E-commerce customer service response method and device, equipment and medium

Abstract

The invention relates to an e-commerce customer service response method and device, equipment and medium in the technical field of e-commerce, and the method comprises the steps: responding to a customer service consultation event triggered by a user, obtaining a target consultation text corresponding to the event, and determining the matching results between the target consultation text and preset dictionaries of a plurality of consultation categories; when the matching result represents that the dictionary is hit, determining a target consultation category to which the target consultation text belongs in a plurality of consultation categories according to the matching result; when all the matching results represent that the dictionary is missed, determining a target consultation category to which the target consultation text belongs in a plurality of consultation categories by adopting a preset intention recognition model; and from a preset question and answer knowledge base of the target consultation category, obtaining a customer service response text corresponding to the matched target consultation text, and pushing the customer service response text to the user. The user consultation can be efficiently and accurately responded, and the consultation customer service satisfaction degree is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of e-commerce, in particular to an e-commerce customer service response method and a corresponding device, computer equipment and computer readable storage medium. BACKGROUND

[0002] In the field of e-commerce customer service question answering, semantic matching models are currently widely used to process user questions. The specific way is to perform semantic matching between the user's input question and the questions stored in the knowledge base, and output the most relevant corresponding answer in the library according to the matching. This approach can to some extent realize the association between questions and answers, but in actual application, it exposes a key problem, that is, it is difficult to accurately distinguish between question types that are semantically similar but essentially different.

[0003] For example, when the user asks "What is the meaning of binding an independent domain name?", which is a question in the definition category, the system may mistakenly match it with "How to bind an independent domain name", which involves the specific operation process. Although there is some relevance from the semantic level, the actual needs of the users they represent and the types of corresponding answers are essentially different.

[0004] The former user's intention is to understand the concept of "binding an independent domain name", which requires a clear and explicit definition explanation, such as explaining that binding an independent domain name means associating an independent domain name with an e-commerce store or other network services, so that users can access the corresponding service through the domain name, and focusing on concept explanation.

[0005] The latter is the user's expectation to obtain specific operation step guidance, such as first entering the domain name service provider background, adding domain name resolution record, selecting corresponding host record, record type and resolution line, and then binding the resolved domain name with the e-commerce store background and a series of operation processes, the answer content is completely different, and the functions cannot be replaced.

[0006] Such erroneous matching phenomena are common in actual application and have serious consequences. On the one hand, it greatly affects the user's experience, so that the user cannot quickly and accurately obtain the information he really needs, which may lead to the user's dissatisfaction with the customer service system, reducing the user's favorability and trust of the e-commerce platform; on the other hand, it also makes it difficult to effectively guarantee the accuracy of the customer service system's reply, and cannot fully play its function of efficiently answering user questions and improving service efficiency, greatly limiting the role of the e-commerce customer service question answering system in improving service quality, optimizing user experience and assisting e-commerce business development, and there is an urgent need for a technical solution to effectively solve such problems to improve and optimize. SUMMARY

[0007] The primary purpose of the present application is to solve at least one of the above problems to provide an e-commerce customer service response method and its corresponding device, computer equipment, computer program product.

[0008] To meet the various purposes of the present application, the present application adopts the following technical solutions:

[0009] An e-commerce customer service response method provided to adapt to one of the purposes of the present application, comprising the following steps:

[0010] In response to a customer service consultation event triggered by a user, a target consultation text corresponding to the event is obtained, and the matching results between the target consultation text and a plurality of consultation category preset dictionaries are determined respectively;

[0011] When the matching result represents a dictionary hit, the target consultation category to which the target consultation text belongs in the plurality of consultation categories is determined according to the matching result;

[0012] When all the matching results represent a dictionary miss, a preset intent recognition model is used to determine the target consultation category to which the target consultation text belongs in the plurality of consultation categories;

[0013] From the preset question and answer knowledge base of the target consultation category, a customer service response text corresponding to the target consultation text is obtained and pushed to the user.

[0014] On the other hand, an e-commerce customer service response device provided to adapt to one of the purposes of the present application comprises an event response module, a first category determination module, a second category determination module, and a user response module. The event response module is used to respond to a customer service consultation event triggered by a user, obtain a target consultation text corresponding to the event, and determine the matching results between the target consultation text and a plurality of consultation category preset dictionaries respectively. The first category determination module is used to determine the target consultation category to which the target consultation text belongs in the plurality of consultation categories according to the matching result when the matching result represents a dictionary hit. The second category determination module is used to determine the target consultation category to which the target consultation text belongs in the plurality of consultation categories using a preset intent recognition model when all the matching results represent a dictionary miss. The user response module is used to obtain a customer service response text corresponding to the target consultation text from the preset question and answer knowledge base of the target consultation category and push it to the user.

[0015] In another aspect, a computer device provided to adapt to one of the purposes of the present application comprises a central processing unit and a memory. The central processing unit is used to call and run a computer program stored in the memory to execute the steps of the e-commerce customer service response method described in the present application.

[0016] In yet another aspect, a computer program product, adapted to another object of the present application, comprises computer programs / instructions which, when executed by a processor, implement the steps of the method described in any of the embodiments of the present application.

[0017] The technical solution of the present application has multiple advantages, including but not limited to the following aspects:

[0018] The present application first responds to the event triggered by the user consulting the customer service, matches the corresponding target consultation text with the preset dictionary of multiple consultation categories respectively to obtain the result, and when the dictionary hits the matching success, the target consultation category can be accurately determined at a speed close to no delay. For the common and standardized problems in the e-commerce customer service field, rule matching shows excellent efficiency, which can realize instant response and ensure that the user obtains the required answer in the shortest time.

[0019] Secondly, the problem of lack of flexibility caused by simply relying on rule matching is avoided. The complex or ambiguous target consultation text provided by serving the user's diverse expression habits enables the intent recognition model to use its semantic ability and generalization ability learned to deeply analyze the user's real consultation intent hidden in complex or unstructured text, so that the target consultation category can still be accurately determined in the blank area that cannot be covered by rules. In this way, rule matching first and model recognition bottom-up, the dual-track classification strategy of organic combination, builds a complete and highly robust automatic classification system, whether it is a high-frequency simple routine problem or a low-frequency complex special problem, the corresponding consultation category can be accurately identified, greatly improving the accuracy and reliability of classification.

[0020] Furthermore, on the basis of accurately positioning the target consultation category, the matching efficiency and quality of the response content are further optimized. By subdividing the question and answer knowledge base, each consultation category corresponds to a dedicated knowledge base, and these knowledge bases are closely focused on solving the user's problems in this category. When searching and matching customer service response texts, only accurate search is needed within the highly relevant knowledge subset, which greatly improves the search efficiency and significantly enhances the pertinence and accuracy of the final pushed response content. This effectively avoids the noise interference and low efficiency problems that may occur when searching in a traditional large-scale unified knowledge base, ensures that the response information received by the user is accurate and practical, and thus greatly optimizes the customer service experience, making the user feel intimate and professional service.

[0021] In summary, the customer service consulting service based on the consulting category is realized, the double-track classification strategy of combining the priority matching dictionary and the intention recognition model as a complementary strategy is provided, the directional retrieval of the knowledge base organized by the category is combined, and the overall performance of the e-commerce customer service is comprehensively improved. The peak demand for the operation resources is reduced, the response speed is significantly improved, the accurate understanding and efficient processing capacity for complex consultation are enhanced, the response accuracy and efficiency are improved, and the user experience is greatly improved. These comprehensive advantages have an important role that cannot be ignored for improving the operation efficiency of the e-commerce platform, enhancing the user satisfaction, and shaping a good service image, and promote the e-commerce industry to a higher level in the key link of customer service. BRIEF DESCRIPTION OF DRAWINGS

[0022] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the following drawings, in which:

[0023] Figure 1 a network architecture of an exemplary e-commerce platform of the present application;

[0024] Figure 2 a flowchart of a typical embodiment of the e-commerce customer service response method of the present application;

[0025] Figure 3 a principle block diagram of the e-commerce customer service response device of the present application;

[0026] Figure 4 a structural schematic diagram of a computer device used in the present application. DETAILED DESCRIPTION

[0027] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the drawings, in which the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0028] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" (or comprise), "comprises" (or comprises) and "comprising" (or comprising), when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is further understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can be present. Furthermore, "connected" or "coupled" as used herein can include wirelessly connected or wirelessly coupled. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0029] Those skilled in the art will appreciate that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0030] As shown in the network architecture, Figure 1 The e-commerce platform 82 is deployed in the Internet to provide corresponding services to its users, and the devices 80 and 81 of the merchant and consumer users of the e-commerce platform 82 are also connected to the Internet to use the services provided by the e-commerce platform.

[0031] The exemplary e-commerce platform 82 provides a supply and demand matching of products and / or services to the public through the Internet infrastructure, in which the products and / or services are provided as commodity information. For the sake of simplicity, the concepts of commodity, product, etc. are used in the present application to refer to the products and / or services in the e-commerce platform 82, which can be physical products, digital products, tickets, service subscriptions, other offline services, etc.

[0032] Real-world entities can access the e-commerce platform 82 in the identity of users to use various online services provided by the e-commerce platform 82 to achieve the purpose of participating in the business activities implemented by the e-commerce platform 82. These entities can be natural persons, legal persons, or social organizations, etc. Corresponding to the two types of entities of merchants and consumers in the business activities, the e-commerce platform 82 correspondingly exists two types of users of merchant users and consumer users. The entities of the product circulation chain in the business activities, including manufacturers, sellers, retailers, logistics providers, etc., can use online services in the e-commerce platform 82 in the identity of merchant users, while the consumers in the business activities, including real or potential consumers, can use online services in the e-commerce platform 82 in the identity of their corresponding consumer users. In actual business activities, the same entity can act as a merchant user and a consumer user, and flexible and variable understanding should be made accordingly.

[0033] The infrastructure for deploying the e-commerce platform 82 mainly includes a backend architecture and front-end devices. The backend architecture runs various online services through a service cluster, including middleware or front-end services for platform parties, services for consumers, services for merchants, etc., to enrich and perfect its service functions. The front-end devices mainly cover terminal devices used by users as clients to access the e-commerce platform 82, including but not limited to various mobile terminals, personal computers, point-of-sale devices, etc. For example, a merchant user can use his terminal device 80 to input product information for his online store, or use the interface opened by the e-commerce platform to generate product information; a consumer user can access the web page of the online store implemented by the e-commerce platform 82 through his terminal device 81, trigger the shopping process through the shopping button provided on the web page, and call various online services provided by the e-commerce platform 82 in the shopping process, so as to achieve the purpose of shopping and ordering.

[0034] In some embodiments, the e-commerce platform 82 can be implemented by a processing facility including a processor and a memory, which stores a set of instructions that, when executed, cause the e-commerce platform 82 to perform the electronic commerce and support functions involved in this application. The processing facility can be part of a server, a client, a network infrastructure, a mobile computing platform, a cloud computing platform, a fixed computing platform, or other computing platforms, and provides electronic components of the e-commerce platform 82, merchant devices, payment gateways, application developers, marketing channels, transportation providers, customer devices, point-of-sale devices, etc.

[0035] The e-commerce platform 82 can be implemented as a cloud computing service, a software as a service (SaaS), an infrastructure as a service (IaaS), a platform as a service (PaaS), a desktop as a service (DaaS), a hosted software as a service, a mobile backend as a service (MBaaS), an information technology management as a service (ITMaaS), and the like online services. In some embodiments, various functional components of the e-commerce platform 82 can be implemented to be suitable for operation on various platforms and operating systems, for example, an administrator user of an online store enjoys the same or similar functions in various embodiments such as iOS, Android, HomonyOS, or web page, and the like.

[0036] The e-commerce platform 82 can implement respective independent stations for various merchants to run their respective online stores, and provide respective business management engine instances for the merchants to establish, maintain, and run one or more online stores in one or more independent stations. The business management engine instance can be used for content management, task automation, and data management of one or more online stores, and can configure various specific business processes of the online store through an interface or a built-in component to support the implementation of business activities. The independent station is the infrastructure of the e-commerce platform 82 with cross-border service functions, and the merchant can maintain its online store based on the independent station in a relatively centralized and autonomous manner. The independent station usually has a domain name and storage space dedicated to the merchant, and different independent stations have relative independence. The e-commerce platform 82 can provide standardized or personalized technical support for a large number of independent stations, so that the merchant user can customize a business management engine instance suitable for itself, and use this business management engine instance to maintain one or more online stores owned by the merchant.

[0037] The online store can be configured and maintained by the merchant user logging in as an administrator to its business management engine instance. With the support of various online services provided by the infrastructure of the e-commerce platform 82, the merchant user can configure various functions in its online store and review various data as an administrator, for example, the merchant user can manage various aspects of its online store, such as viewing recent activities of the online store, updating product catalog of the online store, managing orders, recent visit activities, total order activities, and the like; the merchant user can also view more detailed information about the business and visitors to the online store of the merchant by obtaining reports or metrics, such as displaying sales summary of the overall business of the merchant, specific sales and engagement data of the activity sales marketing channel, and the like.

[0038] The e-commerce platform 82 can provide a communication facility and associated merchant interface for providing electronic communications and marketing, such as collecting and analyzing communication interactions between merchants, consumers, merchant devices, customer devices, point-of-sale devices, etc. using an electronic message aggregation facility, aggregating and analyzing communications, such as for increasing the potential for providing product sales, etc. For example, a consumer can have a question about a product that can generate a conversation between the consumer and a merchant (or an automated processor-based agent representing the merchant), where the communication facility is responsible for the interaction and provides the merchant with analysis on how to improve the probability of a sale.

[0039] In some embodiments, an application suitable for installation to a terminal device can be provided to serve the access needs of different users, so as to enable various users to access the e-commerce platform 82, such as a merchant back-end module of an online store in the e-commerce platform 82, etc. in the terminal device by running the application. In the process of implementing business activities through these functions, the e-commerce platform 82 can implement various functions related to the implementation of business activities as middleware or online services and open corresponding interfaces, and then implant the interface access functions into the application to implement function extension and task implementation. The business management engine can include a series of basic functions, and expose these functions to online services and / or application calls through APIs. Online services and applications use corresponding functions by remotely calling corresponding APIs.

[0040] With the support of various components of the business management engine instance, the e-commerce platform 82 can provide online shopping functions, enabling merchants to establish contact with customers in a flexible and transparent manner, and enabling consumer users to select and purchase goods online, create a product order, provide a delivery address for the goods in the product order, and complete payment confirmation of the product order. Then, the merchant can review and complete or cancel the order.

[0041] An e-commerce customer service response method of the present application can be programmed as a computer program product, deployed in a client or server for running, such as in the exemplary application scenario of the present application, can be deployed and implemented in the server of the e-commerce customer service platform. Thereby, the interface opened after running of the computer program product can be accessed, and the process of the computer program product can be interacted with through a graphical user interface to execute the method.

[0042] Referring to Figure 2 The e-commerce customer service response method of the present application, in a typical embodiment thereof, includes the following steps:

[0043] Step S1100, in response to a customer service consultation event triggered by a user, obtaining target consultation text corresponding to the event, and determining a matching result between the target consultation text and a plurality of preset dictionaries of consultation categories, respectively;

[0044] On an e-commerce platform, users have various ways to consult customer service. For example, when a user is browsing a product detail page and has doubts about the product's material, size, color, etc., the user can click the "customer service consultation" button set on the page, which triggers a customer service consultation event. Or on the order settlement page, the user can initiate consultation through the pop-up customer service chat window in the face of complex payment methods, coupon usage rules, etc., which also triggers a customer service consultation event. In addition, if the user encounters a product return or exchange problem after purchase, the user can find the customer service entry on the order detail page to consult, which triggers a customer service consultation event. Furthermore, the e-commerce platform can provide a customer service entry on the user's personal homepage to facilitate the user's consultation with the customer service at any time, thereby triggering a customer service consultation event.

[0045] When the server of the e-commerce platform detects a customer service consultation event, it can respond to the event and create a session between the user involved in the event and the customer service as a customer service consultation session. In the customer service consultation session, the user and the customer service conduct a question-and-answer dialogue, usually the user provides consultation text, and the customer service provides corresponding response text in response to the consultation text. Thus, the server can obtain the customer service consultation session, which includes all dialogue content between the user and the customer service from beginning to end, and then take the latest consultation text provided by the user in the session as the target consultation text.

[0046] The multiple consultation categories cover various different question types of e-commerce customer service consultation, which are directed to different consultation intents, and those skilled in the art can configure them as needed or based on customer service operation experience. In one embodiment, the multiple consultation categories include definition category, operation category, and reason category. To facilitate understanding, a number of typical consultation texts belonging to different consultation categories are exemplarily listed, such as "What does it mean to bind an independent domain name?", "How to bind an independent domain name?", and "Why do you want to bind an independent domain name?". As can be seen, the semantics of these typical consultation texts are related, but the actual intents and required response contents are completely different, so it is necessary to provide accurate customer service response text required for answering the target consultation text based on the consultation category to which the target consultation text belongs.

[0047] In order to efficiently and accurately locate the target consultation category to which the target consultation text belongs, a preset dictionary for different consultation categories is prepared, which is a set of multiple specific expression texts pre-constructed for the consultation category, and the specific expression text refers to the word used in the semantic environment or language expression habit of the consultation category, which can be a phrase or a word. Those skilled in the art can flexibly prepare the preset dictionary according to the disclosure herein. To facilitate understanding, the preset dictionaries corresponding to each consultation category are exemplarily listed, for example: the dictionary of the definition category includes common words such as "what", "definition", and "concept"; the dictionary of the operation category includes common words such as "how" and "step"; and the dictionary of the reason category includes common words such as "purpose", "meaning", "reason", and "why".

[0048] In an embodiment, for each preset dictionary of a consultation category, each specific expression text in the dictionary is character-matched with a word element in the target consultation text, and the matching result represents whether the dictionary hits, i.e., whether the specific expression text in the dictionary exists in the target consultation text. Accordingly, if the specific expression text in the preset dictionary is a word element in the target consultation text, i.e., the matching result is the consultation category corresponding to the preset dictionary, it represents that the target consultation text hits the specific expression text in the dictionary; if the specific expression text in the preset dictionary is not a word element in the target consultation text, i.e., the matching result is a matching failure or no matching dictionary exists, it represents that the target consultation text does not hit any specific expression text in the dictionary.

[0049] Step S1200, when the matching result represents that the dictionary hits, determining, according to the matching result, a target consultation category to which the target consultation text belongs in the plurality of consultation categories;

[0050] It can be understood that at this time, it means that the target consultation text uses the specific expression text in the hit dictionary in expression, and thus the consultation category corresponding to the dictionary is confirmed as the target consultation category to which the target consultation text belongs.

[0051] Step S1300, when all the matching results represent that the dictionary does not hit, determining, using a preset intent recognition model, a target consultation category to which the target consultation text belongs in the plurality of consultation categories;

[0052] It can be understood that at this time, it means that the target consultation text does not use any specific expression text in any dictionary in expression. For this purpose, in order to continue to locate the target consultation category to which the target consultation text belongs and ensure the accuracy and reliability of the category location, the intent recognition model is used.

[0053] In an embodiment, the model architecture of the intent recognition model is a neural network model suitable for a multi-classification task in a natural field, and each classifier in the neural network model is used to output a classification probability of the consultation text belonging to a corresponding consultation category. The intent recognition model is pre-trained to a convergent state and learns the ability to determine the consultation category to which the input consultation text belongs. Those skilled in the art can flexibly and variably implement the training of the intent recognition model according to the reasoning process of the intent recognition model disclosed herein and disclosed below, so that the intent recognition model has the ability.

[0054] The intent recognition model takes the target consultation text as input. After the deep semantic information of each word unit in the target consultation text is extracted by the hidden layer in the model, the full-text deep semantic information integrating the deep semantic information is obtained. The deep semantic information of the word unit represents the semantic of the word unit itself and the associated semantic between the word unit and other word units. Then, by each classifier in the output of the model, the full-text deep semantic information is mapped to the corresponding classification space to determine the classification probability that the target consultation text belongs to the consultation category corresponding to the classification space. Finally, the consultation category with the highest classification probability and exceeding the preset threshold is selected as the target consultation category. The preset threshold can be set by the person skilled in the art as needed, for example, 0.9.

[0055] In step S1400, from the preset question and answer knowledge base of the target consultation category, the customer service response text matching the target consultation text is obtained and pushed to the user.

[0056] According to the target consultation category to which the target consultation text is positioned, the preset question and answer knowledge base of the target consultation category is called. The target consultation text is matched with each consultation text set in the question and answer knowledge base. Specifically, taking a single consultation text set as an example, the vector similarity algorithm is used to determine the vector similarity between the text semantic vector of the target consultation text and the text semantic vector of the standard consultation text in the set, and the vector similarity between the text semantic vector of the target consultation text and the text semantic vector of each similar consultation text in the set is determined as the second semantic similarity. Then, the maximum semantic similarity is selected from the first semantic similarity and the plurality of second semantic similarities as the consultation matching degree between the target consultation text and the consultation text set. The text semantic vector can be obtained by applying the pre-trained text encoding model to encode. The person skilled in the art can flexibly realize it. In addition, in order to ensure the efficiency of online matching, the text semantic vector of each consultation text in the consultation text set in the question and answer knowledge base of each consultation category can be pre-encoded by the text encoding model, and then stored in association with the corresponding consultation text for calling, so that when online matching, only the target consultation text needs to be encoded in real time by the text encoding model, and the text semantic vector of each consultation text in the consultation text set required for matching can be directly called without online real-time encoding. The vector similarity algorithm can be flexibly realized by the person skilled in the art, for example, the cosine similarity algorithm.

[0057] From the consultation matching degrees between the target consultation text and each consultation text set, the consultation text set with the maximum matching degree and exceeding the preset threshold is selected. The standard response text corresponding to the consultation text set in the question and answer knowledge base is taken as the customer service response text corresponding to the target consultation text, and the customer service response text is pushed to the user to complete the response to the user's consultation customer service.

[0058] It can be understood that for each consultation category, a corresponding question and answer knowledge base can be constructed in advance. The question and answer knowledge base includes a plurality of consultation text sets belonging to the consultation category and corresponding standard answer texts. The consultation text set includes a standard consultation text and a plurality of similar consultation texts, which are highly similar in semantics to the standard consultation text and use different expressions from the standard consultation text. The standard answer text can answer each consultation text in the corresponding consultation text set, and is therefore used to answer these consultation texts. Those skilled in the art can prepare the question and answer knowledge base in advance according to the business needs according to the disclosure.

[0059] According to the above embodiments, it is not difficult to understand that the present application has many advantages over the prior art, at least including:

[0060] The present application first responds to the event triggered by the user consultation customer service, matches the corresponding target consultation text with the preset dictionaries of a plurality of consultation categories to obtain a result, and when the dictionary hits the matching success, the target consultation category can be accurately determined at a speed close to no delay. For common and standardized problems in the e-commerce customer service field, rule matching shows excellent efficiency, which can realize instant response and ensure that users obtain the required answers in the shortest time.

[0061] Secondly, the problem of lack of flexibility caused by simply relying on rule matching is avoided. The complex or ambiguous target consultation text provided by the service to the user's diverse expression habits, in the case that all preset dictionaries cannot hit, the intent recognition model is enabled, which uses the semantic ability and generalization ability learned to deeply analyze the real consultation intent hidden in the complex or unstructured text hidden by the user, so that the target consultation category can still be accurately determined in the blank area that cannot be covered by the rules. In this way, rule matching first and model recognition bottom-up, the dual-track classification strategy of the organic combination of the two, builds a complete and highly robust automatic classification system, whether it is a high-frequency simple routine problem or a low-frequency complex special problem, the corresponding consultation category can be accurately identified, greatly improving the accuracy and reliability of classification.

[0062] Further, based on the accurate positioning of the target consultation category, the matching efficiency and quality of the response content are further optimized. Through the subdivision of the question and answer knowledge base, each consultation category corresponds to a dedicated knowledge base, which focuses on solving the problems of users in this category. When searching for customer service response text, only a precise search is needed within the highly relevant knowledge subset, greatly improving the search efficiency and significantly enhancing the relevance and accuracy of the final pushed response content. This effectively avoids the noise interference and low efficiency problems that may occur in traditional large-scale unified knowledge base search, ensuring that the response information received by the user is accurate and practical, thereby greatly optimizing the customer service experience and making the user feel the thoughtful and professional service.

[0063] In summary, the customer service consultation service provided based on the consultation category is realized, the double-track classification strategy of the priority matching dictionary combined with the intent recognition model as a complementary strategy, and the directional retrieval of the knowledge base organized by categories, which comprehensively improves the overall performance of the e-commerce customer service. Not only does it reduce the peak demand for computing resources and significantly improve response speed, but it also enhances the accurate understanding and efficient processing capacity of complex consultations, improves response accuracy and efficiency, and greatly improves user experience. These comprehensive advantages play an important role in improving the operational efficiency of e-commerce platforms, enhancing user satisfaction, and shaping a good service image, and promote the e-commerce industry to a higher level in the key link of customer service.

[0064] In further embodiments, step S1100, in response to a user triggered customer service consultation event, includes the following steps:

[0065] Step S1110, obtaining a plurality of historical consultation texts, clustering the specific expression texts extracted from the historical consultation texts, and determining the class cluster to which the specific expression texts belong;

[0066] These historical consultation texts can come from the past customer service consultation session records of the e-commerce platform, covering specific expressions of various user consultation problems.

[0067] In one embodiment, a pre-trained named entity recognition model in a convergent state can be used, which has learned the ability to recognize whether each word token in the input text belongs to a specific expression text. Those skilled in the art can flexibly implement the training of the named entity recognition model to make it learn this ability. Thus, for each historical consultation text, the named entity recognition model can be used to efficiently and accurately identify the specific expression text in the historical consultation text and extract the specific expression text.

[0068] In another embodiment, a preset e-commerce dictionary can be used to perform semantic matching between each word unit in each historical consultation text and each e-commerce word in the e-commerce dictionary, to determine each unmatched word unit in the historical consultation text as a specific expression text. The e-commerce dictionary includes a plurality of phrases or words corresponding to e-commerce operation and e-commerce management. These words are greatly different from the specific expression text in expression and semantics. For ease of understanding, the consultation text is "how to bind an independent domain name", in which "bind" and "independent domain name" are word units in the e-commerce dictionary, and "how" is a specific expression text not in the e-commerce dictionary.

[0069] A clustering algorithm is applied to cluster all extracted specific expression texts. In this process, the closer the semantics between specific expression texts, the closer the distance between them in the clustering space. Thus, the relative distance between specific expression texts in the clustering space can be determined according to the semantics between them, and finally, each region in the clustering space is formed by clustering specific expression texts. Each region represents a single cluster, and all specific expression texts in the region belong to the cluster.

[0070] The clustering algorithm can be selected from one of the following algorithms, including but not limited to K-Means clustering algorithm, mean shift clustering algorithm, density-based clustering algorithm (DBSCAN), Gaussian mixture model-based expectation maximization clustering algorithm, agglomerative hierarchical clustering algorithm, and graph community detection clustering algorithm.

[0071] In step S1120, for each cluster, a large language model is used to determine the consultation category corresponding to the cluster according to a plurality of typical specific expression texts selected from the cluster.

[0072] It can be understood that each cluster has a cluster center, so that a plurality of specific expression texts closest to the cluster center can be selected from each cluster as typical specific expression texts in the cluster. The specific number of specific expression texts selected can be set by those skilled in the art as needed.

[0073] Further, a preset category definition template is used to embed all typical specific expression texts of the cluster into the category definition template to obtain a category definition prompt text, which is used to guide the large language model to make a definitional description of the consultation category to which the specific expression texts in the cluster belong, and to express the consultation category in natural language. Those skilled in the art can flexibly configure the category definition template according to the disclosure.

[0074] The large language model is trained to convergence in advance using an extremely large corpus, acquires the ability to generate text describing human language, and has accurate basic text semantic understanding ability and basic logical reasoning ability. The large language model can be Chinchilla, PaLM, LLaMA, Alpaca, Vicuna, etc., and any one of them can be selected as needed.

[0075] Step S1130, constructing a dictionary of the consultation category corresponding to the cluster according to the specific expression text belonging to the cluster.

[0076] An empty dictionary of the consultation category corresponding to the cluster is created, and all specific expression texts belonging to the cluster are added to the dictionary.

[0077] In this embodiment, by clustering the specific expression texts in the historical consultation texts, the potential semantic association and problem type rules in the past customer service consultation data of the e-commerce platform can be deeply mined, and the consultation categories can be accurately determined in a data-driven manner, avoiding the subjective limitations of manually presetting the consultation categories, making the consultation category division more in line with the actual business scenarios and user needs. With the help of the large language model to determine the consultation categories and construct the dictionary for the cluster, the powerful semantic understanding and text generation capabilities are fully utilized, which can ensure the accuracy of the consultation category definition and the richness and representativeness of the dictionary content, thereby laying a solid foundation for subsequent rapid and accurate matching of the target consultation category, effectively improving the automation level and accuracy of the customer service consultation classification, optimizing the user consultation experience, and reducing the workload of manually maintaining the dictionary and category definition, improving the overall flexibility and scalability.

[0078] In a further embodiment, step S1100, obtaining a target consultation text corresponding to the event, and determining the matching results between the target consultation text and the preset dictionaries of multiple consultation categories, includes the following steps:

[0079] Step S1101, obtaining a customer service consultation session corresponding to the customer service consultation event;

[0080] The customer service consultation session includes the dialogue content between the user and the customer service from the beginning to the current entire process of the customer service consultation event, and each consultation text provided by the user and its sending time in the dialogue content, and each response text provided by the customer service and its sending time.

[0081] Step S1102, according to whether the target consultation text in the customer service consultation session belongs to the initial consultation text, when the target consultation text does not belong to the initial consultation text, constructing a rewriting prompt text based on the target consultation text and its previous segment text in the customer service consultation session;

[0082] The sending time of each consultation text in the customer service consultation session is screened to obtain the earliest sending time, and it is determined whether the sending time is the sending time of the target consultation text. If yes, it is confirmed that the target consultation text is the consultation text sent by the user for the first time, that is, the target consultation text belongs to the initial consultation text. Therefore, the target consultation text does not need to be rewritten, because it is impossible to make a reasonable rewriting based on its previous content, and forced rewriting may lead to the rewritten target consultation text deviating from the user's original intention.

[0083] If the sending time of the target consultation text is not the earliest sending time, it is confirmed that the target consultation text is not the consultation text sent by the user for the first time, that is, the target consultation text does not belong to the initial consultation text. The rewriting process of the target consultation text is started. First, the multiple consultation texts before the sending time of the target consultation text in the customer service consultation session are obtained to obtain the previous segment text of the target consultation text. The specific number of the obtained consultation texts can be set by a person skilled in the art as needed. The target consultation text and its previous segment text are embedded in the preset rewriting prompt template to obtain a rewriting prompt text.

[0084] In step S1103, a large language model is used to determine whether the consultation intention expressed by the target consultation text meets the standard according to the rewriting prompt text. When it does not meet the standard, the rewritten target consultation text output by the large language model is obtained. When it meets the standard, the original target consultation text output by the large language model is obtained.

[0085] The rewriting prompt text guides the large language model to determine whether the consultation intention of the target consultation text is clearly and accurately expressed. If the expression is not clear or accurate, the rewritten text is output to optimize the expression and make it more consistent with the accurate communication of the consultation intention. If the expression is clear and accurate, the original text of the target consultation text is directly output without rewriting. A person skilled in the art can flexibly modify the rewriting prompt template according to the disclosure.

[0086] In this embodiment, by judging whether the target consultation text is the initial consultation text after obtaining the target consultation text and rewriting it accordingly, the context coherence and semantic integrity of the customer service consultation session are fully considered. For non-initial consultation texts, the expression of the target consultation text can be optimized based on the previous segment text to make the consultation intention clearer and more accurate, avoiding errors in subsequent matching or deviations in response due to ambiguous or incomplete text expression. The introduction of the large language model for rewriting judgment and text optimization can realize intelligent text quality control to make the subsequent matching process more accurate, improve the accuracy and reliability of the entire customer service response service, increase the probability of obtaining accurate responses for users, enhance the trust and satisfaction of users for the customer service, and ensure the efficiency and professionalism of the customer service consultation processing.

[0087] In a further embodiment, step S1300 includes the following steps of:

[0088] Step S1310, constructing a classification prompt text based on the target consultation text, the classification prompt text containing a plurality of exemplary classification examples, the exemplary classification examples including a typical consultation text and its intended consultation category in the plurality of consultation categories;

[0089] It can be understood that for each consultation category, the typical specific expression text in the historical consultation text corresponding to the class cluster of the consultation category can be obtained as the typical consultation text of the consultation category, and the consultation category is taken as the intended consultation category to which the typical consultation text belongs, and then the typical consultation text and the intended consultation category to which it belongs are spliced into an exemplary classification example.

[0090] Step S1320, determining the target consultation category of the target consultation text in the plurality of consultation categories by using a large language model according to the classification prompt text.

[0091] The classification prompt text is used to guide the consultation classification demonstration made by the large language model with reference to the plurality of exemplary classification examples, so as to determine the target consultation category of the target consultation text in the plurality of consultation categories. Those skilled in the art can flexibly configure the classification prompt template according to the disclosure.

[0092] In the embodiment, the classification prompt text containing the exemplary classification examples is constructed based on the target consultation text, and the target consultation category is determined by using the large language model, which combines the exemplary learning and the powerful semantic reasoning ability of the large language model, can more accurately understand and grasp the deep semantics of the target consultation text, and avoid the classification errors caused by simply relying on the model. Especially when the semantics of the consultation text is complex, ambiguous or ambiguous, the exemplary classification examples provide clear reference for the model, which helps to further improve the classification accuracy. This classification method guided by examples enhances the generalization ability and adaptability of the intent recognition model in dealing with diversified and complex consultation texts, so that the determination of the target consultation category is more in line with the actual business scenario and the real intention of the user, thereby ensuring the pertinence and effectiveness of the subsequent response content, optimizing the quality of customer service and user experience.

[0093] In a further embodiment, step S1400 includes the following steps of:

[0094] Step S1410, determining the target consultation text corresponding to the speculative response text by using a preset customer service response model;

[0095] The customer service response model is pre-trained to a convergent state to learn the ability to accurately determine the appropriate response text according to the consultation text. Thus, the target consultation text can be taken as the input of the customer service response model to obtain the response text generated by the inference of the model as the speculative response text. Those skilled in the art can flexibly train the customer service response model to a convergent state according to the model ability disclosed herein, or can refer to the disclosure of the subsequent embodiments for implementation.

[0096] In step S1420, each set of consultation texts and the corresponding standard response texts in the preset Q&A knowledge base of the target consultation category are matched with the target consultation text and the corresponding speculative response text to determine the corresponding consultation matching degree and response matching degree.

[0097] Taking a single set of consultation texts and the corresponding standard response texts as an example, the vector similarity algorithm is used to determine the vector similarity between the text semantic vector of the target consultation text and the text semantic vector of the standard consultation text in the set, and the vector similarity between the text semantic vector of the target consultation text and the text semantic vector of each similar consultation text in the set is determined as the second semantic similarity. Then, the maximum semantic similarity is selected from the first semantic similarity and the plurality of second semantic similarities as the consultation matching degree between the target consultation text and the set of consultation texts. In addition, the vector similarity algorithm is also used to determine the semantic similarity between the speculative response text and the standard response text as the response matching degree.

[0098] Similarly, in order to ensure the efficiency of online matching, the text semantic vector of each standard response text in the Q&A knowledge base can be pre-encoded by applying the text encoding model, and then the text semantic vectors are stored in association with the corresponding standard response texts for calling, so that when online matching, only the speculative response text needs to be encoded in real time by applying the text encoding model, and the text semantic vector of the standard response text required for matching can be directly called without online real-time encoding.

[0099] In step S1430, the corresponding standard response text is selected as the target customer service response text according to the consultation matching degree and the response matching degree.

[0100] For each pair of knowledge in the Q&A knowledge base, i.e., a single set of consultation texts and the corresponding standard response texts, the consultation matching degree and the response matching degree of the pair of knowledge are multiplied by the respective weights, and then the comprehensive matching degree is obtained by adding them up. Then, the standard response text in the knowledge pair with the highest comprehensive matching degree and exceeding the preset threshold is selected as the target customer service response text.

[0101] In this embodiment, the customer service response model is used to determine the speculative response text, and the customer service response text is matched with the set of consultation texts and standard response texts in the question and answer knowledge base. The matching degree is comprehensively evaluated from two dimensions of consultation and response. Compared with matching from a single dimension, the knowledge base item that best matches the target consultation text can be more accurately located, ensuring that the pushed customer service response text is not only accurate in problem understanding, but also appropriate, complete and in line with user needs in response content. Through comprehensive matching degree optimization of the standard response text, the value of the question and answer knowledge base is fully tapped, the associated information of each item in the knowledge base is fully utilized, the supporting role of the knowledge base for customer service response is maximized, the accuracy and quality of the response are further improved, the response errors or insufficient cases caused by one-sided matching are reduced, the customer service response process is optimized in all aspects, the user obtains more accurate and satisfactory response service, and the professionalism and reliability image of the e-commerce platform customer service are strengthened.

[0102] In further embodiments, after step S1400, obtaining the customer service response text corresponding to the target consultation text from the preset question and answer knowledge base of the target consultation category and pushing the customer service response text to the user, the following steps are included:

[0103] Step S1500, obtaining the feedback satisfaction degree of the user receiving the customer service response text;

[0104] In the e-commerce customer service response process, after the user receives the response text pushed by the customer service, the satisfaction degree is a key indicator for measuring the quality of the response. The feedback satisfaction degree can be obtained in various ways, such as setting a satisfaction survey on the display page of the customer service consultation session, the user can score and evaluate this customer service response, generally with options such as "very satisfied", "satisfied", "average", "dissatisfied", "very dissatisfied", or monitoring the user's subsequent behavior after receiving the response, such as whether to ask the same or similar questions again (which may indicate dissatisfaction), whether to give good or bad reviews, etc., and indirectly infer the feedback satisfaction degree based on these behaviors.

[0105] Step S1510, when the feedback satisfaction degree of the customer service response text meets the standard, the target consultation text and the customer service response text corresponding thereto are taken as a training sample and a corresponding supervision label;

[0106] When the user feedback satisfaction degree reaches the pre-set qualified standard (for example, the satisfaction degree evaluation is "satisfied" or above), it indicates that this customer service response is successful and can accurately answer the user's consultation question. At this time, the target consultation text initially proposed by the user is taken as a training sample, and the corresponding customer service response text is taken as a supervision label. Such sample-label pair can effectively represent a reasonable consultation and response relationship.

[0107] Step S1520, add the training sample and its supervised label to a training set, and use the training set to train the customer service response model to a convergent state, so that the customer service response model learns the ability to determine the corresponding suspected response text of the consultation text.

[0108] The training samples and labels obtained from the actual satisfactory responses are continuously accumulated and expanded to the training set of the customer service response model. When the total number of incremental training samples in the training set reaches a preset threshold, the training sample and its supervised label that have existed in the training set for the longest time are removed from the training set, and the threshold can be set by a person skilled in the art as needed. Then, the training set is called to supervise the training of the customer service response model by iteratively calling the real and high-quality dialogue content in the training set, so that the model learns more accurate response patterns and rules, gradually reduces errors, and finally reaches the convergent state of the training, that is, the model shows stable and high-quality performance on the training set, and finally has the ability to accurately determine the appropriate suspected response text according to the consultation text, thereby guaranteeing the overall quality and accuracy of the customer service response, and enabling the customer service response model to better cope with various user consultation scenarios.

[0109] In the embodiment, by obtaining the user feedback satisfaction as a training sample and a supervised label, a high-quality training set is constructed for continuously optimizing the customer service response model, forming a closed-loop optimization mechanism from user actual experience to model improvement, so that the model can continuously learn and adjust based on real user satisfaction data, accurately grasp the user's response expectations for different consultation contents, gradually improve the response accuracy and adaptability of the model, and ensure that the customer service always meets the changes in user needs and business development. Regularly update the training set, remove old samples, ensure the freshness and representativeness of the training data, and keep the model in high-performance training with the times, provide more high-quality and actual demand-oriented customer service responses for users, promote the continuous improvement of customer service quality, and enhance user stickiness and platform competitiveness.

[0110] In a further embodiment, after the customer service response text corresponding to the target consultation text is pushed to the user, step S1400 includes the following steps:

[0111] Step S1600, each non-target consultation category in the plurality of consultation categories is taken as an extended consultation category.

[0112] After the target consultation category to which the target consultation text belongs is accurately positioned, and the corresponding customer response text is obtained and pushed to the user according to the target consultation category, the rest of the consultation categories that are not selected are redefined as extended consultation categories. Based on the concept of comprehensive and in-depth service to users, considering that when a user consults a problem, he or she may have potential doubts or needs about other related aspects, by converting the non-target consultation categories into extended consultation categories, a foundation is laid for further exploration and provision of more extensive and rich related information, aiming to broaden the service scope and improve the integrity and satisfaction of the user's consultation experience.

[0113] Step S1610, obtaining an extended consultation text corresponding to the target consultation text from a preset question and answer knowledge base of the extended consultation category;

[0114] Relying on the preset question and answer knowledge base corresponding to each extended consultation category. These question and answer knowledge bases, like the knowledge base of the target consultation category, are carefully constructed and filled in advance, covering a wealth of consultation texts and corresponding response contents under the corresponding consultation category. Based on the target consultation text, matching algorithms such as vector similarity are used to compare and analyze the consultation text set in each extended consultation category knowledge base, so as to screen out the extended consultation text with high semantic similarity or correlation with the target consultation text. It can be seen that the deep mining and full use of knowledge base resources aims to mine the expansion information that may have implicit correlation with the user's initial consultation content from different angles and different fields, further enrich the range of customer service content available to users, and meet the user's deeper or more extensive consultation demands.

[0115] Step S1620, pushing the extended consultation text obtained for each of the extended consultation categories to the user.

[0116] After matching a series of extended counseling texts related to the extended counseling category, the texts are sorted and pushed to the user. In terms of content presentation, the content can be sorted according to the matching degree or integrated and displayed in a format that is easy for the user to understand and browse. Through this operation, the user not only receives accurate answers to the initial counseling question, but also obtains additional extended information filtered from the non-target counseling category that has potential correlation value. This approach helps users to more comprehensively understand the associated counseling response content, and even guides users to discover new concerns or new ideas for solving problems, shortens the time and effort of user counseling, greatly enhances the initiative, foresight of customer service, and the depth and breadth of interaction between users and customer service, further improves the user's sense of acquisition and satisfaction during the counseling process, and highlights the thoughtful and thoughtful e-commerce customer service, which helps to build a good user relationship and platform service image.

[0117] In this embodiment, by converting the non-target counseling category into an extended counseling category and pushing the user with the matched extended counseling text, the initial counseling of the user is taken as the starting point, and the potential associated counseling content is actively explored and mined, the service boundary is widened, and the user is provided with a more extensive and in-depth information acquisition channel, which meets the diversified and deep-seated needs of the user when counseling a problem, improves the completeness and richness of the user's counseling experience. By reasonably utilizing the question and answer knowledge base of each extended counseling category, the value of the knowledge base resources is maximized, the transformation of customer service from passive response to active recommendation is realized, the interaction mode between the user and the customer service is optimized, the user feels the thoughtful and thoughtful all-round service, effectively shortens the time cost of the user's independent exploration of related problems, improves the user's counseling efficiency, further enhances the user's trust and satisfaction with the e-commerce platform customer service, and helps the e-commerce platform to build a positive and good user relationship and service image.

[0118] Please refer to Figure 3, provided by one of the purposes of the present application, is a functional embodiment of the e-commerce customer service response method of the present application. On the other hand, provided by one of the purposes of the present application, an e-commerce customer service response device comprises an event response module 1100, a first category determination module 1200, a second category determination module 1300, and a user response module 1400. The event response module 1100 is used to respond to a customer service consultation event triggered by a user, obtain a target consultation text corresponding to the event, and determine the matching results between the target consultation text and a plurality of consultation category preset dictionaries respectively. The first category determination module 1200 is used to determine the target consultation category to which the target consultation text belongs in the plurality of consultation categories according to the matching result when the matching result represents a dictionary hit. The second category determination module 1300 is used to determine the target consultation category to which the target consultation text belongs in the plurality of consultation categories by using a preset intent recognition model when all the matching results represent a dictionary miss. The user response module 1400 is used to obtain a customer service response text corresponding to the target consultation text from a preset question and answer knowledge base of the target consultation category, and push the customer service response text to the user.

[0119] In further embodiments, the event response module 1100 comprises a text clustering submodule for obtaining a plurality of historical consultation texts, clustering specific expression texts extracted from the historical consultation texts, and determining the class cluster to which the specific expression texts belong. The category determination submodule is used to determine the consultation category corresponding to each class cluster by using a large language model according to a plurality of typical specific expression texts selected from the class cluster. The dictionary construction submodule is used to construct the dictionary of the consultation category corresponding to the class cluster according to the specific expression texts belonging to the class cluster.

[0120] In further embodiments, the event response module 1100 comprises a session acquisition submodule for acquiring a customer service consultation session corresponding to a customer service consultation event. The first prompt construction submodule is used to determine whether the target consultation text in the customer service consultation session belongs to the initial consultation text. When the target consultation text does not belong to the initial consultation text, the rewriting prompt text is constructed based on the target consultation text and its previous text in the customer service consultation session. The rewriting determination submodule is used to determine whether the consultation intent expressed by the target consultation text meets the standard by using a large language model according to the rewriting prompt text. When it does not meet the standard, the rewritten target consultation text output by the large language model is obtained. When it meets the standard, the unrewritten target consultation text output by the large language model is obtained.

[0121] In a further embodiment, the second category determining module 1300 comprises: a second prompt constructing submodule configured to construct a classification prompt text based on the target consultation text, the classification prompt text comprising a plurality of exemplary classification examples, the exemplary classification examples including a typical consultation text and an intended consultation category to which the typical consultation text belongs; and a category determining submodule configured to determine, by using a large language model, a target consultation category to which the target consultation text belongs from the plurality of consultation categories according to the classification prompt text.

[0122] In a further embodiment, the second category determining module 1300 comprises: a response inferring submodule configured to determine an inferred response text corresponding to the target consultation text by using a preset customer service response model; a matching degree determining submodule configured to perform matching between each consultation text set and the corresponding standard response text in a preset question and answer knowledge base of the target consultation category, and the target consultation text and the corresponding inferred response text, to determine corresponding consultation matching degrees and response matching degrees; and a response optimizing submodule configured to optimize the corresponding standard response text according to the consultation matching degrees and the response matching degrees, as a target customer service response text.

[0123] In a further embodiment, the user response module 1400 further comprises: a user feedback submodule configured to obtain a feedback satisfaction degree of the user receiving the customer service response text; a training set constructing submodule configured to, when the feedback satisfaction degree of the customer service response text meets a standard, correspond the target consultation text and the customer service response text as a training sample and a corresponding supervision label; and a model training submodule configured to add the training sample and the supervision label to a training set, and train the customer service response model to a convergent state by using the training set, so that the customer service response model learns the ability to determine an inferred response text corresponding to a consultation text.

[0124] In a further embodiment, the user response module 1400 further comprises: an expanded category submodule configured to take each non-target consultation category in the plurality of consultation categories as an expanded consultation category; a text matching submodule configured to obtain an expanded consultation text corresponding to the target consultation text from a preset question and answer knowledge base of the expanded consultation category; and a text pushing submodule configured to push the expanded consultation text corresponding to each of the expanded consultation categories to the user.

[0125] To solve the above technical problems, the embodiments of the present application further provide a computer device. As shown in Figure 4As shown, the internal structure diagram of the computer device is shown. The computer device includes a processor, a computer readable storage medium, a memory and a network interface connected by a system bus. Among them, the computer readable storage medium of the computer device stores an operating system, a database and computer readable instructions, the database can store control information sequence, and the computer readable instructions are executed by the processor to enable the processor to implement an e-commerce customer service response method. The processor of the computer device is used to provide computing and control capability to support the operation of the entire computer device. The memory of the computer device can store computer readable instructions, which, when executed by the processor, can enable the processor to execute the e-commerce customer service response method of the present application. The network interface of the computer device is used to connect and communicate with the terminal. Those skilled in the art can understand, Figure 4 The structure shown in the above embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0126] In the present embodiment, the processor is used to execute the specific functions of each module and its sub-modules in Figure 3 The memory stores the program codes and various data required to execute the above-mentioned modules or sub-modules. The network interface is used for data transmission between the user terminal or the server. The memory in the present embodiment stores the program codes and data required to execute all modules / sub-modules in the e-commerce customer service response device of the present application, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

[0127] The present application also provides a storage medium storing computer readable instructions, which, when executed by one or more processors, enable the one or more processors to perform the steps of the e-commerce customer service response method of any embodiment of the present application.

[0128] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments of the present application can be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of each method. Among them, the storage medium can be a computer readable storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0129] In summary, the present application can efficiently and accurately answer user inquiries and improve the satisfaction of customer service.

[0130] Those skilled in the art can understand that the steps, measures and schemes in the various operations, methods and processes discussed in the present application can be alternated, changed, combined or deleted. Further, other steps, measures and schemes in the various operations, methods and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined or deleted. Further, the steps, measures and schemes in the prior art with the various operations, methods and processes disclosed in the present application can also be alternated, changed, rearranged, decomposed, combined or deleted.

[0131] The above only describes some embodiments of the present application, and it should be pointed out that those skilled in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. An e-commerce customer service response method, characterized in that: The steps include: Responding to a customer service consultation event triggered by a user, obtaining the target consultation text corresponding to the event, and determining the matching results between the target consultation text and preset dictionaries of multiple consultation categories; When the matching result represents a dictionary hit, determining a target consulting category to which the target consulting text belongs among multiple consulting categories according to the matching result; When all the matching results indicate a dictionary miss, a preset intention recognition model is used to determine a target consultation category to which the target consultation text belongs among multiple consultation categories; From the preset question and answer knowledge base of the target consultation category, obtain the customer service response text corresponding to the target consultation text and push it to the user.

2. The e-commerce customer service response method according to claim 1, characterized in that: Responding to a customer service consultation event triggered by a user includes the following steps: Acquire multiple historical consultation texts, cluster specific expression texts extracted from the historical consultation texts, and determine the cluster to which the specific expression texts belong; For each cluster, a large language model is used to determine the consultation category corresponding to the cluster based on multiple typical specific expression texts selected from the cluster; According to the specific expression text belonging to the cluster, a dictionary of the consultation category corresponding to the cluster is constructed.

3. The e-commerce customer service response method according to claim 1, characterized in that: Obtaining the target consultation text corresponding to the event and determining the matching results between the target consultation text and the preset dictionaries of multiple consultation categories respectively include the following steps: Get the customer service consultation session corresponding to the customer service consultation event; According to whether the target consultation text in the customer service consultation session belongs to the starting consultation text, if the target consultation text does not belong to the starting consultation text, constructing a rewriting prompt text based on the target consultation text and its previous text fragment in the customer service consultation session; A large language model is used to determine whether the consultation intention expressed by the target consultation text meets the standard based on the rewritten prompt text. If it does not meet the standard, the rewritten target consultation text output by the large language model is obtained. If it meets the standard, the unrewritten target consultation text output by the large language model is obtained.

4. The e-commerce customer service response method according to claim 1, characterized in that: Determining the target consultation category to which the target consultation text belongs among multiple consultation categories using a preset intention recognition model includes the following steps: Constructing a classification prompt text based on the target consultation text, wherein the classification prompt text includes a plurality of exemplary classification samples, and the exemplary classification samples include a typical consultation text and an intended consultation category among the plurality of consultation categories to which it belongs; A large language model is used to determine the target consultation category to which the target consultation text belongs among multiple consultation categories based on the classification prompt text.

5. The e-commerce customer service response method according to claim 1, characterized in that: Obtaining a customer service response text that matches the target consultation text from a preset question-and-answer knowledge base of the target consultation category and pushing it to the user includes the following steps: Using a preset customer service response model to determine the inferred response text corresponding to the target consultation text; Matching each consultation text set and its corresponding standard answer text in the preset question-answer knowledge base of the target consultation category with the target consultation text and its corresponding inferred answer text to determine the corresponding consultation matching degree and answer matching degree; According to the consultation matching degree and the response matching degree, a corresponding standard response text is selected as the target customer service response text.

6. The e-commerce customer service response method according to claim 1, characterized in that: After obtaining a customer service response text that matches the target consultation text from a preset question-and-answer knowledge base of the target consultation category and pushing it to the user, the following steps are included: Obtaining the user's feedback satisfaction with the customer service response text; When the customer service response text meets the feedback satisfaction standard, the target consultation text and the customer service response text are used as training samples and their corresponding supervision labels; The training samples and their supervision labels are added to a training set, and the training set is used to train the customer service response model to a convergence state, so that the model can learn the ability to determine the inferred response text corresponding to the consultation text.

7. The e-commerce customer service response method according to claim 1, characterized in that: After obtaining a customer service response text that matches the target consultation text from a preset question-and-answer knowledge base of the target consultation category and pushing it to the user, the following steps are included: Treat each non-target consultation category in multiple consultation categories as an expanded consultation category; Obtaining an extended consultation text corresponding to the target consultation text from a preset question-and-answer knowledge base of the extended consultation category; The extended consultation texts obtained corresponding to the respective extended consultation categories are pushed to the user.

8. An e-commerce customer service response device, characterized in that: include: An event response module is used to respond to customer service consultation events triggered by users, obtain the target consultation text corresponding to the event, and determine the matching results between the target consultation text and preset dictionaries of multiple consultation categories; A first category determination module is configured to determine, when the matching result represents a dictionary hit, a target consulting category to which the target consulting text belongs among multiple consulting categories according to the matching result; A second category determination module is configured to, when all the matching results indicate a dictionary miss, determine the target consultation category to which the target consultation text belongs among the multiple consultation categories using a preset intention recognition model; The user response module is used to obtain the customer service response text corresponding to the target consultation text from the preset question and answer knowledge base of the target consultation category and push it to the user.

9. A computer device comprising a central processing unit and a memory, characterized in that: The central processing unit is configured to call and run a computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that It stores a computer program implemented according to the method described in any one of claims 1 to 7 in the form of computer-readable instructions, and when the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

Images