Multi-language intelligent question and answer method and device, equipment and medium

Abstract

The invention relates to a multi-language intelligent question and answer method and device, equipment and a medium in the technical field of e-commerce, and the method comprises the steps: obtaining a question text provided by a user, and determining a question language of the question text; when the question language is consistent with a knowledge language corresponding to the question and answer knowledge base, determining a pre-stored question matched with the question text in the question and answer knowledge base, and taking an answer text corresponding to the pre-stored question as a target answer text; when the question language is inconsistent with the knowledge language, translating the question text into a target question text corresponding to the knowledge language, determining a pre-stored question matched with the target question text in a question and answer knowledge base, and translating an answer text corresponding to the pre-stored question into a target answer text corresponding to the knowledge language; and sending the target answer text to the user. Only one question and answer knowledge base of a specific language needs to be maintained, and the management and maintenance workload of the question and answer knowledge base is remarkably reduced.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to a multilingual intelligent question-answering method and its device, equipment, and medium. Background Art

[0002] In the field of e-commerce technology, providing customer service is crucial. It not only responds to the inquiries and needs of buyers of goods instantly, but also provides personalized shopping suggestions and / or solutions, thus significantly improving consumers' shopping experience and satisfaction. For merchants, high-quality customer service is a key factor in building customer trust, enhancing brand loyalty and promoting sales. In addition, the introduction of automated customer service has greatly reduced the workload of manual customer service and improved overall operational efficiency.

[0003] However, in traditional multilingual customer service scenarios, it is often necessary to maintain one or more independent knowledge bases and answer systems for each language. In this model, answers are retrieved directly from the knowledge base of the corresponding language based on the language of the question text provided by the user. However, the maintenance of multiple knowledge bases not only increases the complexity of management and maintenance, but also leads to difficulties in knowledge updating and synchronization. Furthermore, if the knowledge bases in different languages ​​are not updated in a timely manner or maintained insufficiently, the service quality will differ between different languages, thus affecting user satisfaction.

[0004] In view of the shortcomings of traditional practices, the applicant has been engaged in research in related fields for a long time and has taken a different approach to solve industry problems in the field of e-commerce technology. Summary of the invention

[0005] The primary purpose of the present application is to solve at least one of the above problems and provide a multilingual intelligent question answering method and its device, equipment, and medium.

[0006] In order to meet the various objectives of this application, this application adopts the following technical solutions:

[0007] A multilingual intelligent question answering method provided to meet one of the purposes of this application comprises the following steps:

[0008] Obtain the question text provided by the user and determine the question language of the question text;

[0009] When the question language is consistent with the knowledge language corresponding to the question-answering knowledge base, a pre-stored question in the question-answering knowledge base that matches the question text is determined, and the answer text corresponding to the pre-stored question is used as the target answer text;

[0010] When the question language is inconsistent with the knowledge language, the question text is translated into a target question text corresponding to the knowledge language, a pre-stored question matching the target question text is determined in the question-answering knowledge base, and the answer text corresponding to the pre-stored question is translated into a target answer text corresponding to the knowledge language;

[0011] The target answer text is sent to the user.

[0012] On the other hand, a multilingual intelligent question-answering device is provided to meet one of the purposes of the present application, comprising:

[0013] The question text acquisition module is used to obtain the question text provided by the user and determine the question language of the question text;

[0014] An answer text matching module is used to determine a pre-stored question in the question and answer knowledge base that matches the question text when the question language is consistent with the knowledge language corresponding to the question and answer knowledge base, and use the answer text corresponding to the pre-stored question as the target answer text;

[0015] A translation conversion module, for translating the question text into a target question text corresponding to the knowledge language when the question language is inconsistent with the knowledge language, determining a pre-stored question in the question-answering knowledge base that matches the target question text, and translating the answer text corresponding to the pre-stored question into a target answer text corresponding to the knowledge language;

[0016] The answer text sending module is used to send the target answer text to the user.

[0017] On the other hand, a computer device provided to meet one of the purposes of the present application includes a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the multilingual intelligent question-answering method described in the present application.

[0018] On the other hand, a computer program product provided to meet another purpose of the present application includes a computer program / instruction, which, when executed by a processor, implements the steps of the method described in any embodiment of the present application.

[0019] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0020] This application identifies the question language of the question text provided by the user, and determines whether the question language is consistent with the knowledge language in the question and answer knowledge base, and then determines whether the question text needs to be translated into a target question text corresponding to the knowledge language. Finally, the untranslated question text or the translated target question text is matched with the pre-stored questions in the question and answer knowledge base to determine the corresponding answer text and send it to the user as the target answer text. In this process, only one question and answer knowledge base in a specific language needs to be maintained, rather than multiple independent knowledge bases in different languages, thereby significantly reducing the management and maintenance workload of the question and answer knowledge base. Since the update and maintenance of a single knowledge base is more centralized and efficient, it can quickly respond to changes in new information and knowledge.

[0021] In addition, this application can quickly respond to user queries and improve overall question-answering efficiency through automated language recognition and translation processes. At the same time, it can adapt to the needs of different user groups and provide effective e-commerce customer service regardless of the language used. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0023] Figure 1 The network architecture of the e-commerce platform exemplified in this application;

[0024] Figure 2 A flowchart of a typical embodiment of the multilingual intelligent question-answering method of the present application;

[0025] Figure 3 This is a functional block diagram of the multilingual intelligent question-answering device of the present application;

[0026] Figure 4 A schematic diagram of the structure of a computer device used in this application. DETAILED DESCRIPTION

[0027] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0028] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0029] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

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

[0031] The exemplary e-commerce platform 82 provides supply and demand matching of products and / or services to the general public with the help of Internet infrastructure. In the e-commerce platform 82, products and / or services are provided as commodity information. To simplify the description, the concepts of commodity, product, etc. are used in this application to refer to the products and / or services in the e-commerce platform 82, which may specifically be physical products, digital products, tickets, service subscriptions, other offline services, etc.

[0032] In reality, all entities can access the e-commerce platform 82 as users, use various online services provided by the e-commerce platform 82, and achieve the purpose of participating in the business activities achieved 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, merchants and consumers in business activities, the e-commerce platform 82 has two types of users, merchant users and consumer users. All entities in the product distribution chain in business activities, including manufacturers, sellers, retailers, logistics providers, etc., can use online services in the e-commerce platform 82 as merchant users, while consumers in business activities, including real or potential consumers, can use online services in the e-commerce platform 82 as their corresponding consumer users. In actual business activities, the same entity can act as both a merchant user and a consumer user, and this should be understood flexibly.

[0033] The infrastructure used to deploy the e-commerce platform 82 mainly includes the backend architecture and frontend equipment. The backend architecture runs various online services through the service cluster, including middleware or frontend services for the platform, services for consumers, services for merchants, etc., to enrich and improve its service functions; the frontend equipment mainly covers the terminal devices used by users to access the e-commerce platform 82 as clients, including but not limited to various mobile terminals, personal computers, point-of-sale devices, etc. For example, merchant users can enter product information for their online stores through their terminal devices 80, or generate their product information using the interfaces opened by the e-commerce platform; consumer users can access the webpage of the online store implemented by the e-commerce platform 82 through their terminal devices 81, trigger the shopping process through the shopping buttons provided on the webpage, and call various online services provided by the e-commerce platform 82 during the shopping process, so as to achieve the purpose of shopping orders.

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

[0035] The e-commerce platform 82 can be implemented as online services such as cloud computing services, software as a service (SaaS), infrastructure as a service (IaaS), platform as a service (PaaS), desktop as a service (DaaS), hosted software as a service, mobile backend as a service (MBaaS), information technology management as a service (ITMaaS), etc. In some embodiments, the 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, for an online store, its administrator users enjoy the same or similar functions regardless of various embodiments such as iOS, Android, HomonyOS, or web pages.

[0036] The e-commerce platform 82 can realize the corresponding independent station for each merchant to run its corresponding online store, and provide the merchant with the corresponding business management engine instance for the merchant 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 various specific business processes of the online store can be configured through interfaces or built-in components to support the implementation of business activities. The independent station is the infrastructure of the e-commerce platform 82 with cross-border service functions. Merchants can maintain their online stores more centrally and autonomously based on the independent station. The independent station usually has a domain name and storage space dedicated to the merchant, and different independent stations are relatively independent. The e-commerce platform 82 can provide standardized or personalized technical support for a large number of independent stations, so that merchant users can customize their own business management engine instance and use this business management engine instance to maintain one or more online stores they own.

[0037] The online store can implement backend configuration and maintenance by having the merchant user log in to its business management engine instance as an administrator. 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 as an administrator, view various data, etc. For example, the merchant user can manage various aspects of its online store, such as viewing the online store's recent activities, updating the online store's product catalog, managing orders, recent visit activities, total order activities, etc.; the merchant user can also view more detailed information about the business and visitors to the merchant's online store by obtaining reports or metrics, such as showing a sales summary of the merchant's overall business, specific sales and participation data of active sales marketing channels, etc.

[0038] The e-commerce platform 82 may provide communication facilities and associated merchant interfaces for providing electronic communications and marketing, such as utilizing electronic message aggregation facilities to collect and analyze communication interactions between merchants, consumers, merchant devices, customer devices, point-of-sale devices, etc., aggregating and analyzing communications, such as for increasing the potential for providing product sales, etc. For example, a consumer may have a question related to a product, which may generate a dialogue between the consumer and the merchant (or an automated processor-based agent on behalf of the merchant), where the communication facility is responsible for interacting and providing the merchant with analysis on how to increase the probability of a sale.

[0039] In some embodiments, an application suitable for installation in a terminal device can be provided to serve the access needs of different users, so that various users can access the e-commerce platform 82 by running the application in the terminal device, such as the merchant backend module of the online store in the e-commerce platform 82, etc. In the process of implementing business activities through these functions, the e-commerce platform 82 can implement various functions related to supporting the implementation of business activities as middleware or online services and open corresponding interfaces, and then implant the toolkit corresponding to the interface access function into the application to implement function expansion 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 connections with customers in a flexible and transparent manner. Consumer users can select items online, create product orders, provide the delivery address of the goods in the product order, and complete the payment confirmation of the product order. Then, the merchant can review and complete or cancel the order. The review component carried by the business management engine instance can realize the compliance use of business processes to ensure that the order is suitable for fulfillment before actual fulfillment. Orders may sometimes be fraudulent and need to be verified (such as ID card checks). There is a payment method that requires merchants to wait to ensure that funds are received, which can play a role in preventing such risks, etc. Order risks may be generated by fraud detection tools submitted by third parties through order risk APIs, etc. Before fulfillment, merchants may need to obtain payment information or wait to receive payment information in order to mark the order as paid, so that the merchant can prepare to deliver the product. Such situations can be reviewed accordingly. The review process can be implemented by the fulfillment component. Merchants can use the fulfillment component to review, adjust work, and trigger related fulfillment services, such as: manual fulfillment services, which are used when merchants pick and pack products in boxes, purchase shipping labels and enter their tracking numbers, or simply mark items as fulfilled; custom fulfillment services, which can define email notifications; API fulfillment services, which can trigger third-party applications to create fulfillment records in third parties; legacy fulfillment services, which can trigger custom API calls from the business management engine to third parties; gift card fulfillment services. You can provide the generation of numbers and activate gift cards. Merchants can use the order printer application to print shipping orders. The fulfillment process can be executed when the items are packed in boxes and ready for shipping, tracking, delivery, consumer receipt verification, etc.

[0041] It can be seen that the services provided by the e-commerce platform are based on products. The corresponding commodity data is the basic data of the e-commerce platform. The commodity information is provided through the commodity data. The mining and utilization of commodity data is the basis for realizing various technical services, including the use of user transaction data and commodity data in the commodity data of the e-commerce platform to provide basic services for the operation of the data processing system. Therefore, the data processing system can run on any one or more servers of the e-commerce platform's cluster, so as to realize various functions by using the various commodity data provided by the e-commerce platform.

[0042] A multilingual intelligent question-answering method of the present application can be programmed as a computer program product and deployed in a client or server for execution. For example, in the exemplary application scenario of the present application, it can be deployed and implemented in a server of an e-commerce customer service platform, thereby accessing an interface opened after the computer program product is running and performing human-computer interaction with the process of the computer program product through a graphical user interface to execute the method.

[0043] See also Figure 2 The multilingual intelligent question answering method of the present application, in its typical embodiment, comprises the following steps:

[0044] Step S5100: Obtain the question text provided by the user and determine the question language of the question text;

[0045] E-commerce platforms can provide e-commerce customer service services by providing users with multiple entrances for using e-commerce customer service services, for example: integrating real-time chat windows, customer service links, customer service icons, customer service buttons, etc. in the website or mobile application that carries the e-commerce platform. Users can use the e-commerce customer service service through any of the said entrances to consult the corresponding e-commerce customer service. When users use the e-commerce customer service service, the e-commerce platform creates a conversation window between the user and the e-commerce customer service so that the user can consult the questions they want to know through the conversation window.

[0046] In one embodiment, a plurality of preset question texts are provided in the conversation window. The question texts are pre-set by the operator of the e-commerce platform by guessing the questions that the user may ask. If the question text contains the questions that the user wants to ask, the user can directly select from these preset question texts without the user having to reorganize the language to ask the question, which saves the user's time cost and improves the conversation efficiency. Figure 1 When the question text is correct, the user can also re-edit the question text by manually editing and inputting the question text or inputting the question text by language, etc. At this time, the conversation window will monitor the conversation between the user and the e-commerce customer service in real time to obtain the question text provided by the user.

[0047] In another embodiment, when the question text is input by the user, due to the different grammatical structures, vocabulary expressions and language habits of different languages, only by accurately identifying the question language can the corresponding language be selected to answer the user's question. In order to overcome this problem, the e-commerce platform can adopt a variety of technical means to identify the language of the question text. In an exemplary example, an algorithm for language identification can be built into the e-commerce platform. After obtaining the question text provided by the user, the built-in algorithm can directly identify the language of the question text; in another exemplary example, a trained language recognition model is used. The model is trained with a large amount of multilingual text data and can quickly determine the language of the question text according to the characteristics of the question text (such as character frequency, vocabulary distribution, grammatical structure, etc.); in another exemplary example, the question text is input into a large language model, and the language recognition instruction is used to control the large language model to recognize the question text. These methods ensure that the language of the question text can be accurately identified, so that the corresponding language can be selected to answer the questions raised by the user.

[0048] Step S5200: When the question language is consistent with the knowledge language corresponding to the question-answering knowledge base, a pre-stored question matching the question text in the question-answering knowledge base is determined, and the answer text corresponding to the pre-stored question is used as the target answer text;

[0049] The Q&A knowledge base is a database preset in the e-commerce platform, which contains a large number of pre-stored questions and their corresponding answer texts. The Q&A knowledge base is formed by collecting and organizing common questions and standard answers in the field of e-commerce customer service, aiming to provide corresponding answers to users' inquiries. The content in the Q&A knowledge base can come from historical customer consultation records, FAQ manuals, product usage guides, terms of service, and other relevant customer service documents. The content has been reviewed and edited to ensure the accuracy and reliability of the answer text. It should be noted that the Q&A knowledge base only has pre-stored questions and their corresponding answer texts corresponding to one specific language.

[0050] The question language of the question text determined in the previous step is compared with the knowledge language corresponding to the question-answering knowledge base. If the two languages ​​are consistent, the question text in the question language is directly used as the query basis. In one embodiment, before the answer matching operation is performed on the question text, the question text needs to be rewritten and standardized. This process is mainly to extract the core intention of the question text (i.e., understand the user's intention) and convert the question text into a more standardized text that is closer to the pre-stored question format in the question-answering knowledge base. Natural language processing technology can be used to understand the semantics of the question text, and then the text is adjusted to eliminate informal expressions and grammatical errors, ensuring that the rewritten question text is more in line with the standards of written language while maintaining the original meaning. In addition, the key words in the rewritten question text are extracted and standardized, such as unifying terms, removing unnecessary stop words or word form restoration, etc., so as to more accurately match the pre-stored questions in the question-answering knowledge base. In another embodiment, a large language model is called to rewrite the question text. For specific operations, please refer to the subsequent specific implementation methods, which will not be repeated here.

[0051] Based on the rewritten and standardized question text, it is matched with the pre-stored questions in the question and answer knowledge base. After determining the pre-stored questions that match the question text in the question and answer knowledge base, the answer text corresponding to the pre-stored question is used as the target answer text. In one embodiment, the e-commerce platform adopts keyword matching. First, the rewritten and standardized question text is subjected to keyword extraction, and the core words and phrases in the question text are identified using natural language processing technology. These keywords are the core intentions of the user's questions and can directly reflect the user's needs. Then the extracted keywords are matched with the pre-stored questions in the question and answer knowledge base. Each pre-stored question in the question and answer knowledge base is pre-marked with several keywords, which are the core elements of these pre-stored questions. By calculating the overlap between the keywords in the question text and the keywords in the pre-stored questions, it is determined whether the two match. The higher the overlap, the higher the match between the question text and the pre-stored questions.

[0052] In another embodiment, the e-commerce platform adopts a matching algorithm based on semantic similarity. The algorithm first converts the rewritten and standardized question text and the pre-stored questions in the question and answer knowledge base into vector representations. Specifically, the text is encoded using a pre-trained language model (such as BERT, GPT, etc.) to generate a high-dimensional semantic vector. The generated semantic vector captures the deep semantic information of the question text and can better reflect the similarity between texts. Then the cosine similarity between the question text vector and each pre-stored question vector in the question and answer knowledge base is calculated. The higher the cosine similarity value, the closer the semantics of the two texts are. The highest similarity value calculated is compared with the preset similarity threshold. If the highest similarity value calculated is higher than the similarity threshold, the answer text corresponding to the pre-stored question is extracted from the question and answer knowledge base and returned to the user as the target answer text. Among them, the similarity threshold is set by those skilled in the art based on experience or experimental data. In order to deal with the polysemous or ambiguous problems that may exist in the question and answer knowledge base, a multi-round dialogue mechanism can also be introduced to clarify the intention through further interaction with the user to ensure that the answer text finally returned can accurately meet the user's needs. Through this matching algorithm based on semantic similarity, e-commerce platforms can efficiently and accurately find answer texts that match user question texts from the question-and-answer knowledge base, thereby improving user consultation satisfaction and efficiency.

[0053] Step S5300: When the question language is inconsistent with the knowledge language, the question text is translated into a target question text corresponding to the knowledge language, a pre-stored question matching the target question text is determined in the question-answer knowledge base, and the answer text corresponding to the pre-stored question is translated into a target answer text corresponding to the knowledge language;

[0054] If the question language is inconsistent with the knowledge language, the e-commerce platform first needs to translate the question text provided by the user into the language corresponding to the question and answer knowledge base, that is, the knowledge language. In one embodiment, the e-commerce platform uses a neural network-based translation model to translate the question text. The translation model is trained through a large number of bilingual parallel corpus data sets and can learn the mapping relationship between different languages. When the question language of the question text input by the user is inconsistent with the knowledge language of the question and answer knowledge base, the e-commerce platform will input the question text into the pre-trained translation model, and the translation model will automatically translate the question text in the question language into the target question text corresponding to the knowledge language according to the knowledge language. Further, in order to ensure the accuracy of the translation, the translation model will combine the contextual information and conversation context in the conversation during the translation process to keep the core intention of the translated target question text unchanged. In addition, in order to deal with the ambiguity and polysemy problems that may occur during the translation process, the translation model uses an attention mechanism to focus on the key parts of the input question text to ensure the accuracy and fluency of the translation result. In another embodiment, the large language model can be controlled to translate the question text through corresponding translation prompt instructions, that is, the question text in the question language is translated into the target question text in the knowledge language. For specific steps, please refer to the subsequent specific implementation method, which will not be repeated here.

[0055] The target question text after translation is rewritten and standardized, and then similarity matching is performed with each pre-stored question in the question and answer knowledge base to determine the pre-stored question that matches the target question text, and then the answer text corresponding to the pre-stored question can be determined. After determining the answer text corresponding to the pre-stored question, since the language of the user's question is inconsistent with the language of the question and answer knowledge base, the extracted answer text also needs to be translated back to the language of the user's question, that is, the target answer text. In one embodiment, the e-commerce platform uses the same neural network-based translation model as the question text translation to implement the translation. The translation model automatically translates the answer text in the knowledge language into the target answer text in the question language according to the knowledge language of the target answer text and the language of the user's question. In another embodiment, the e-commerce platform can control the large language model to translate the answer text through corresponding translation prompt instructions. The large language model will translate the answer text in the knowledge language into the target answer text in the user's question language according to the translation prompt instruction. For specific steps, please refer to the subsequent specific implementation method, which will not be repeated here.

[0056] Through the above steps, the e-commerce platform can efficiently and accurately complete the translation, matching and translation of question texts and answer texts when the question language is inconsistent with the knowledge language, thereby providing users with satisfactory answers and improving user consultation satisfaction and efficiency.

[0057] Step S5400: Send the target answer text to the user.

[0058] After determining the target answer text, the e-commerce platform will send the target answer text to the user to answer the user's consultation questions. In one embodiment, the e-commerce platform displays the target answer text to the user in real time through a conversation window. The conversation window is an interface for users to interact with e-commerce customer service, and users can view the e-commerce platform's reply in this window. When the target answer text is generated, the e-commerce platform will display it directly in the conversation window, and users can see the answer immediately. In order to improve the user experience, the e-commerce platform can also format the target answer text, such as adding titles, paragraphs, bold keywords, etc., to make the target answer text clearer and easier to read.

[0059] In another embodiment, the e-commerce platform can send the target answer text to the user via email or text message. For example, when a user submits a question on the e-commerce platform website but does not log in or use the real-time chat function, the e-commerce platform can send the target answer text to the user via the email address or mobile phone number provided by the user in the form of email or text message. This method is suitable for situations where it is inconvenient for the user to use the real-time chat function or the user wants to receive the answer offline.

[0060] In order to ensure that users can receive the answer text in time, the e-commerce platform can also introduce a push notification mechanism. For example, when the target answer text is generated, the e-commerce platform can send a notification to the user's mobile device through the push notification function of the mobile application, informing the user that the answer is ready and providing a link or button that the user can click to directly view the answer text. This method can improve the user's response speed and ensure that the user can obtain the required information in time.

[0061] Through the above steps, the e-commerce platform can efficiently and accurately send the target answer text to the user, answer the user's consultation questions, and improve the user's consultation satisfaction and efficiency.

[0062] It can be known from the typical embodiments of the present application that the technical solution of the present application has many advantages, including but not limited to the following aspects:

[0063] This application identifies the question language of the question text provided by the user, and determines whether the question language is consistent with the knowledge language in the question and answer knowledge base, and then determines whether the question text needs to be translated into the target question text corresponding to the knowledge language. Finally, the untranslated question text or the translated target question text is matched with the pre-stored questions in the question and answer knowledge base, and the corresponding answer text is determined and sent to the user as the target answer text. Specifically, this application significantly improves the translation accuracy and matching efficiency of the e-commerce customer service by translating the question text provided by the user into the target question text corresponding to the knowledge language, and rewriting and similarity matching. By using a large language model for translation and rewriting, the e-commerce platform can efficiently and accurately translate the question text provided by the user into the language corresponding to the question and answer knowledge base, and further highlight the core intention of the user. By matching the rewritten target question text with the pre-stored questions in the question and answer knowledge base, the corresponding answer text is finally determined, and the answer text is translated into the language of the user's question. The implementation of this step can ensure that the user can accurately understand the answer content without having to translate it by himself, thereby improving user satisfaction.

[0064] In this process, only one question-and-answer knowledge base in a specific language needs to be maintained, rather than multiple independent knowledge bases in different languages, which significantly reduces the management and maintenance workload of the question-and-answer knowledge base. Since the update and maintenance of a single knowledge base is more centralized and efficient, it can quickly respond to changes in new information and knowledge.

[0065] In addition, this application can quickly respond to user queries and improve overall question-answering efficiency through automated language recognition and translation processes. At the same time, it can adapt to the needs of different user groups and provide effective e-commerce customer service regardless of the language used.

[0066] In a further embodiment, obtaining the question text provided by the user includes the following steps:

[0067] Step S5110: obtaining an initial question text and a previous question text provided by a user, wherein the previous question text is obtained based on the corresponding initial question text in a previous round of question-answering;

[0068] In multiple rounds of dialogue, the e-commerce platform needs to obtain the initial question text and the previous question text provided by the user in order to better understand the user's intention and context information. Among them, a round of question and answer contains a question text of the user and a corresponding answer text of the e-commerce customer service. The initial question text is the question text entered by the user in the current round, and the previous question text is the text asked by the user in the previous round of question and answer that matches the pre-stored question. The previous question text is obtained based on the initial question text of the previous round, that is, the question text used by the user in the previous round of question and answer to match the pre-stored question is a further standardization of the initial question text.

[0069] In order to obtain the initial question text and the previous question text, the e-commerce platform will monitor the conversation window between the user and the e-commerce customer service in real time. When the user enters a new question text in the conversation window, the e-commerce platform will identify it as the initial question text. At the same time, the e-commerce platform will also record the user's question text standardized based on the corresponding initial question text in the previous round of questions and answers, and use it as the previous question text for use in subsequent specific implementation methods.

[0070] Step S5120: When the correlation between the initial question text and the previous question text is greater than a preset threshold, the initial question text and the previous question text are combined to generate a question text corresponding to the current round of question and answer;

[0071] In a multi-round dialogue, the e-commerce platform needs to determine the correlation between the initial question text and the previous question text to determine whether it is necessary to combine the previous question text to generate the question text of the current round (i.e., this round). Correlation refers to the semantic similarity and logical correlation between the initial question text and the previous question text. In one embodiment, the text is first encoded using a pre-trained language model (such as BERT, GPT, etc.) to generate a high-dimensional semantic vector, in which similar texts are closer in the vector space. Then, a measurement method such as cosine similarity is used to calculate the similarity between the two semantic vectors, and its value ranges from -1 to 1. The closer the value is to 1, the higher the correlation between the two texts. The calculated similarity value can be directly used as the correlation between the initial question text and the previous question text. Then a preset threshold is set. When the calculated correlation exceeds the preset threshold, the e-commerce platform determines that the question text currently input by the user is related to the question text of the previous round, that is, it is necessary to combine the previous question text to generate the question text of the current round. Among them, the threshold is set by a technician in this field as needed. Specifically, the e-commerce platform will process the initial question text and the previous question text to generate a comprehensive question text. This comprehensive question text will contain the contextual information of the previous question text to better understand the user's intention. Specifically, the e-commerce platform will extract key words and phrases from the initial question text and the previous question text. These key words and phrases contain the core intention and important information of the question. Then, the e-commerce platform will generate a comprehensive question text based on these key words and phrases. The comprehensive question text will retain the core intention of the initial question text and combine the contextual information of the previous question text to ensure that the generated text can accurately reflect the user's intention. After generating the comprehensive question text, the e-commerce platform will use it as the question text of the current round for subsequent question and answer processing.

[0072] Step S5130: When the correlation between the initial question text and the previous question text is less than a preset threshold, the initial question text is used as the corresponding question text for this round of question and answer.

[0073] When the correlation between the initial question text and the previous question text is less than the preset threshold, the e-commerce platform will consider that the question text entered by the user in this round is not related to the question text in the previous round, so there is no need to combine the previous question text to generate the question text of the current round. At this time, the e-commerce platform will use the initial question text as the question text of the current round and ignore the influence of the previous question text.

[0074] In this embodiment, by obtaining the initial question text and the previous question text provided by the user, and generating the question text of the current round based on the correlation between the two, the e-commerce platform's ability to understand and respond in multiple rounds of dialogue is significantly improved. When the correlation between the initial question text and the previous question text is greater than a preset threshold, a comprehensive question text is generated by combining the two. This method helps to preserve the continuity of the dialogue and ensure that the answer can be made in accordance with the user's complete intention. This not only improves the accuracy of the answer, but also reduces the need for users to repeat questions or provide additional contextual information, thereby enhancing the user experience.

[0075] In a further embodiment, when the question language is inconsistent with the knowledge language, the question text is translated into a target question text corresponding to the knowledge language, a pre-stored question matching the target question text is determined in the question-answer knowledge base, and the answer text corresponding to the pre-stored question is translated into a target answer text corresponding to the knowledge language, including the following steps:

[0076] Step S5310: When the question language is inconsistent with the knowledge language, embed the question text into a preset first translation prompt template, obtain a first translation prompt instruction and input it into the large language model to control the large language model to translate the question text in the question language into a target question text corresponding to the knowledge language;

[0077] When the question language provided by the user is inconsistent with the knowledge language of the question-answering knowledge base, the e-commerce platform needs to translate the question text in the question language input by the user into the target question text corresponding to the knowledge language, so as to match it with the pre-stored questions in the knowledge language in the question-answering knowledge base. The e-commerce platform adopts a translation method based on a large language model, such as a translation agent (an intelligent agent that can complete certain specific tasks autonomously), the principle of which is to control the large language model to generate specific outputs by using different prompts. Specifically, the e-commerce platform will embed the question text provided by the user into a preset first translation prompt template to generate a first translation prompt instruction. The first translation prompt template contains some guiding sentences and placeholders to indicate how the large language model processes the input text.

[0078] For example, the first translation prompt template can be designed as: "Please translate the following question text from [question language] to [knowledge language]: [question text]". In the first translation prompt template, "[question language]" and "[knowledge language]" are placeholders for specifying the language of the input text and the target language, while "[question text]" is the question text provided by the user. By embedding the question text into the first translation prompt template, the e-commerce platform can generate a complete first translation prompt instruction, for example: "Please translate the following question text from English to Chinese: How do I return a product?".

[0079] The e-commerce platform will input the generated first translation prompt instruction into the pre-trained large language model. The large language model will automatically translate the question text in the question language into the target question text corresponding to the knowledge language based on the language information and question text in the first translation prompt instruction. Since the large language model has powerful natural language understanding and generation capabilities, it can capture the deep semantic information of the question text and generate an accurate and fluent target question text.

[0080] Through this translation method based on a large language model, the e-commerce platform can efficiently and accurately translate the question text provided by users into the language corresponding to the question and answer knowledge base, laying the foundation for the subsequent matching and answering process.

[0081] Step S5320: embed the target question text into a preset rewriting prompt template, and input the obtained first rewriting prompt instruction into the large language model to rewrite the target question text to further highlight the user's intention;

[0082] In order to ensure that the pre-stored questions in the question-answering knowledge base can accurately match the user's question text, the e-commerce platform needs to further rewrite and optimize the translated target question text. The purpose is to refine the core intent of the target question text and convert it into a more standardized text that is closer to the format of the pre-stored questions in the question-answering knowledge base. The e-commerce platform adopts a rewriting method based on a large language model, such as a rewriting agent, whose function is to perform a specific rewriting operation on the input target question text. Specifically, the e-commerce platform embeds the translated target question text into a preset rewriting prompt template and generates a first rewriting prompt instruction. The rewriting prompt template contains some guiding sentences and placeholders to indicate how the large language model processes the input text. For example, the rewriting prompt template can be designed as: "Please rewrite the following question text into [pre-set format] to highlight the user's core intent: [target question text]". In the rewriting prompt template, the preset format is obtained based on the format of the pre-stored questions. "[Target question text]" is the translated question text. By embedding the target question text into the rewriting prompt template, the e-commerce platform can generate a complete first rewriting prompt instruction.

[0083] The e-commerce platform will input the generated first rewriting prompt instruction into the pre-trained large language model. The large language model will automatically generate a rewritten question text based on the sentence in the first rewriting prompt instruction and the target question text. Through this rewriting method based on the large language model, the e-commerce platform can efficiently and accurately rewrite the translated target question text, further highlight the user's core intention, and improve the accuracy and efficiency of question answering.

[0084] Step S5330: perform similarity matching on the rewritten target question text and each pre-stored question in the question-answer knowledge base, determine the pre-stored question that matches the target question text, and then determine the answer text corresponding to the pre-stored question;

[0085] The e-commerce platform will perform similarity matching between the rewritten target question text and each pre-stored question in the question-and-answer knowledge base. This is done to find the pre-stored question that is semantically closest to the target question text and extract its corresponding answer text. This can be achieved by using a matching algorithm based on semantic similarity. Specifically, the e-commerce platform will convert the rewritten target question text and all pre-stored questions in the question-and-answer knowledge base into vector representations. The text is encoded using a pre-trained language model (such as BERT, GPT, etc.) to generate a high-dimensional semantic vector. The cosine similarity between the rewritten target question text vector and each pre-stored question vector in the question-and-answer knowledge base is then calculated. By calculating the cosine similarity, the e-commerce platform can determine the pre-stored question that is semantically closest to the target question text.

[0086] In order to ensure the accuracy of the match, the e-commerce platform will set a similarity threshold. When the cosine similarity between the target question text and a pre-stored question exceeds the threshold, the two are determined to be matched successfully. In one embodiment, after the match is successful, the e-commerce platform will extract multiple successfully matched answer texts from the question-and-answer knowledge base. First, the answer text corresponding to the pre-stored question with the highest similarity is sent to the user, and a reply satisfaction window pops up at the same time. The user can choose to evaluate the satisfaction of the target answer text by clicking, etc. If the user's choice is characterized by dissatisfaction with the answer, the answer text of the pre-stored question with the second highest similarity will be pushed to the user as the target answer text. If the user chooses to be dissatisfied with the answer again, the answer text of the pre-stored question with the third highest similarity will be pushed to the user as the target answer text; and so on.

[0087] Through this matching algorithm based on semantic similarity, the e-commerce platform can efficiently and accurately find pre-stored questions that match the rewritten target question text from the question-and-answer knowledge base, and extract the corresponding answer text, thereby improving user consultation satisfaction and efficiency.

[0088] Step S5340: embed the answer text into a preset second translation prompt template, obtain a second translation prompt instruction and input it into the large language model to control the large language model to translate the answer text in the knowledge language into the target answer text corresponding to the question language.

[0089] After determining the answer text, the e-commerce platform needs to translate the answer text back to the question language corresponding to the question text so that the user can accurately understand the answer content. Here, the large language model can use the translation agent. It should be noted that the translation agent is different from the target question text in the above step by translating the question text in the question language into the knowledge language. The essential difference is that the prompts of the two are different, that is, the generated translation prompt instructions are different. The e-commerce platform will embed the answer text in the question and answer knowledge base into a preset second translation prompt template to generate a second translation prompt instruction, for example: "Please translate the following answer text from Chinese to English: You can return the goods by following the steps below.".

[0090] The second translation prompt instruction will control the large language model to translate the answer text in the knowledge language into the target answer text corresponding to the question language. For example, if the user asks a question in English and the answer text is in Chinese, the second translation prompt instruction will instruct the large language model to translate the Chinese answer text into English. Through this translation method based on the large language model, the target answer text that the user can understand is translated.

[0091] In this embodiment, by translating the question text provided by the user into the target question text corresponding to the knowledge language, and performing rewriting and similarity matching, the translation accuracy and matching efficiency of the e-commerce customer service are significantly improved. By using a large language model for translation and rewriting, the question text provided by the user can be efficiently and accurately translated into the language corresponding to the question and answer knowledge base, and the core intention of the user can be further highlighted. Through a matching algorithm based on semantic similarity, the pre-stored questions that match the rewritten target question text are accurately determined, and the corresponding answer text is extracted. Finally, by translating the answer text back to the language of the user's question, the e-commerce platform can ensure that the user can accurately understand the answer content without having to translate it by themselves, thereby improving user satisfaction.

[0092] In a further embodiment, the rewritten target question text is matched with each pre-stored question in the question-answer knowledge base for similarity, and a pre-stored question matching the target question text is determined, and then an answer text corresponding to the pre-stored question is determined, including the following steps:

[0093] Step S5331: input the target question text and each pre-stored question in the question-answer knowledge base into a pre-trained semantic vector model, and extract the semantic vectors of the target question text and each pre-stored question;

[0094] The target question text and each pre-stored question in the question-answering knowledge base are input into a pre-trained semantic vector model. The semantic vector model is a deep learning model, such as BERT, GPT or other similar pre-trained language models. The semantic vector model is trained with a large amount of text data and can capture the deep semantic information of the text. When the target question text and pre-stored questions are input into the semantic vector model, the semantic vector model encodes the text and generates a high-dimensional semantic vector. These semantic vectors capture the semantic features of the input text and can reflect the meaning and intent of the input text. In this way, the e-commerce platform can convert the text into a numerical representation, which is convenient for subsequent similarity calculations and matching operations.

[0095] Step S5332, calculating the similarity between the semantic vector of the target question text and the semantic vectors of each pre-stored question, and determining the pre-stored question with the highest similarity value;

[0096] Calculate the similarity between the semantic vector of the target question text and the semantic vector of each pre-stored question in the question-answering knowledge base. Common similarity calculation methods include cosine similarity, Euclidean distance, etc. Cosine similarity is a commonly used method that measures the similarity between two vectors by calculating the cosine value of the angle between them. The value range of cosine similarity is between -1 and 1, and the closer the value is to 1, the more similar the two vectors are. The e-commerce platform calculates the cosine similarity between the semantic vector of the target question text and the semantic vector of each pre-stored question one by one, and records the similarity value of each pre-stored question. Then compare these similarity values ​​to find the pre-stored question with the highest similarity value. In this way, the e-commerce platform can efficiently and accurately determine the pre-stored question that best matches the target question text.

[0097] Step S5333: When the highest similarity value is greater than a preset similarity threshold, the corresponding pre-stored question is determined to be a pre-stored question in the question-and-answer knowledge base that matches the target question text, and then the answer text corresponding to the pre-stored question is determined.

[0098] The calculated highest similarity value is compared with the preset similarity threshold. The similarity threshold is a preset value used to determine whether the match between the target question text and the pre-stored question is high enough. If the highest similarity value is greater than the preset similarity threshold, it is determined that the target question text matches the pre-stored question successfully. At this time, the e-commerce platform will extract the answer text corresponding to the pre-stored question from the question and answer knowledge base. The answer text is the standard answer pre-stored in the question and answer knowledge base, which has been reviewed and edited to ensure its accuracy and reliability. In this way, the e-commerce platform can efficiently and accurately determine the pre-stored questions that match the target question text, and extract the corresponding answer text, laying the foundation for the subsequent answer translation and sending process.

[0099] In this embodiment, the target question text and the pre-stored question are respectively input into the pre-trained semantic vector model, and the corresponding semantic vector is extracted, so that the text is converted into a numerical representation, which is convenient for subsequent similarity calculation and matching operations. By calculating the cosine similarity between the semantic vector of the target question text and the semantic vector of each pre-stored question, the pre-stored question that best matches the target question text can be determined efficiently and accurately. When the highest similarity value is greater than the preset similarity threshold, the answer text corresponding to the pre-stored question is extracted from the question-answering knowledge base to ensure its accuracy and reliability.

[0100] In a further embodiment, after calculating the similarity between the semantic vector of the target question text and the semantic vectors of each pre-stored question and determining the pre-stored question with the highest similarity value, the following steps are included:

[0101] Step S6100: When the highest similarity is less than a preset similarity threshold, the target question text is embedded in the answer generation template, and the generated answer generation instruction controls the large language model to generate a hypothetical answer text corresponding to the target question text;

[0102] When the highest similarity value between the target question text and the pre-stored questions in the question-answering knowledge base is less than the preset similarity threshold, it is determined that the target question text does not match any pre-stored questions in the question-answering knowledge base. At this time, the e-commerce platform needs to generate a hypothetical answer text. Specifically, the target question text is embedded in a preset answer generation template to generate an answer generation instruction. The answer generation template is used to instruct the large language model how to process the input text. For example, the answer generation template can be designed as: "Please generate a hypothetical answer based on the following question text: [target question text]". By embedding the target question text into the answer generation template, a complete answer generation instruction is generated. The e-commerce platform will input the generated answer generation instruction into the pre-trained large language model. To control the large language model to automatically generate a hypothetical answer text. The large language model captures the deep semantic information of the target question text and generates a reasonable and fluent hypothetical answer text for use in subsequent specific embodiments.

[0103] Step S6200: determine whether there is a matching answer text in the question-answer knowledge base for the assumed answer text. If there is, embed the matching answer text into a third translation prompt template, generate a third translation prompt instruction and input it into the large language model to control the large language model to translate the answer text in the knowledge language into a target answer text corresponding to the question language, and send the target answer text to the user.

[0104] After generating the hypothetical answer text, the hypothetical answer text is matched with each answer text in the question-answer knowledge base, and the matching method includes semantic matching, etc. Specifically, the e-commerce platform can quantify the similarity between the hypothetical answer text and each pre-stored answer text in the question-answer knowledge base. If the similarity between the hypothetical answer text and a pre-stored answer text in the question-answer knowledge base exceeds a preset threshold, the e-commerce platform will consider the hypothetical answer text to be valid and extract the pre-stored answer text that matches it. Then translate the matching answer text back to the language of the user's question by embedding the answer text into a preset third translation prompt template. The generated third translation prompt instruction controls the large language model to automatically translate the answer text in the knowledge language into the target answer text corresponding to the question language. Finally, the e-commerce platform will send the generated target answer text to the user. This step is to use the hypothetical answer text to match the answer text in the question-answer knowledge base when the user expression does not correspond to the expression of the pre-stored question, the context information is missing, the target question text is complex in intent, etc., resulting in the target question text not matching the pre-stored question in the question-answer knowledge base, which can solve the situation where the corresponding answer text cannot be found due to the above-mentioned situations.

[0105] Step S6300: If it does not exist, obtain the artificial answer text corresponding to the question text and send it to the user as the target answer text.

[0106] After determining that there is no matching answer text in the question-and-answer knowledge base for the hypothetical answer text. At this time, since there is no answer text matching the hypothetical answer text in the question-and-answer knowledge base, the e-commerce platform will turn to manual customer service to obtain the corresponding answer text. The user's question text is forwarded to the manual customer service team, who will provide detailed answers based on the user's specific questions. The manual customer service team combines the relevant information and knowledge base of the e-commerce platform to generate an accurate and detailed answer text. The answer text will be reviewed and confirmed by the manual customer service to ensure its accuracy and reliability. After obtaining the manual answer text, the e-commerce platform will send it to the user as the target answer text.

[0107] In this embodiment, when the target question text does not match the pre-stored questions in the question-and-answer knowledge base, the e-commerce platform can use the large language model to generate a reasonable hypothetical answer text, and determine whether it matches the answer text in the question-and-answer knowledge base through a semantic matching method. If there is a matching answer text, the e-commerce platform will translate it back to the language of the user's question and send it to the user, thereby quickly and accurately solving the user's problem. If the hypothetical answer text does not have a matching answer text in the question-and-answer knowledge base, the e-commerce platform will turn to manual customer service to obtain the answer to ensure that the user can get a satisfactory answer. This multi-level response mechanism not only improves the intelligence level of e-commerce customer service, but also enhances its adaptability and accuracy in complex situations, significantly improving the satisfaction and efficiency of user consultation.

[0108] In a further embodiment, after determining whether there is a matching answer text in the question-answer knowledge base, the following steps are included:

[0109] Step S7100: When the assumed answer text has a matching answer text in the question-answer knowledge base, the pre-stored question corresponding to the answer text and the question text are used to form a translation instance;

[0110] If the hypothetical answer text matches a pre-stored answer text in the question-answer knowledge base, the e-commerce platform determines that the hypothetical answer text is valid and extracts the pre-stored answer text that matches it. At this time, the e-commerce platform will form a translation instance with the pre-stored question corresponding to the pre-stored answer text and the question text provided by the user to provide data support for subsequent e-commerce translation model training.

[0111] Step S7200: Collect multiple translation instances to form a translation instance set. When the total number of translation instances in the translation instance set is greater than a preset number threshold, construct a fine-tuning training set based on the translation instance set.

[0112] After generating and recording multiple translation instances in the previous step, these translation instances are combined into a translation instance set. The translation instance set contains multiple translation instances, each of which contains the user's question text and pre-stored questions. The e-commerce platform will continue to monitor and update the translation instance set to ensure that it contains enough translation instances. When the total number of translation instances in the translation instance set is greater than the preset quantity threshold, it can be used to construct a fine-tuning training set. The preset quantity threshold is set by technicians in this field based on experience or experimental data to ensure the scale and quality of the fine-tuning training set. In this way, the e-commerce platform can use rich translation examples to provide strong support for subsequent e-commerce translation model training.

[0113] Step S7300: Use the fine-tuning training set to train a preset e-commerce translation model to obtain an e-commerce translation model that has been trained to convergence, which is used to translate the question text into a target question text corresponding to the knowledge language.

[0114] After constructing the fine-tuning training set, the e-commerce platform will use the fine-tuning training set to train the preset e-commerce translation model. The e-commerce translation model is a translation model based on a neural network, such as a Transformer model. The e-commerce platform will input the translation examples in the fine-tuning training set into the e-commerce translation model for supervised training. During the training process, the e-commerce translation model learns how to translate the question text into the target question text corresponding to the knowledge language based on the input question text and the corresponding pre-stored questions. Through continuous iterative training, the e-commerce translation model will gradually optimize its translation capabilities until the model converges, that is, the performance of the model is no longer significantly improved. The e-commerce translation model trained to convergence can execute step S5300 to translate the question text into the target question text corresponding to the knowledge language. In this way, the e-commerce platform can continuously improve the performance of the e-commerce translation model to ensure that it can provide high-quality translation services in practical applications.

[0115] In one embodiment, the e-commerce translation model is pre-trained, and the fine-tuning training set is added to the original training samples to increase the number and diversity of the original training samples, so that the e-commerce translation model can better adapt to the complexity and variability in actual applications. During the fine-tuning process, the e-commerce translation model further optimizes its parameters according to the new training set to improve the accuracy and relevance of the translation. The e-commerce translation model can not only maintain the general translation ability learned in the pre-training stage, but also enhance its adaptability to newly emerging expressions, thereby providing more accurate and efficient translation services in actual e-commerce customer service scenarios.

[0116] In this embodiment, if it is assumed that the answer text matches the answer text in the question and answer knowledge base, it means that the target question text does not match the pre-stored question. This may be due to the emergence of new expressions in the e-commerce customer service field or the instability of the trained e-commerce translation model, resulting in inaccurate translation when translating the question text into the target question text. It should be noted that the pre-stored question corresponding to the answer text is the standard text after the question text is translated. By collecting these translation examples and forming a translation example set, it can be used to train the e-commerce translation model, thereby improving the accuracy of the e-commerce translation model. By optimizing the e-commerce translation model, question and answer errors caused by inaccurate translation can be reduced, and the overall efficiency and user satisfaction of e-commerce customer service can be improved.

[0117] In a further embodiment, before obtaining the question text provided by the user, the following steps are included:

[0118] Step S5000: Acquire a multilingual parallel corpus dataset in the field of e-commerce customer service, wherein the multilingual parallel corpus dataset includes a plurality of parallel corpus data pairs, wherein the parallel corpus data pairs include question texts in different languages ​​collected in the field of e-commerce customer service and corresponding translation texts in knowledge languages;

[0119] Obtain multilingual parallel corpus datasets in the field of e-commerce customer service. These parallel corpus datasets contain question texts in different languages ​​collected in the field of e-commerce customer service and their corresponding translation texts in knowledge languages. Each parallel corpus data pair consists of a question text and its corresponding translation text, ensuring that the translation text is an accurate translation of the knowledge language. E-commerce platforms can obtain these parallel corpus data through a variety of channels, such as extracting from historical customer consultation records, FAQ manuals, product usage guides, terms of service, and other relevant customer service documents. In addition, e-commerce platforms can also translate the question texts raised by users into knowledge languages ​​through manual annotation to form new parallel corpus data pairs. By collecting and organizing these multilingual parallel corpus data, e-commerce platforms can build a rich dataset to provide strong support for subsequent e-commerce translation model training.

[0120] Step S5001: input the multilingual parallel corpus dataset into the e-commerce translation model for training, and use the translated text of the knowledge language as a supervision label to train the e-commerce translation model to a convergence state. The e-commerce translation model is used to translate the question text into a target question text corresponding to the knowledge language.

[0121] After obtaining a multilingual parallel corpus dataset, the multilingual parallel corpus dataset is input into a preset e-commerce translation model for training. The e-commerce translation model is a neural network-based model. The e-commerce translation model is pre-trained with a large number of bilingual parallel corpus datasets and is able to learn the mapping relationship between different languages. Through continuous iterative training, the e-commerce translation model will gradually optimize its translation capabilities and learn how to translate the question text into the target question text corresponding to the knowledge language. The training process will continue until the performance of the e-commerce translation model is no longer significantly improved, that is, the model reaches a convergence state. In particular, the e-commerce translation model is used to translate the question text into the target question text corresponding to the knowledge language in step S5300.

[0122] Furthermore, after the e-commerce translation model training is completed, the e-commerce translation model will continue to be continuously trained and updated with the help of the embodiment constituted by steps S7100 to S7300. This process further trains the e-commerce translation model by using a fine-tuning training set, which can continuously improve its ability to accurately translate question texts and realize the iterative upgrade mechanism of the e-commerce translation model. This continuous optimization and updating ensures that the e-commerce translation model can adapt to the ever-changing language usage habits and the latest developments in the e-commerce field, thereby maintaining its translation accuracy and relevance in the long term. Through this training method that combines pre-training and continuous fine-tuning, the e-commerce translation model can play a more stable and efficient role and provide users with more accurate services.

[0123] In this embodiment, by using a multilingual parallel corpus data set specific to the e-commerce customer service field for training, the e-commerce translation model can learn e-commerce-specific terms and expressions, thereby improving the accuracy and relevance of translation. Specifically, using a multilingual parallel corpus data set specific to the e-commerce customer service field for training, the e-commerce translation model can deeply learn and master the professional terms and expressions in the e-commerce field. This training method enables the e-commerce translation model to more accurately identify and translate e-commerce-specific vocabulary and phrases, such as "discount code", "size guide", "return and exchange policy", etc. when handling customer inquiries, ensuring that these professional terms are accurately converted between different languages. At the same time, the e-commerce translation model can better understand the context and context in e-commerce conversations by learning a large number of actual conversations, so as to maintain the intention and nuances of the original text during translation, and avoid misunderstandings caused by cultural differences or improper understanding of the context. Targeted training helps the model to provide more natural and close translation results that are close to the habits of target language users when facing complex e-commerce scenarios, such as promotional activities, product feature descriptions, customer reviews, etc.

[0124] See also Figure 3, a multilingual intelligent question-answering device provided to meet one of the purposes of the present application is a functional embodiment of the multilingual intelligent question-answering method of the present application. On the other hand, the device is a multilingual intelligent question-answering device provided to meet one of the purposes of the present application, including a question text acquisition module 5100, an answer text matching module 5200, a translation conversion module 5300 and an answer text sending module 5400, wherein the question text acquisition module 5100 is used to obtain the question text provided by the user and determine the question language of the question text; the answer text matching module 5200 is used to match the question language with the answer text when the question language is When the knowledge language corresponding to the question and answer knowledge base is consistent, the pre-stored question in the question and answer knowledge base that matches the question text is determined, and the answer text corresponding to the pre-stored question is used as the target answer text; the translation conversion module 5300 is used to translate the question text into the target question text corresponding to the knowledge language when the question language is inconsistent with the knowledge language, determine the pre-stored question in the question and answer knowledge base that matches the target question text, and translate the answer text corresponding to the pre-stored question into the target answer text corresponding to the knowledge language; the answer text sending module 5400 is used to send the target answer text to the user.

[0125] In a further embodiment, the question text acquisition module 5100 includes: a question text acquisition submodule, which is used to acquire an initial question text and a previous question text provided by a user, wherein the previous question text is obtained in a previous round of question and answer based on the corresponding initial question text; an associated text generation submodule, which is used to combine the initial question text and the previous question text to generate a question text corresponding to this round of question and answer when the degree of association between the initial question text and the previous question text is greater than a preset threshold; and a question text determination submodule, which is used to use the initial question text as the corresponding question text for this round of question and answer when the degree of association between the initial question text and the previous question text is less than a preset threshold.

[0126] In a further embodiment, the translation conversion module 5300 includes: a question text translation submodule, which is used to embed the question text into a preset first translation prompt template when the question language is inconsistent with the knowledge language, obtain a first translation prompt instruction and input it into the large language model to control the large language model to translate the question text in the question language into a target question text corresponding to the knowledge language; a question text rewriting submodule, which is used to embed the target question text into a preset rewriting prompt template, obtain a first rewriting prompt instruction and input it into the large language model to rewrite the target question text to further highlight the user's intention; a similarity matching submodule, which is used to perform similarity matching on the rewritten target question text and each pre-stored question in the question-answering knowledge base, respectively, to determine the pre-stored question matching the target question text, and then determine the answer text corresponding to the pre-stored question; an answer text translation submodule, which is used to embed the answer text into a preset second translation prompt template, obtain a second translation prompt instruction and input it into the large language model to control the large language model to translate the answer text in the knowledge language into the target answer text corresponding to the question language.

[0127] In a further embodiment, the similarity matching submodule includes: a semantic vector extraction submodule, which is used to input the target question text and each pre-stored question in the question and answer knowledge base into a pre-trained semantic vector model, respectively, to extract the semantic vectors of the target question text and each pre-stored question; a similarity value calculation submodule, which is used to calculate the similarity value between the semantic vector of the target question text and the semantic vectors of each pre-stored question, and determine the pre-stored question with the highest similarity value; a pre-stored question determination submodule, which is used to determine that the corresponding pre-stored question is a pre-stored question in the question and answer knowledge base that matches the target question text when the highest similarity value is greater than a preset similarity threshold, and then determine the answer text corresponding to the pre-stored question.

[0128] In a further embodiment, after the similarity value calculation submodule, it includes: a hypothetical answer generation submodule, which is used to embed the target question text into the answer generation template when the highest similarity is less than a preset similarity threshold, and the generated answer generation instruction controls the large language model to generate a hypothetical answer text corresponding to the target question text; an answer text matching submodule, which is used to determine whether there is a matching answer text for the hypothetical answer text in the question and answer knowledge base, and if so, embed the matching answer text into the third translation prompt template, generate a third translation prompt instruction and input it into the large language model to control the large language model to translate the answer text in the knowledge language into the target answer text corresponding to the question language, and send the target answer text to the user; a manual reply submodule, which is used to obtain the manual answer text corresponding to the question text when it does not exist, and send it to the user as the target answer text.

[0129] In a further embodiment, after the answer text matching submodule, it includes: a translation instance construction submodule, which is used to, when it is assumed that there is a matching answer text in the question and answer knowledge base, construct a translation instance with the pre-stored question and question text corresponding to the answer text; a fine-tuning training set construction submodule, which is used to collect multiple translation instances to form a translation instance set, and when the total number of translation instances in the translation instance set is greater than a preset number threshold, construct a fine-tuning training set based on the translation instance set; a model training submodule, which is used to use the fine-tuning training set to train a preset e-commerce translation model to obtain an e-commerce translation model trained to convergence, which is used to translate the question text into a target question text corresponding to the knowledge language.

[0130] In a further embodiment, the question text acquisition module 5100 includes: a data set acquisition submodule for acquiring a multilingual parallel corpus data set in the field of e-commerce customer service, wherein the multilingual parallel corpus data set includes multiple parallel corpus data pairs, and the parallel corpus data pairs include question texts in different languages ​​collected in the field of e-commerce customer service and their corresponding translation texts in knowledge languages; a translation model training submodule for inputting the multilingual parallel corpus data set into the e-commerce translation model for training, and using the translation text in the knowledge language as a supervision label to train the e-commerce translation model to a convergence state, wherein the e-commerce translation model is used to translate the question text into a target question text corresponding to the knowledge language.

[0131] In order to solve the above technical problems, the present application also provides a computer device. Figure 4 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via 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 may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a multilingual intelligent question-answering method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the multilingual intelligent question-answering method of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0132] In this embodiment, the processor is used to execute Figure 3 The memory stores the program code and various data required to execute the above modules or submodules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all modules / submodules in the multilingual intelligent question-answering device of the present application, and the server can call the program code and data of the server to execute the functions of all submodules.

[0133] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the multilingual intelligent question-answering method of any embodiment of the present application.

[0134] 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 implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0135] In summary, this application can provide users with timely and accurate responses.

[0136] It will be understood by those skilled in the art that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, changed, combined, or deleted. Furthermore, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted. Furthermore, the steps, measures, and schemes in the various operations, methods, and processes in the prior art that are open source and in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0137] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A multilingual intelligent question answering method, characterized in that: The steps include: Obtain the question text provided by the user and determine the question language of the question text; When the question language is consistent with the knowledge language corresponding to the question-answering knowledge base, a pre-stored question in the question-answering knowledge base that matches the question text is determined, and the answer text corresponding to the pre-stored question is used as the target answer text; When the question language is inconsistent with the knowledge language, the question text is translated into a target question text corresponding to the knowledge language, a pre-stored question matching the target question text is determined in the question-answering knowledge base, and the answer text corresponding to the pre-stored question is translated into a target answer text corresponding to the knowledge language; The target answer text is sent to the user.

2. The multilingual intelligent question-answering method according to claim 1, characterized in that: Get the question text provided by the user, including: Obtaining an initial question text and a previous question text provided by a user, wherein the previous question text is obtained based on the corresponding initial question text in a previous round of question-answering; When the correlation between the initial question text and the previous question text is greater than a preset threshold, the initial question text and the previous question text are combined to generate a question text corresponding to the current round of question and answer; When the correlation between the initial question text and the previous question text is less than a preset threshold, the initial question text is used as the corresponding question text for this round of question and answer.

3. The multilingual intelligent question answering method according to claim 1, characterized in that: When the question language is inconsistent with the knowledge language, the question text is translated into a target question text corresponding to the knowledge language, a pre-stored question matching the target question text is determined in the question-answer knowledge base, and the answer text corresponding to the pre-stored question is translated into a target answer text corresponding to the knowledge language, including: When the question language is inconsistent with the knowledge language, the question text is embedded in a preset first translation prompt template to obtain a first translation prompt instruction which is input into the large language model to control the large language model to translate the question text in the question language into a target question text corresponding to the knowledge language; The target question text is embedded in a preset rewriting prompt template, and the obtained first rewriting prompt instruction is input into the large language model to rewrite the target question text to further highlight the user's intention; The rewritten target question text is matched with each pre-stored question in the question-answering knowledge base for similarity, and the pre-stored question matching the target question text is determined, and then the answer text corresponding to the pre-stored question is determined; The answer text is embedded in a preset second translation prompt template to obtain a second translation prompt instruction which is input into the large language model to control the large language model to translate the answer text in the knowledge language into a target answer text corresponding to the question language.

4. The multilingual intelligent question-answering method according to claim 3, characterized in that: The rewritten target question text is matched with each pre-stored question in the question-answering knowledge base for similarity, and a pre-stored question matching the target question text is determined, and then an answer text corresponding to the pre-stored question is determined, including: Input the target question text and each pre-stored question in the question-answer knowledge base into a pre-trained semantic vector model, respectively, and extract the semantic vectors of the target question text and each pre-stored question; Calculating the similarity between the semantic vector of the target question text and the semantic vectors of each pre-stored question, and determining the pre-stored question with the highest similarity value; When the highest similarity value is greater than a preset similarity threshold, the corresponding pre-stored question is determined to be a pre-stored question in the question-and-answer knowledge base that matches the target question text, and then the answer text corresponding to the pre-stored question is determined.

5. The multilingual intelligent question-answering method according to claim 4, characterized in that: Calculating the similarity between the semantic vector of the target question text and the semantic vectors of each pre-stored question, and determining the pre-stored question with the highest similarity value, comprising: When the highest similarity is less than a preset similarity threshold, the target question text is embedded in the answer generation template, and the generated answer generation instruction controls the large language model to generate a hypothetical answer text corresponding to the target question text; Determine whether there is a matching answer text for the hypothetical answer text in the question-answer knowledge base, and if so, embed the matching answer text into a third translation prompt template, generate a third translation prompt instruction and input it into the large language model to control the large language model to translate the answer text in the knowledge language into a target answer text corresponding to the question language, and send the target answer text to the user; If it does not exist, the artificial answer text corresponding to the question text is obtained and sent to the user as the target answer text.

6. The multilingual intelligent question-answering method according to claim 5, characterized in that: After judging whether the hypothetical answer text has a matching answer text in the question-answer knowledge base, it includes: When the assumed answer text has a matching answer text in the question-answer knowledge base, the pre-stored question corresponding to the answer text and the question text are used as a translation instance; Collecting multiple translation instances to form a translation instance set, and when the total number of translation instances in the translation instance set is greater than a preset number threshold, constructing a fine-tuning training set based on the translation instance set; The preset e-commerce translation model is trained using the fine-tuning training set to obtain an e-commerce translation model that has been trained to convergence, and is used to translate the question text into a target question text corresponding to the knowledge language.

7. The multilingual intelligent question answering method according to any one of claims 1 to 6, characterized in that: Before getting the user-supplied question text, include: Acquire a multilingual parallel corpus dataset in the field of e-commerce customer service, wherein the multilingual parallel corpus dataset includes a plurality of parallel corpus data pairs, wherein the parallel corpus data pairs include question texts in different languages ​​collected in the field of e-commerce customer service and corresponding translation texts in knowledge languages; The multilingual parallel corpus dataset is input into the e-commerce translation model for training, and the translated text of the knowledge language is used as a supervision label to train the e-commerce translation model to a convergence state. The e-commerce translation model is used to translate the question text into a target question text corresponding to the knowledge language.

8. A multi-language intelligent question-answering device, characterized in that: include: The question text acquisition module is used to obtain the question text provided by the user and determine the question language of the question text; An answer text matching module is used to determine a pre-stored question in the question and answer knowledge base that matches the question text when the question language is consistent with the knowledge language corresponding to the question and answer knowledge base, and use the answer text corresponding to the pre-stored question as the target answer text; A translation conversion module, for translating the question text into a target question text corresponding to the knowledge language when the question language is inconsistent with the knowledge language, determining a pre-stored question in the question-answering knowledge base that matches the target question text, and translating the answer text corresponding to the pre-stored question into a target answer text corresponding to the knowledge language; The answer text sending module is used to send the target answer text to the user.

9. A computer device comprising a central processing unit and a memory, characterized in that: The central processing unit is used to call and run the 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