultation text, generate the current progress, current business status, and business status basis of the standard intent category; S36: The execution status, current progress, current business status, and business status of the standard intent category are used as the solution for the standard intent category; S37: Take the solutions for all standard intent categories in the consultation intent identification features as the business solutions corresponding to the consultation intent identification features.

[0013] Furthermore, step S4 involves compliance verification of the business solution based on the laws and regulations of the regions involved in the consultation and the platform rules, including: The compliance verification includes verification of regional regulations, platform rules, time limits, and liability and fee compliance. The verification outputs for the regional regulations compliance verification, platform rules compliance verification, time limit compliance verification, and liability and fee compliance verification are 0 or 1, where 1 indicates that the verification has passed and 0 indicates that the verification has failed. The verification outputs of the regional regulatory compliance verification, platform rule compliance verification, time limit compliance verification, and liability and cost compliance verification are weighted. If the weighted result is higher than the preset verification threshold, it means that the business solution has passed the compliance verification.

[0014] Furthermore, in step S4, if the business solution passes compliance verification, the corresponding customer service content is generated based on the cross-border customer service service specifications, including: The customer service content includes simplified expressions of the solutions for each standard intent category in the order identification and business solutions. The expression of the text is adjusted according to the target language of customer service, the region involved in the consultation, and the compliance verification process. The expressed text is then translated into the target language of customer service and used as the customer service content.

[0015] The present invention also provides a cross-border customer service system based on intelligent agents. The cross-border customer service system is composed of multiple intelligent agents, including a cross-border consultation preprocessing intelligent agent, a cross-border intent recognition intelligent agent, a business solution generation intelligent agent, and a compliance verification and customer service content generation intelligent agent, so as to realize the technical steps of the cross-border customer service method based on intelligent agents as described above.

[0016] Compared with existing technologies, this invention proposes a cross-border customer service method and system based on intelligent agents, which has the following beneficial effects: First, this invention achieves deep semantic optimization of customer service consultation text through a multilingual ambiguity resolution algorithm. Specifically, this invention transforms standardized word sequence into dense word vectors, and then constructs business coding features containing four dimensions—region, logistics, payment, and time—for multi-semantic words, achieving structured quantification of semantic dimensions. Furthermore, by extracting related order information to form business context features, and combining similarity calculation with a four-dimensional mismatch penalty mechanism, it accurately calculates the scores of each candidate semantic term, automatically selecting true semantics that highly match the order status, regional scenario, logistics stage, and payment attributes. This effectively eliminates comprehension biases caused by polysemy, colloquial ambiguity, and language mixing in cross-border scenarios, significantly improving the accuracy and business relevance of semantic representation. Simultaneously, the semantic optimization results are deeply integrated with order information, providing high-purity, unambiguous semantic input for subsequent intent recognition and solution generation, greatly improving the robustness and reliability of cross-border customer service systems under multilingual, cross-regional, and strong business constraints.

[0017] Meanwhile, this invention employs a pooling method with logarithmically increasing temporal weights to perform global feature fusion on the semantically optimized word vector sequence. The weights increase smoothly without abrupt changes, focusing on the core business intent at the end of sentences. This adapts to the colloquial and short text characteristics of cross-border consultations, enhancing the ability to capture key information and resist interference. Furthermore, it calculates the matching degree between the global semantic vector of the consultation text and the standard features of scene intent in the knowledge base using cosine similarity, and combines this with adaptive correction of sequence length to obtain intent confidence, reducing noise interference from short texts and making the confidence distribution more closely match the true intent distribution. Simultaneously, relying on a semantic extraction model with a Transformer structure, it simultaneously optimizes temporal pooling parameters and feature extraction parameters under a unified training framework, further improving intent matching accuracy. Through confidence threshold filtering, it outputs a highly reliable set of standard intent categories, effectively avoiding interference from weakly related intents. This provides a reliable basis for subsequent business solution generation and compliance verification, significantly improving the intelligence level and processing accuracy of the cross-border customer service system. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a cross-border customer service method based on intelligent agents, as provided in an embodiment of the present invention.

[0019] Figure 2 This is a structural diagram of a cross-border customer service system provided in an embodiment of the present invention.

[0020] Figure 3 This is an experimental comparison diagram provided for one embodiment of the present invention. Detailed Implementation

[0021] The realization of the objectives, functional characteristics, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0022] This invention provides a cross-border customer service method and system based on intelligent agents. The executing entity of the intelligent agent-based cross-border customer service method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this invention: a server, a terminal, etc. In other words, the intelligent agent-based cross-border customer service method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0023] Reference Figure 1 Embodiment 1 of the present invention is as follows: A cross-border customer service method based on intelligent agents, the method comprising: S1: Obtain customer service consultation requests from cross-border users, standardize the customer service consultation requests to obtain the target language of customer service, the region involved in the consultation, and the standard text of customer service consultation, and use a cross-border multilingual ambiguity resolution algorithm to perform semantic optimization processing on the standard text of customer service consultation to obtain semantically optimized text of customer service consultation.

[0024] Specifically, step S1 involves obtaining customer service inquiries from cross-border users and standardizing those inquiries, including: S11: The customer service inquiry request from the cross-border user includes the cross-border user's IP address and the customer service inquiry text. The cross-border user's IP address is matched by region, and the matched region is used as the region involved in the inquiry. S12: Use a cross-border multilingual fast recognition algorithm to identify the language of the customer service consultation text, and use the identified language as the target language for customer service. As an embodiment of the present invention, the process of using a cross-border multilingual fast recognition algorithm to identify the language of the customer service inquiry text is as follows: S121: Load the preset cross-border common language library and algorithm recognition threshold. The cross-border common language library contains language feature vectors of various mainstream cross-border trade languages. The language feature vectors include character frequency features, grammatical structure features and cross-border specific vocabulary features. The algorithm recognition threshold is 0.85 by default. The character frequency features are the frequency of occurrence of each character in the mainstream cross-border trade languages; the grammatical structure features include the sentence structure of the mainstream cross-border trade languages ​​and commonly used prepositions / conjunctions; and the cross-border specific vocabulary features include the expression of high-frequency words in cross-border trade in the mainstream cross-border trade languages. S122: Extract text features from the customer service consultation text to obtain the text feature vector of the customer service consultation text, wherein the text feature vector includes text character frequency features, text grammatical structure features, and text cross-border specific vocabulary features. Specifically, by statistically analyzing the frequency of each character (including letters, punctuation, and special characters) in customer service consultation texts, text character frequency features are generated, with a focus on capturing exclusive high-frequency characters of each mainstream cross-border trade language (such as the Spanish "ñ" and the French "é"). Extract sentence structure (such as the order of subject, verb, and object) and common prepositions / conjunctions (such as "in" in English and "dans" in French) from customer service consultation texts to generate text grammatical structure features; Match the cross-border specific vocabulary features of the mainstream cross-border trade languages ​​in the cross-border common language feature library, such as the corresponding expressions of high-frequency words such as customs, tariff, and logistics in different mainstream cross-border trade languages, and extract the proportion of overlapping words between customer service consultation text and cross-border specific vocabulary features as the text cross-border specific vocabulary features. S123: Extract the proportion of garbled symbols and emoticons in the customer service consultation text. Based on the extracted proportion, perform language matching between the text feature vector of the customer service consultation text and the language feature vector of the mainstream cross-border trade language to obtain the text matching score of the mainstream cross-border trade language. Select the mainstream cross-border trade language with a text matching score higher than the algorithm recognition threshold as the candidate language. Specifically, the formula for calculating the text matching score of the mainstream cross-border trade languages ​​is as follows: ; in, This represents the text matching score for the Lth mainstream cross-border trade language, where Len represents the number of mainstream cross-border trade languages ​​in the cross-border common language database. Represents the frequency characteristics of text characters. This represents the character frequency characteristics of the Lth mainstream cross-border trade language. Representing the syntactic structural features of the text, This represents the grammatical structural features of the Lth mainstream cross-border trade language. For the function that calculates feature similarity, set This is the cosine similarity calculation function. This indicates the proportion of garbled characters and emoticons in the customer service inquiry text. This indicates the proportion of overlapping words between the customer service inquiry text and the cross-border specific vocabulary features of the Lth mainstream cross-border trade language. Indicates the cross-border vocabulary enhancement factor, the default setting. It is 1.2. This represents the matching weight coefficient, set to the default value. It is 0.55; S124: Obtain the official language of the region involved in the consultation, and select the candidate language that overlaps with the official language and has the highest text matching score as the target language for customer service; Optionally, if no candidate language is selected, the language with the highest text matching score from the official languages ​​is selected as the target language for customer service. S13: Perform text preprocessing on the customer service consultation text to obtain the word sequence corresponding to the customer service consultation text, which serves as the standard text for customer service consultation. The text preprocessing includes word segmentation, stop word removal, business terminology standardization, and order association. Specifically, the word segmentation tool is the jieba word segmentation tool. Stop words include modal words, conjunctions, and redundant function words without business meaning. The business terminology is uniformly processed by using a cross-border customer service terminology mapping library to uniformly replace synonymous expressions, colloquial expressions, and multilingual mixed expressions with standard business terms. The order number is extracted from the customer service consultation text, and the order number is matched with the order in the cross-border business system to obtain the associated order information. The word sequence processed by the business terminology is concatenated with the associated order information (with the word sequence first and the associated order information second) to form the standard customer service consultation text. The standard customer service consultation text includes a sequence of phrases with standardized business terminology and associated order information, including order identifier, payment information, delivery area, delivery deadline, and current logistics status. As an embodiment of the present invention, taking the customer service inquiry text "Hola pedido 7K91A2 si pago hoy elimpuesto ¿pueden interceptar el paquete y reenviarlo antes del viernes? Noquiero pagar doble envío." (Hello, order 7K91A2, if I pay the tax today, can you intercept the package and resend it before Friday? I don't want to pay double the shipping fee.) as an example, after removing the stop words: Hola, si, el, pueden, el, y, No, quiero, and standardizing them according to the cross-border terminology database: pedido → order, impuesto → tax, interceptar → logistics interception, paquete → package, reenviarlo → resend, envío → shipping fee, doble → The second time, the corresponding business terminology sequence is uniformly processed as follows: (order, 7K91A2, payment, today, tax, logistics interception, package, re-delivery, before Friday, payment, second time, shipping fee). The order number 7K91A2 is extracted and associated with the order information: order identifier (MLMX20260310_7K91A2), payment information (paid), delivery area (Mexico City), delivery deadline (before January 10, 2026), and current logistics status (in customs clearance). S14: The target language of customer service, the region involved in the consultation, and the standard text of customer service consultation are used as the standardized processing result of the customer service consultation request.

[0025] It should be noted that this invention accurately identifies the region involved in the consultation based on the user's IP address, avoiding misjudgment of the region; secondly, it achieves highly robust multilingual recognition by matching three-dimensional features of character frequency, grammatical structure, and cross-border specific vocabulary, combined with adaptive adjustment of the scrambled character ratio, ensuring the accuracy of the target language; finally, through word segmentation, noise reduction, terminology standardization, and order association, the original consultation text is transformed into standardized text containing standard business terms and complete order information. The entire process is closed-loop and controllable, effectively eliminating multilingual ambiguity, colloquial interference, and cross-regional rule differences, laying a solid data foundation for subsequent accurate intent recognition and compliance solution generation, and significantly improving the response speed and processing accuracy of the cross-border customer service system.

[0026] Step S1, based on the target language of customer service, employs a cross-border multilingual ambiguity resolution algorithm to perform semantic optimization processing on the standard customer service consultation text, and also includes: S15: Perform word vector representation on the standard customer service consultation text to obtain the word vector sequence of the word group sequence in the standard customer service consultation text; Specifically, a pre-trained model (such as the BERT model or the Word2Vec model) is used to represent word vectors in the standard customer service consultation text; S16: Traverse the word vector sequence to obtain word vectors with multiple semantic meanings in the target language of customer service, and generate business coding features of the word vectors in different semantic meanings; As an embodiment of the present invention, the business coding features include regional adaptation coding, logistics status adaptation coding, payment attribute coding, and time attribute coding; The region adaptation code is a numerical form between 0 and 4, representing the type of country / region to which the semantic meaning is usually applicable, where 0 indicates no region involved, 1 indicates the Americas, 2 indicates Europe, 3 indicates Asia, and 4 indicates global applicability. The logistics status adaptation code is a numerical form between 0 and 4, representing the logistics stage to which the semantic meaning usually applies. 0 indicates no logistics meaning, 1 indicates pending payment / pending shipment, 2 indicates customs clearance / pending tax payment, 3 indicates transportation, and 4 indicates delivery / receipt / rejection. The payment attribute is encoded in numerical form between 0 and 4, representing the type of fee to which the semantic meaning usually applies, where 0 indicates no fee meaning, 1 indicates paid, 2 indicates unpaid / pending payment, 3 indicates tax-related, and 4 indicates freight-related. The time attribute is encoded in a numerical form between 0 and 3, representing the time type to which the semantic meaning usually applies. 0 indicates no time meaning, 1 indicates a point in time (e.g., today, tomorrow), 2 indicates a time limit (e.g., deadline, expiration), and 3 indicates a time-limited commitment. S17: Extract the business context features from the word vector sequence, and calculate the semantic meaning score between the business encoding features and the business context features of different semantic meanings; Specifically, word vectors corresponding to the associated order information are extracted from the word vector sequence, and business encoding features of the extracted word vectors are generated as business context features in the word vector sequence; The formula for calculating the semantic meaning score between the business coding features and business context features of different semantic meanings is as follows: ; ; in, Business coding features representing semantic meaning e Business context features Semantic meaning scores between them , These represent the business coding features respectively. The code includes region adaptation code, logistics status adaptation code, payment attribute code, and time attribute code. , These represent the business context features in turn. The code includes region adaptation code, logistics status adaptation code, payment attribute code, and time attribute code. A function for calculating feature similarity. Indicates business coding characteristics Business context features The mismatch penalty parameter between them Represents the penalty function, for ,like If they are not equal, then =1, otherwise =0, All represent penalty weights, default settings. The values ​​are 0.2, 0.3, 0.2, and 0.3 respectively. This represents the penalty coefficient, the default setting. It is 0.6; S18: Select the semantic sense with the highest semantic sense score as the true semantic sense of the word vector, and represent the true semantic sense with word vectors. Replace the word vectors with multiple semantic senses in step S16 with the word vector representation of the true semantic sense to obtain the semantically optimized word vector sequence of the customer service consultation standard text. S19: The semantically optimized word vector sequence, the target language of customer service, the regions involved in the consultation, and the standard text of customer service consultation are used as the semantically optimized text of customer service consultation.

[0027] S2: An intent recognition algorithm combining a cross-border customer service scenario knowledge base is used to identify the consultation intent of the semantically optimized text of the customer service consultation, and to generate consultation intent recognition features.

[0028] Specifically, step S2 employs an intent recognition algorithm that combines a cross-border customer service scenario knowledge base to identify the consultation intent of the semantically optimized customer service consultation text, including: S21: Extract the semantically optimized word vector sequence from the semantically optimized text of the customer service inquiry; S22: Perform temporal pooling on the semantically optimized word vector sequence to obtain the global semantic vector of the consultation text; Specifically, the temporal pooling formula for the semantically optimized word vector sequence is as follows: ; ; in, Let N represent the nth word vector in the semantically optimized word vector sequence, where N represents the sequence length of the semantically optimized word vector sequence. Word vectors Temporal weights, This represents the global semantic vector of the consultation text. Represents the logarithmic function. This represents the trainable temporal pooling weight parameters. This represents the trainable temporal pooling bias parameters. This indicates the activation function, with the Sigmoid function selected by default. Specifically, for this temporal weight calculation formula, the word vectors at the beginning of the word vector sequence after semantic optimization are given a basic weight, and the weight is smoothly increased as the word vector position moves to the end. This avoids the weight abrupt changes caused by linear weighting, improves the stability and robustness of temporal enhancement, and allows word vectors closer to the end of the sentence to receive higher weights. This aligns with the user's expression habit in cross-border customer service scenarios of placing the core intent and business actions in the latter half of the sentence, thereby strengthening and focusing on the core semantic position. It can effectively highlight the core business semantics in the latter half of the text, enhance the focusing ability and anti-interference ability of intent recognition, and adapt to the semantic extraction needs of short texts and conversational consultations in cross-border customer service.

[0029] S23: Calculate the similarity and intent confidence between the global semantic vector of the consultation text and the standard features of the scenario intent corresponding to the standard intent category in the cross-border customer service scenario knowledge base, wherein the formulas for calculating the similarity and intent confidence are as follows: ; ; in, Represents the global semantic vector of the consultation text. With the dth standard intent category in the cross-border customer service scenario knowledge base Corresponding scene intent standard features Similarity between them Let L represent the L2 norm, and D represent the number of standard intent categories in the cross-border customer service scenario knowledge base. Represents the global semantic vector of the consultation text. With the dth standard intent category in the cross-border customer service scenario knowledge base Confidence of intent between them Indicates the selection of a set The maximum value in, where N represents the sequence length of the semantically optimized word vector sequence. Represents a logarithmic function; Specifically, the standard intent categories include, but are not limited to, logistics inquiry, tax payment, package interception, re-delivery, refusal to accept and return, refund application, customs clearance progress, timeliness commitment, liability attribution, and cost bearing; by constructing intent description text for standard intent categories and using a semantic extraction model to extract the semantic features of the intent description text as the scenario intent standard features corresponding to the standard intent categories, the scenario intent standard features and standard intent categories are constructed into a cross-border customer service scenario knowledge base; Furthermore, the semantic extraction model is a Transformer model structure, and obtains global semantic vectors of consultation text under multiple standard intent categories. With the goal of maximizing the similarity between the global semantic vectors of consultation text and the scene intent standard features of the corresponding standard intent categories, the model parameters, temporal pooling weight parameters, and temporal pooling bias parameters in the semantic extraction model are optimized and trained. Optionally, the parameter optimization training method adopts gradient descent algorithm or Adam optimizer. S24: Retain standard intent categories with intent confidence levels higher than a preset confidence threshold, and use the retained standard intent categories and their corresponding intent confidence levels as features for identifying consultation intent.

[0030] Specifically, the preset confidence threshold is set to 0.7 by default.

[0031] S3: Based on the optimized text of the customer service consultation semantics and the consultation intent recognition features, use a graph neural network to calculate the matching degree between the consultation intent recognition features and the business nodes, and generate a business solution corresponding to the consultation intent recognition features.

[0032] Specifically, step S3, which utilizes a graph neural network to calculate the matching degree between consultation intent recognition features and business nodes, includes: S31: The business nodes include logistics query nodes, logistics modification nodes, payment nodes, and after-sales nodes, and a graph association matrix between the standard intent category and the business nodes in the consultation intent recognition features is constructed. Specifically, if there is a business association between a business node and a standard intent category, the corresponding position in the graph association matrix is ​​set to 1; otherwise, it is set to 0. S32: Combining the graph association matrix, a graph neural network is used to perform graph convolution calculation on the intent confidence corresponding to the standard intent category in the consultation intent recognition features to obtain the matching association degree between the standard intent category and the business node; Specifically, the graph convolution calculation formula is as follows: ; in, This represents the degree of matching correlation between the standard intent category m and the j-th business node. The first to fourth business nodes are, respectively, the logistics query node, the logistics modification node, the payment node, and the after-sales node. This represents the element values ​​of the standard intent category m and the j-th service node in the graph association matrix. This represents the node degree of the standard intent category m in the graph association matrix. This represents the degree of the j-th business node in the graph association matrix. This represents the intent confidence level for standard intent category m; S33: For any standard intent category in the consultation intent recognition features, calculate the multi-dimensional constraint score of the standard intent category, select the business node with the highest matching correlation, and calculate the execution score of the standard intent category based on the fixed business weight of the business node and the matching correlation between the standard intent category and the business node. As an embodiment of the present invention, associated order information is obtained, wherein the associated order information includes order identifier, payment information, delivery area, delivery deadline, and current logistics status. Based on the associated order information, a multi-dimensional constraint score for the standard intent category is generated, wherein the multi-dimensional constraint score includes a payment constraint score, a logistics constraint score, and a time limit constraint score. The calculation formula for the multi-dimensional constraint score is as follows: ; ; in, The multidimensional constraint score represents the standard intent category m. This represents the payment constraint score for standard intent category m. This represents the logistic constraint score for standard intention category m. This represents the time constraint score for standard intent category m, where a higher constraint score indicates a higher feasibility of the order in the corresponding dimension. Payment constraint scores and logistics constraint scores for standard intent categories are generated based on preset value selection rules, which include: The standard intent category is tax payment. If the payment information is unpaid taxes, the payment constraint score is 1. The standard intent category is tax payment. If the payment information indicates that the tax has been paid, the payment constraint score is 0. The standard intent category is refund; if the payment information indicates that payment has been made, the payment constraint score is 1. The default score for payment constraint in other standard intent categories and payment information is 0.5; The standard intent category is package interception. If the current logistics status is pending shipment, the logistics constraint score is 1. The standard intent category is package interception. If the current logistics status is in customs clearance / delivery, the logistics constraint score is 0. The standard intent category is to query logistics. If the current logistics status is any logistics status, the logistics constraint score is 1. The default score for logistics constraints, including other standard intent categories and the current logistics status, is 0.5. Based on the matching relevance, the fixed business weight of the business node, and the multi-dimensional constraint score of the standard intent category, the execution score of the standard intent category is calculated, wherein the formula for calculating the execution score is: ; in, This represents the execution score for standard intent category m. This indicates the degree of matching and correlation between the standard intent category m and the selected business node. This represents the score of multi-dimensional constraints based on fixed business weights of business nodes. The weighted result This represents the correlation control coefficient, set to the default value. It is 0.5; Furthermore, the default weights for the payment constraint score, logistics constraint score, and time limit constraint score of the logistics query node are set to 0.15, 0.55, and 0.3, respectively; the default weights for the payment constraint score, logistics constraint score, and time limit constraint score of the logistics modification node are set to 0.2, 0.5, and 0.3, respectively; the default weights for the payment constraint score, logistics constraint score, and time limit constraint score of the payment node are set to 0.6, 0.2, and 0.2, respectively; and the default weights for the after-sales node are set to 0.25, 0.4, and 0.35, respectively. S34: Based on the execution score, generate the execution status of the standard intent category, wherein the execution status of the standard intent category includes full execution, partial execution, and failure to execute.

[0033] Specifically, the higher the execution score, the stronger the feasibility of the standard intent category in the current order's actual state, and the more suitable it is to provide the corresponding business solution to the user. If the execution score is greater than or equal to 0.75, the execution status is fully executed; if the execution score is greater than or equal to 0.45 but less than 0.75, the execution status is partially executed; if the execution score is less than 0.45, the execution status is unexecuted.

[0034] It should be noted that this invention constructs a graph association matrix that links standard intent categories with business nodes, and uses graph convolution to achieve accurate matching association calculation. Based on real order information, it generates three-dimensional constraint scores for payment, logistics, and time limits, employing differentiated weighting for different business nodes to make the execution score more closely reflect the realities of cross-border business. The execution score comprehensively reflects the intent matching degree and order executability, clearly defining three states—fully executed, partially executed, and unexecuted—through clear thresholds. This enables a quantitative assessment of fulfillment capabilities, effectively avoiding unreasonable promises and responses that are detached from the actual order situation, improving the authenticity and reliability of business solutions, and providing quantifiable and traceable execution evidence for subsequent compliance verification and customer service responses. This enhances the stability and accuracy of the cross-border system in complex cross-border scenarios.

[0035] Step S3, which generates a business solution corresponding to the consultation intent identification features based on the execution status of the standard intent category, also includes: S35: Based on the execution status of the standard intent category and the order association processing result in the standard customer service consultation text, generate the current progress, current business status, and business status basis of the standard intent category; Specifically, the current progress refers to the current stage of the logistics status (awaiting shipment / in customs clearance / in delivery / signed for), and the current business status is a natural language description of the execution status, which is generated based on the combination of related order information; S36: The execution status, current progress, current business status, and business status of the standard intent category are used as the solution for the standard intent category; As an embodiment of the present invention, taking two standard intent categories, tax payment and package interception, as examples, with associated order information of Mexico, pending customs clearance, unpaid taxes, and closed cutoff window, the execution scores are: tax payment 0.82 (fully executed), package interception 0.28 (cannot be executed). The corresponding solutions are as follows: Tax payment: Current progress: Customs clearance pending tax payment | Current business status: Taxes not paid | Business status based on: Mexican customs clearance rules; Package Interception: Current Progress: Order Cut-off Window Closed | Current Business Status: Cannot Be Intercepted | Business Status Basis: Warehouse Order Cut-off Time Limit Has Expired; S37: Take the solutions for all standard intent categories in the consultation intent identification features as the business solutions corresponding to the consultation intent identification features.

[0036] S4: Conduct compliance verification of the business solution based on the laws and regulations of the regions involved in the consultation and the platform rules, and generate customer service content corresponding to the business solution in conjunction with cross-border customer service service specifications.

[0037] Specifically, step S4 involves compliance verification of the business solution based on the laws and regulations of the region where the consultation is conducted, as well as the platform rules, including: The compliance verification includes verification of regional regulations, platform rules, time limits, and liability and fee compliance. The regional regulatory compliance verification is used to verify whether the current business status in the business solution and the basis for the business status comply with the relevant legal provisions of the consulting region regarding consumer protection, tariffs, customs clearance, returns, and refusal of acceptance. The platform rule compliance verification is used to verify whether the current business status in the business solution and the basis for the business status comply with the platform's constraints on order cut-off, re-delivery, taxes, freight, refunds, and timeliness commitments. The time limit compliance check is used to verify whether the current time is within the allowable range of the receiving deadline, warehouse order cut-off window, and customs clearance time. The liability and cost compliance verification is used to verify whether the attribution of liability and the allocation of costs (taxes, freight) comply with the cost allocation rules of the region and platform involved in the consultation. The verification outputs for the regional regulations compliance verification, platform rules compliance verification, time limit compliance verification, and liability and fee compliance verification are 0 or 1, where 1 indicates that the verification has passed and 0 indicates that the verification has failed. The verification outputs of the regional regulatory compliance verification, platform rule compliance verification, time limit compliance verification, and liability and cost compliance verification are weighted. If the weighted result is higher than the preset verification threshold, it means that the business solution has passed the compliance verification.

[0038] Specifically, the weighting coefficients for the verification outputs of regional regulatory compliance verification, platform rule compliance verification, time limit compliance verification, and liability and fee compliance verification are set to 0.3, 0.3, 0.2, and 0.2 respectively by default; the default verification threshold is set to 0.75.

[0039] If the business solution passes compliance verification in step S4, the corresponding customer service content will be generated based on the cross-border customer service service specifications, including: The customer service content includes simplified expressions of the solutions for each standard intent category in the order identification and business solutions. The expression of the text is adjusted according to the target language of customer service, the region involved in the consultation, and the compliance verification process. The expressed text is then translated into the target language of customer service and used as the customer service content.

[0040] As an embodiment of the present invention, taking order identifier MLMX20260310_7K91A2 and two standard intent categories, namely tax payment and package interception, as examples, the Chinese description of the generated customer service content is as follows: "Dear customer, thank you for your inquiry. Your order number is MLMX20260310_7K91A2. The processing status of your request is as follows:" Package interception request: Your package has already left the warehouse and entered the customs clearance process. The order interception window is closed, and package interception is not possible. This response complies with Mexican customs regulations and the platform's order interception rules.

[0041] Tax Payment Request: Your package's customs clearance process is temporarily stalled due to unpaid taxes. You can complete the tax payment in your personal account management center. After payment, the package will resume customs clearance and delivery. This response complies with Mexican import tariff regulations and the platform's tax payment rules.

[0042] If you have any further questions, please feel free to contact us. Example 2: As another embodiment of the present invention, this embodiment provides a cross-border customer service system based on intelligent agents. The cross-border customer service system consists of multiple intelligent agents, including a cross-border consultation preprocessing intelligent agent, a cross-border intent recognition intelligent agent, a business solution generation intelligent agent, and a compliance verification and customer service content generation intelligent agent, to implement the technical steps of the cross-border customer service method based on intelligent agents as described in Example 1; refer to... Figure 2 The diagram shows the structure of the cross-border customer service system. Each intelligent agent works collaboratively in a progressive order. The output of the previous intelligent agent serves as the input of the next intelligent agent. They share core basic data (such as associated order information, cross-border customer service scenario knowledge base, platform rules, etc.) and respectively complete the functions of consultation preprocessing, intent recognition, solution generation, compliance verification, and customer service content generation, jointly realizing the intelligent closed-loop processing of cross-border customer service consultation.

[0043] Specifically, the consultation preprocessing agent is used to obtain customer service consultation requests from cross-border users, standardize the customer service consultation requests to obtain the target language of customer service, the region involved in the consultation, and the standard text of customer service consultation. Based on the target language of customer service, a cross-border multilingual ambiguity resolution algorithm is used to perform semantic optimization processing on the standard text of customer service consultation to obtain semantically optimized text of customer service consultation. The cross-border intent recognition agent uses an intent recognition algorithm that combines a cross-border customer service scenario knowledge base to recognize the consultation intent of the semantically optimized text of customer service consultation and generate consultation intent recognition features. The business solution generating agent optimizes the text based on the semantics of customer service inquiries and the features of inquiry intent recognition. It then uses a graph neural network to calculate the degree of matching between the features of inquiry intent recognition and business nodes, and generates a business solution corresponding to the features of inquiry intent recognition. The compliance verification and customer service content generation intelligent agent performs compliance verification on the business solution based on the laws and regulations of the regions involved in the consultation and the platform rules, and generates the corresponding customer service content for the business solution in conjunction with the cross-border customer service service specifications.

[0044] Example 3: This invention collects 5,000 real cross-border e-commerce customer service inquiries (covering Mexico, Spain, the United States, etc.), involving inquiries such as logistics tracking, package interception, tax payment, refunds, customs clearance, and reshipment. It compares the performance of the cross-border customer service system described in this invention with that of a traditional customer service system (based on keyword + manual rule matching) in terms of inquiry processing efficiency, intent recognition accuracy, commitment compliance rate, and user satisfaction. (Refer to...) Figure 3 The experimental comparison chart shows that the cross-border customer service intelligent system described in this invention, through multilingual ambiguity resolution, accurate intent recognition, graph neural network business reasoning, order quantification constraints, and compliance verification mechanisms, achieves a 22.4% improvement in intent recognition accuracy, a 29.6% improvement in response compliance rate, an 82.0% reduction in average response time, and a 21.3% improvement in user satisfaction compared to traditional customer service systems. It has significant technical advantages and practical value in cross-border multilingual, multi-regional, and multi-business scenarios.

[0045] It should be noted that the terms "comprising," "including," or any other variations thereof used herein are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0046] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0047] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A cross-border customer service method based on intelligent agents, characterized in that, The method includes: S1: Obtain customer service consultation requests from cross-border users, standardize the customer service consultation requests to obtain the target language of customer service, the region involved in the consultation, and the standard text of customer service consultation, and use a cross-border multilingual ambiguity resolution algorithm to perform semantic optimization processing on the standard text of customer service consultation to obtain semantically optimized text of customer service consultation. S2: An intent recognition algorithm combining a cross-border customer service scenario knowledge base is used to identify the consultation intent of the semantically optimized text of the customer service consultation, and to generate consultation intent recognition features; S3: Based on the optimized text of the customer service consultation semantics and the consultation intent recognition features, use a graph neural network to calculate the matching degree between the consultation intent recognition features and the business nodes, and generate a business solution corresponding to the consultation intent recognition features; S4: Conduct compliance verification of the business solution based on the laws and regulations of the regions involved in the consultation and the platform rules, and generate customer service content corresponding to the business solution in conjunction with cross-border customer service service specifications.

2. The cross-border customer service method based on intelligent agents as described in claim 1, characterized in that, Step S1 involves obtaining customer service inquiries from cross-border users and standardizing these inquiries, including: S11: The customer service inquiry request from the cross-border user includes the cross-border user's IP address and the customer service inquiry text. The cross-border user's IP address is matched by region, and the matched region is used as the region involved in the inquiry. S12: Use a cross-border multilingual fast recognition algorithm to identify the language of the customer service consultation text, and use the identified language as the target language for customer service. S13: Perform text preprocessing on the customer service consultation text to obtain the word sequence corresponding to the customer service consultation text, which serves as the standard text for customer service consultation. The text preprocessing includes word segmentation, stop word removal, business terminology standardization, and order association. S14: The target language of customer service, the region involved in the consultation, and the standard text of customer service consultation are used as the standardized processing result of the customer service consultation request.

3. The cross-border customer service method based on intelligent agents as described in claim 2, characterized in that, Step S1, based on the target language of customer service, employs a cross-border multilingual ambiguity resolution algorithm to perform semantic optimization processing on the standard customer service consultation text, and also includes: S15: Perform word vector representation on the standard customer service consultation text to obtain the word vector sequence of the word group sequence in the standard customer service consultation text; S16: Traverse the word vector sequence to obtain word vectors with multiple semantic meanings in the target language of customer service, and generate business coding features of the word vectors in different semantic meanings; S17: Extract the business context features from the word vector sequence, and calculate the semantic meaning score between the business encoding features and the business context features of different semantic meanings; S18: Select the semantic sense with the highest semantic sense score as the true semantic sense of the word vector, and represent the true semantic sense with word vectors. Replace the word vectors with multiple semantic senses in step S16 with the word vector representation of the true semantic sense to obtain the semantically optimized word vector sequence of the customer service consultation standard text. S19: The semantically optimized word vector sequence, the target language of customer service, the regions involved in the consultation, and the standard text of customer service consultation are used as the semantically optimized text of customer service consultation.

4. The cross-border customer service method based on intelligent agents as described in claim 1, characterized in that, In step S2, an intent recognition algorithm combining a cross-border customer service scenario knowledge base is used to identify the consultation intent of the semantically optimized customer service consultation text, including: S21: Extract the semantically optimized word vector sequence from the semantically optimized text of the customer service inquiry; S22: Perform temporal pooling on the semantically optimized word vector sequence to obtain the global semantic vector of the consultation text; S23: Calculate the similarity and intent confidence between the global semantic vector of the consultation text and the standard features of the scenario intent corresponding to the standard intent category in the cross-border customer service scenario knowledge base, wherein the formulas for calculating the similarity and intent confidence are as follows: ； ； in, Represents the global semantic vector of the consultation text. With the dth standard intent category in the cross-border customer service scenario knowledge base Corresponding scene intent standard features Similarity between them Let L represent the L2 norm, and D represent the number of standard intent categories in the cross-border customer service scenario knowledge base. Represents the global semantic vector of the consultation text. With the dth standard intent category in the cross-border customer service scenario knowledge base Confidence of intent between them Indicates the selection of a set The maximum value in, where N represents the sequence length of the semantically optimized word vector sequence. Represents a logarithmic function; S24: Retain standard intent categories with intent confidence levels higher than a preset confidence threshold, and use the retained standard intent categories and their corresponding intent confidence levels as features for identifying consultation intent.

5. The cross-border customer service method based on intelligent agents as described in claim 1, characterized in that, Step S3 utilizes a graph neural network to calculate the matching degree between the consultation intent recognition features and the business nodes, including: S31: The business nodes include logistics query nodes, logistics modification nodes, payment nodes, and after-sales nodes, and a graph association matrix between the standard intent category and the business nodes in the consultation intent recognition features is constructed. S32: Combining the graph association matrix, a graph neural network is used to perform graph convolution calculation on the intent confidence corresponding to the standard intent category in the consultation intent recognition features to obtain the matching association degree between the standard intent category and the business node; S33: For any standard intent category in the consultation intent recognition features, calculate the multi-dimensional constraint score of the standard intent category, select the business node with the highest matching correlation, and calculate the execution score of the standard intent category based on the fixed business weight of the business node and the matching correlation between the standard intent category and the business node. S34: Based on the execution score, generate the execution status of the standard intent category, wherein the execution status of the standard intent category includes full execution, partial execution, and failure to execute.

6. The cross-border customer service method based on intelligent agents as described in claim 5, characterized in that, Step S3, which generates a business solution corresponding to the consultation intent identification features based on the execution status of the standard intent category, also includes: S35: Based on the execution status of the standard intent category and the order association processing result in the standard customer service consultation text, generate the current progress, current business status, and business status basis of the standard intent category; S36: The execution status, current progress, current business status, and business status of the standard intent category are used as the solution for the standard intent category; S37: Take the solutions for all standard intent categories in the consultation intent identification features as the business solutions corresponding to the consultation intent identification features.

7. The cross-border customer service method based on intelligent agents as described in claim 1, characterized in that, Step S4 involves compliance verification of the business solution based on the laws and regulations of the regions involved in the consultation and the platform rules, including: The compliance verification includes verification of regional regulations, platform rules, time limits, and liability and fee compliance. The verification outputs for the regional regulations compliance verification, platform rules compliance verification, time limit compliance verification, and liability and fee compliance verification are 0 or 1, where 1 indicates that the verification has passed and 0 indicates that the verification has failed. The verification outputs of the regional regulatory compliance verification, platform rule compliance verification, time limit compliance verification, and liability and cost compliance verification are weighted. If the weighted result is higher than the preset verification threshold, it means that the business solution has passed the compliance verification.

8. The cross-border customer service method based on intelligent agents as described in claim 7, characterized in that, If the business solution passes compliance verification in step S4, the corresponding customer service content will be generated based on the cross-border customer service service specifications, including: The customer service content includes simplified expressions of the solutions for each standard intent category in the order identification and business solutions. The expression of the text is adjusted according to the target language of customer service, the region involved in the consultation, and the compliance verification process. The expressed text is then translated into the target language of customer service and used as the customer service content.

9. A cross-border customer service system based on intelligent agents, characterized in that, The cross-border customer service system consists of multiple intelligent agents, including a cross-border consultation preprocessing intelligent agent, a cross-border intent recognition intelligent agent, a business solution generation intelligent agent, and a compliance verification and customer service content generation intelligent agent, to implement the technical steps of the cross-border customer service method based on intelligent agents as described in any one of claims 1-8.

Images