Large language model-based after-sales service management method, storage medium and device

Abstract

The invention discloses an after-sales service management method based on a large language model, a storage medium and an electronic device, and relates to the field of after-sales services of smart home, and the after-sales service management method based on the large language model comprises the steps that a platform end receives session information sent by an after-sales end; under the condition that historical session data corresponding to the session information exists in the database, based on the session information and the historical session data, a first intention recognition model is adopted to determine a recognition result corresponding to the session information, and a corresponding target verbal skill configuration template is determined; calling a reply language model to obtain reply information corresponding to the session information, and feeding back the recognition result and the reply information to the after-sales terminal; and the after-sales terminal determines an after-sales service corresponding to the reply information based on the identification result, and feeds back the reply information and an after-sales service scheme corresponding to the after-sales service to the user terminal. The understanding ability of an after-sales system is improved through a large language model, and the problems that an existing after-sales service system is limited in understanding ability, insufficient in emotion recognition and the like are solved.

Description

Technical Field

[0001] This application relates to the field of smart home after-sales service, and more specifically, to an after-sales service management method, storage medium, and device based on a large language model. Background Technology

[0002] Traditional after-sales customer service systems primarily rely on rule-based systems, keyword matching, limited natural language processing techniques, and some early machine learning algorithms. These systems typically include automated call distribution systems, interactive voice response systems, and text-based online chatbots.

[0003] Traditional after-sales service systems often struggle to understand complex language structures and contexts, easily leading to misunderstandings of customer intentions. For issues outside the predefined scope, they lack emotional recognition capabilities, potentially failing to provide satisfactory answers and effectively handling emotional or complex communication. Furthermore, while existing systems may save costs during initial deployment, long-term software upgrades and maintenance can become a burden.

[0004] For users, existing technology performs well for simple problems, but for complex or rare problems, the user experience may be poor, potentially leading to decreased customer satisfaction. Existing after-sales services often suffer from limited understanding, poor flexibility, insufficient emotional awareness, high upgrade or maintenance costs, and inconsistent customer experiences, resulting in problems such as inconsistent user order information, incomplete collection of user needs, and reliance on service personnel to reconfirm with users after order dispatch. Summary of the Invention

[0005] This application provides an after-sales service management method, apparatus, equipment, and computer storage medium based on a large language model, to solve the problems of limited understanding ability, poor flexibility, insufficient emotion recognition, high upgrade or maintenance costs, and inconsistent customer experience in existing after-sales service systems.

[0006] Firstly, this application provides an after-sales service management method based on a large language model, applied to a platform. The method includes:

[0007] Receive session information sent by the back-end support team;

[0008] If historical session data corresponding to the session information exists in the database, the recognition result corresponding to the session information is determined based on the session information and the historical session data using a first intent recognition model. The recognition result includes key information and intent information, wherein the key information is used to indicate information in the session information that is associated with the target after-sales business.

[0009] Based on the historical conversation data, the key information, and the intent information, the corresponding target dialogue configuration template is determined;

[0010] The large language model is invoked, and pre-trained based on the target speech configuration template to obtain the response language model;

[0011] The key information, intent information, and target dialogue configuration template are analyzed and combined using the response language model to obtain the response information corresponding to the conversation information, and the recognition result and the response information are fed back to the sales backend.

[0012] Optionally, determining the recognition result corresponding to the session information using a first intent recognition model based on the session information and the historical session data includes:

[0013] Determine the text information corresponding to the session information and the historical session data;

[0014] The first intent recognition model is used to perform intent recognition processing on the text information;

[0015] If the first intent recognition model fails to recognize the intent, the large language model is invoked to perform language analysis on the text information to obtain the recognition result corresponding to the conversation information.

[0016] Optionally, the step of calling a large language model to perform language analysis on the text information to obtain the recognition result corresponding to the conversation information includes:

[0017] A large language model is invoked, and the large language model is pre-trained based on multiple candidate intent templates to obtain a second intent recognition model;

[0018] The second intent recognition model is invoked to perform language analysis processing on the text information to obtain the first intent information and key information corresponding to the text information;

[0019] Based on the first intent information, the intent information of the session information is determined, and the intent information and the key information are used as the identification result;

[0020] Optionally, determining the corresponding target script configuration template based on the historical session data, the key information, and the intent information includes:

[0021] The session scenario of the session information is determined based on at least one of the key information, the intent information, and the historical session data;

[0022] Based on the intent information and the conversation scenario, the corresponding target dialogue configuration template is determined from multiple candidate dialogue configuration templates;

[0023] Optionally, before determining the recognition result corresponding to the session information using an intent recognition model based on the session information and the historical session data, the method further includes:

[0024] Determine whether there exists historical interaction data in the database that corresponds to the session information and the interval between the interaction time and the current time is less than a preset duration;

[0025] If so, determine whether the number of historical interaction data is greater than a preset number;

[0026] If the number of historical interaction data is greater than a preset number, a preset number of historical interaction data whose interaction time is closest to the current time will be used as the historical session data.

[0027] If the amount of historical interaction data is not greater than a preset amount, the historical interaction data will be used as the historical session data.

[0028] Optionally, the method further includes:

[0029] If the historical session data is not present in the database, the recognition result corresponding to the session information is determined using a first intent recognition model based on the session information.

[0030] Secondly, this application provides an after-sales service management method based on a large language model, applied to the after-sales backend. This method includes:

[0031] Forward session information sent by the user to the platform.

[0032] The system receives the recognition result and response information corresponding to the session information fed back by the platform. The recognition result is determined by the first intent recognition model of the platform based on the session information and the corresponding historical session data. The recognition result includes key information and intent information. The key information is used to indicate information related to after-sales business in the session information. The response information is determined by the response language model of the platform based on the key information, the intent information and the script configuration template.

[0033] Based on the key information and the intent information, the after-sales service corresponding to the response information is determined, and the after-sales service plan corresponding to the response information and the after-sales service is fed back to the user terminal.

[0034] Optionally, determining the after-sales service corresponding to the response information based on the key information and the intent information includes:

[0035] Based on the key information and the intent information, the corresponding after-sales service is determined, and the corresponding business processing module for the after-sales service is determined, wherein different intent information corresponds to different business processing modules;

[0036] Based on the after-sales service, the business processing module, and the response text, the after-sales service plan is determined.

[0037] Thirdly, this application provides an after-sales service management device based on a large language model, applied to a platform, the device comprising:

[0038] The receiving module is used to receive session information sent by the back-end sales department;

[0039] The determination module is used to determine the recognition result corresponding to the session information based on the session information and the historical session data when the database contains historical session data corresponding to the session information, using a first intent recognition model. The recognition result includes key information and intent information, wherein the key information is used to indicate information in the session information that is associated with the target after-sales business.

[0040] The determining module is further configured to determine the corresponding target dialogue configuration template based on the historical session data, the key information, and the intent information;

[0041] The calling module is used to call the large language model and pre-train the large language model based on the target speech configuration template to obtain the response language model;

[0042] The calling module is also used to call the response language model to analyze and combine the key information, the intent information and the target speech configuration template to obtain the response information corresponding to the conversation information;

[0043] The feedback module is used to send the identification results and the response information back to the after-sales service department.

[0044] Optionally, the device may also include: an identification module;

[0045] The determining module is further configured to determine the text information corresponding to the session information and the historical session data;

[0046] The recognition module is used to perform intent recognition processing on the text information using the first intent recognition model.

[0047] The calling module is also used to call a large language model to perform language analysis processing on the text information in the event that the first intent recognition model fails to recognize the text, so as to obtain the recognition result corresponding to the conversation information.

[0048] Optionally, the calling module is further configured to call a large language model and pre-train the large language model based on multiple candidate intent templates to obtain a second intent recognition model;

[0049] The calling module is further configured to call the second intent recognition model to perform language analysis processing on the text information to obtain the first intent information and key information corresponding to the text information;

[0050] The determining module is further configured to determine the intent information of the session information based on the first intent information, and use the intent information and the key information as the identification result;

[0051] Optionally, the determining module is further configured to determine the session scenario of the session information based on at least one of the key information, the intent information, and the historical session data;

[0052] The determining module is further configured to determine the corresponding target dialogue configuration template from multiple candidate dialogue configuration templates based on the intent information and the conversation scenario;

[0053] Optionally, the device further includes: a determination module;

[0054] The determining module is further configured to determine whether there is historical interaction data in the database that corresponds to the session information and the interval between the interaction time and the current time is less than a preset duration;

[0055] The judgment module is used to determine whether the number of historical interaction data is greater than a preset number if the condition is met.

[0056] The determining module is further configured to, when the number of historical interaction data is greater than a preset number, select a preset number of historical interaction data whose interaction time is closest to the current time as the historical session data.

[0057] The determining module is further configured to, when the number of historical interaction data is not greater than a preset number, use the historical interaction data as the historical session data;

[0058] Optionally, the determining module is further configured to, in the case that the historical session data does not exist in the database, determine the recognition result corresponding to the session information based on the session information using a first intent recognition model.

[0059] Fourthly, this application provides an after-sales service management device based on a large language model, applied to the after-sales service department. The device includes:

[0060] The forwarding module is used to forward session information sent by the user client to the platform.

[0061] The receiving module is used to receive the recognition result and response information corresponding to the session information fed back by the platform. The recognition result is determined by the first intent recognition model of the platform based on the session information and the corresponding historical session data. The recognition result includes key information and intent information. The key information is used to indicate information related to after-sales business in the session information. The response information is determined by the response language model of the platform based on the key information, the intent information and the script configuration template.

[0062] The determination module is used to determine the after-sales service corresponding to the response information based on the key information and the intent information;

[0063] The feedback module is used to provide the user terminal with the response information and the after-sales service plan corresponding to the after-sales service.

[0064] Optionally, the determining module is further configured to determine the corresponding after-sales service based on the key information and the intent information, and to determine the business processing module corresponding to the after-sales service, wherein different intent information corresponds to different business processing modules;

[0065] The determining module is further configured to determine the after-sales service plan based on the after-sales service, the service processing module, and the reply text.

[0066] Fifthly, this application provides a computer-readable storage medium comprising a stored program, wherein the program executes at runtime to implement the after-sales service management method based on a large language model as described in the first aspect and various possible implementations of the first aspect and the second aspect and various possible implementations of the second aspect.

[0067] Sixthly, this application provides an electronic device, comprising:

[0068] Memory;

[0069] processor;

[0070] The memory contains computer programs;

[0071] The processor is configured to execute the computer program to implement the after-sales service management method based on a large language model as described in the first aspect and various possible implementations of the first aspect and various possible implementations of the second aspect.

[0072] This application provides an after-sales service management method based on a large language model. The method involves receiving session information from the back-end system via a platform. If historical session data corresponding to the session information exists in the database, a first intent recognition model is used to determine the recognition result based on the session information and the historical session data. The recognition result includes key information and intent information, where the key information indicates information in the session information related to the target after-sales service. Based on the historical session data, key information, and intent information, a corresponding target dialogue configuration template is determined. The large language model is invoked and pre-trained based on the target dialogue configuration template to obtain a response language model. The response language model is then used to analyze and combine the key information, intent information, and target dialogue configuration template to obtain the response information corresponding to the session information. The recognition result and response information are then fed back to the back-end system. The back-end system determines the after-sales service corresponding to the response information based on the recognition result and provides the user with the corresponding after-sales service plan. By invoking a large language model, the understanding capabilities of the after-sales service system are enhanced, the system's corpus model is improved, and complex user queries are better understood, including context, tone, and underlying intent. This allows for more accurate problem parsing and the handling of ambiguous or incomplete information. By inferring complete conversation information from the context, the frequency of transfers to human customer service is reduced, the self-service problem-solving rate is increased, and time and resources are saved. This solves the problems of limited understanding capabilities, poor flexibility, and insufficient emotion recognition in the existing after-sales service system. Attached Figure Description

[0073] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0074] Figure 1 This is a schematic diagram of the hardware environment for an after-sales service management method based on a large language model according to an embodiment of this application;

[0075] Figure 2 This is an interactive schematic diagram of an after-sales service management method based on a large language model according to an embodiment of this application;

[0076] Figure 3 This is a flowchart of an after-sales service management method based on a large language model according to an embodiment of this application. Figure 1 ;

[0077] Figure 4 This is a flowchart of an after-sales service management method based on a large language model according to an embodiment of this application. Figure 2 ;

[0078] Figure 5 This is a schematic diagram of the structure of an after-sales service management device based on a large language model according to an embodiment of this application. Figure 1 ;

[0079] Figure 6 This is a schematic diagram of the structure of an after-sales service management device based on a large language model according to an embodiment of this application. Figure 2 ;

[0080] Figure 7 This is a schematic diagram of the structure of an after-sales service management device based on a large language model according to an embodiment of this application.

[0081] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Detailed Implementation

[0082] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0083] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0084] First, let me explain the terms used in this application:

[0085] Large Language Models (LLMs) are neural network models consisting of billions or even hundreds of billions of parameters. They learn the structure, syntax, semantics, and usage habits of a language through massive amounts of text data. These models are typically based on the Transformer architecture and utilize self-attention mechanisms to process text. LLMs can handle a variety of natural language tasks, such as text classification, question answering, and dialogue, and are an important pathway to artificial intelligence. LLMs are trained using massive datasets, becoming ultra-large-scale deep learning models, which is the powerful foundation for their ability to recognize, translate, predict, or generate text or other content.

[0086] Regular expressions, also known as regulated expressions (often abbreviated as regex, regexp, or RE in code), are text patterns that include ordinary characters (such as letters from a to z) and special characters (called "metacharacters"), and are a concept in computer science. Regular expressions use a single string to describe and match a series of strings that match a certain syntax rule. They are commonly used to search for and replace text that conforms to a specific pattern (rule). A regular expression is a logical formula for string manipulation; it uses predefined specific characters and combinations of these characters to form a "rule string," which expresses a filtering logic for strings.

[0087] Information anonymization refers to the process of transforming sensitive information using anonymization rules to reliably protect sensitive and private data. Data anonymization can be divided into two types: static data anonymization and dynamic data anonymization. Static data anonymization is generally used in scenarios involving the external distribution of data. It saves the modified data for use by the data user. Dynamic data anonymization is generally used in scenarios directly connected to production data. It processes the data during the acquisition process without modifying the original data.

[0088] Intent recognition is a crucial task in natural language processing, aiming to determine the intent or goal expressed by user-input text. It plays a vital role in applications such as dialogue systems, intelligent assistants, and search engines. The basic process of intent recognition typically includes data preprocessing, feature extraction, model training, and intent classification. Data preprocessing mainly involves cleaning, standardizing, and segmenting the raw text; feature extraction extracts useful information for intent recognition, such as keywords and parts of speech; model training uses labeled data to train a classifier capable of accurately recognizing the user's input intent; and finally, intent classification feeds the user's input text into the trained classifier to determine its intent category.

[0089] According to one aspect of the embodiments of this application, an after-sales service management method based on a large language model is provided. This after-sales service management method based on a large language model is widely applicable to whole-house intelligent digital control application scenarios such as smart homes, smart home ecosystems, and intelligence house ecosystems. Optionally, in this embodiment, the above-mentioned after-sales service management method based on a large language model can be applied to, for example... Figure 1 The hardware environment shown consists of terminal device 102 and server 104. For example... Figure 1 As shown, server 104 is connected to terminal device 102 via a network and can be used to provide services (such as after-sales service) to the terminal or clients installed on the terminal. A database can be set up on the server or independently of the server to provide data storage services for server 104. Cloud computing and / or edge computing services can be configured on the server or independently of the server to provide data processing services for server 104.

[0090] The aforementioned network may include, but is not limited to, at least one of the following: wired network, wireless network. The aforementioned wired network may include, but is not limited to, at least one of the following: wide area network, metropolitan area network, local area network. The aforementioned wireless network may include, but is not limited to, at least one of the following: Wi-Fi (Wireless Fidelity), Bluetooth. The terminal device 102 may not be limited to PC, mobile phone, tablet computer, smart air conditioner, smart range hood, smart refrigerator, smart oven, smart stove, smart washing machine, smart water heater, smart washing equipment, smart dishwasher, smart projector, smart TV, smart clothes rack, smart curtains, smart audio-visual equipment, smart socket, smart speaker, smart speaker box, smart fresh air equipment, smart kitchen and bathroom equipment, smart bathroom equipment, smart robot vacuum cleaner, smart window cleaning robot, smart mopping robot, smart air purifier, smart steam oven, smart microwave oven, smart water heater, smart air purifier, smart water dispenser, smart door lock, etc.

[0091] Traditional after-sales customer service systems primarily rely on rule-based systems, keyword matching, limited natural language processing techniques, and some early machine learning algorithms. These systems typically include automated call distribution systems, interactive voice response systems, and text-based online chatbots.

[0092] Traditional after-sales service systems often struggle to understand complex language structures and contexts, making them prone to misunderstanding customer intentions. For issues outside the predefined scope, they lack emotional recognition capabilities, and the system may not be able to provide satisfactory answers or effectively handle emotional or complex emotional exchanges.

[0093] While existing systems may save costs during initial deployment, long-term software upgrades and maintenance can become a burden. For users, existing technology performs well for simple problems, but for complex or rare problems, the user experience may be poor, potentially leading to decreased customer satisfaction.

[0094] Because existing after-sales services often suffer from limited understanding, poor flexibility, insufficient emotional recognition, high upgrade or maintenance costs, and inconsistent customer experiences, problems such as inconsistent user order information, incomplete collection of user needs information, and reliance on service personnel to reconfirm with users after order dispatch occur.

[0095] To address the aforementioned issues, this application provides an after-sales service management method based on a large language model. By introducing a large language model into the platform to intelligently generate response scripts for after-sales issues, upon receiving conversation information from the back-end, if historical conversation data corresponding to the conversation information exists in the database, an intent recognition model is used to determine the recognition result corresponding to the conversation information based on the conversation information and historical conversation data, and the corresponding target script configuration template is determined. The large language model is then invoked to analyze and combine relevant information to obtain the response information corresponding to the conversation information. The recognition result and response information are then fed back to the back-end. This solves the problems of limited understanding ability, poor flexibility, insufficient emotion recognition, high upgrade or maintenance costs, and inconsistent customer experience in existing after-sales service systems.

[0096] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0097] Figure 2 This is an interactive diagram illustrating an after-sales service management method based on a large language model, provided as an embodiment of this application. The implementing entities in this embodiment may include, for example, a user terminal, an after-sales service backend, and a platform terminal, such as... Figure 2 As shown in the figure, this embodiment provides an after-sales service management method based on a large language model, including:

[0098] S101: The user sends session information to the after-sales service department.

[0099] In this context, "user terminal" refers to the user's device, which may include mobile phones, websites, chat windows, etc. "Back-end" refers to the programs or web pages that may include after-sales services such as customer service, technical support, and after-sales data analysis.

[0100] Understandably, users enter the information they need to inquire about or provide feedback on the user's device and click the send button. The input information can include text, images, videos, or other media files, depending on the interaction types supported by the after-sales system; this solution does not impose any restrictions. The after-sales server receives the request and data from the user's device, and the system can verify the user's identity and permissions, confirming that a legitimate user is sending session information.

[0101] Specifically, the back-end system can integrate conversational functionality to present different business processing solutions based on user inquiries and recognition results. These solutions could include intelligent reply scripts, work order editing cards, access links, service intervention reminders, images, and videos.

[0102] S102: The sales backend forwards the session information to the platform.

[0103] The platform side refers to the server that can analyze and process session information. The back-end can perform anonymization on the session information. First, it can identify sensitive information in the session information to determine which information needs to be anonymized. Then, it can select an appropriate anonymization method based on the type and sensitivity of the sensitive information. The specific anonymization method can be selected according to the actual situation, and the sensitive information can be anonymized through corresponding code or scripts.

[0104] Understandably, the back-end can securely store the anonymized data and transmit it to the platform. During storage, appropriate security measures, such as encrypted storage and access control, can be taken to protect the data security. During transmission, a secure transmission protocol needs to be used to ensure the data security during transmission.

[0105] S103: If the platform has historical session data corresponding to the session information in the database, it uses an intent recognition model to determine the recognition result corresponding to the session information based on the session information and the historical session data.

[0106] The identification results include key information and intent information. The database can store historical data of each user interaction and intent identification results, providing context for the large model and also for user data analysis. When historical data is needed, it can be traversed and relevant historical data can be returned for subsequent data processing. Historical session data related to the current session is retrieved from the database, which may include previous user queries, dialogue content, system responses, timestamps, etc. Based on the obtained historical session data and session information forwarded by the back-end, which may include text, voice, images, video, or other forms of input, the platform can process the data based on different input session information, integrating it into a data type that the server can recognize and process. Specific data processing methods are not detailed here.

[0107] Understandably, the platform can extract useful features from historical conversation data, such as users' historical query patterns, topic shifts, sentiment tendencies, and response times. Subsequently, it can extract features from the current conversation information, such as keywords, sentence structure, and sentiment analysis. After extracting key information, the features from the historical and current conversations can be combined to form a complete conversation representation. This combination process can be achieved through methods such as concatenation, weighting, and attention mechanisms. From this, the key information within the conversation can be derived.

[0108] It should be noted that the platform can select a suitable intent recognition model based on the application scenario and data characteristics. This intent recognition model can be pre-set and trained on the platform. For example, the intent recognition model can include a CCG model, a regularization model, or other training models that can be used for corpus intent recognition. The intent recognition model can be optimized by adjusting model parameters, optimizing the algorithm, or introducing new features to improve the accuracy and efficiency of intent recognition.

[0109] Specifically, the preprocessed current session information is input into a trained intent recognition model. The model can then perform intent recognition based on the input session information and features from historical session data. Subsequently, the model outputs the intent recognition result corresponding to the current session information. This result can be, for example, a specific intent label, a probability distribution, or a sorted list of intents, thus deriving the intent information corresponding to the session information.

[0110] Through the above process, the platform can make full use of historical session data combined with session information to assist the decision-making process of the intent recognition model, thereby improving the accuracy and efficiency of recognizing current session information.

[0111] S104: The platform determines the corresponding target script configuration template based on historical conversation data, key information, and intent information.

[0112] The script configuration template can be a pre-set script library. Based on feedback from the platform, the target script configuration template is determined. For example, a clear list of intents can be defined based on business needs and historical data. Each intent represents a possible user need or goal, and one or more script templates can be designed for each intent. These templates can cover different response styles, tones, and possible user feedback. By establishing a mapping relationship between intents and script templates, it is ensured that a suitable script template can be found for each intent.

[0113] In one possible implementation, the conversation scenario of the conversation information is determined based on at least one of key information, intent information, and historical conversation data; and the corresponding target speech configuration template is determined from multiple candidate speech configuration templates based on the intent information and the conversation scenario.

[0114] The platform can infer the current conversation scenario based on key information, intent information, and relevant information from historical conversation data. The conversation scenario can involve multiple aspects, such as the user's current needs, emotional state, and context. Key information analysis can extract crucial information from user input, such as time, location, people, and events, which helps in understanding the conversation's context. Alternatively, intent information can be interpreted, and combined with the results of the intent recognition model, to clarify the user's specific goals or needs. Furthermore, historical conversation data can be reviewed to understand the user's previous interactions and possible conversation paths. By comprehensively analyzing all of this information, the platform can infer the general scenario of the current conversation.

[0115] Understandably, the platform can set up a candidate dialogue configuration template library. This library contains multiple dialogue templates for different scenarios and intents. These templates can be designed based on factors such as business needs, user profiles, and product characteristics, and can be continuously optimized and updated. Based on intent information and conversation scenarios, the platform can match and select from the candidate dialogue configuration template library. For example, it can combine factors such as the context of the conversation, the user's emotional state, and interaction history to select the dialogue template that best fits the current conversation scenario. Through the above matching and selection process, the platform can ultimately determine one or more target dialogue configuration templates, which will serve as the basis for the system's response and interaction with the user.

[0116] Optionally, conversation scenarios can include, for example, initial response scenarios, intermediate scenarios, and ending scenarios. For instance, the initial response scenario can indicate the beginning of the conversation, typically occurring when a user first initiates a request or inquiry. At this stage, the system has not yet established a complete interaction history with the user, and therefore relies primarily on the initial information input by the user to respond. Furthermore, the intermediate scenario can indicate the main body of the conversation, occurring after the initial response and before the end. At this stage, the system has already established a certain interaction history with the user and can dynamically adjust the script configuration template based on user feedback, i.e., historical conversation data, key information, and intent information. As another example, the ending scenario can indicate the closing stage of the conversation, marking the end of the current conversation. At this stage, the system can summarize the conversation content, confirm whether the user's after-sales needs have been met, and may provide follow-up support or guide the user to perform other operations.

[0117] Specifically, the script template for the first-call scenario can be determined based on historical conversation information. If there is no historical conversation data, or if the previous historical conversation data was from an ended scenario, the current scenario can be considered the first-call scenario. In this case, if there are previous conversation records for the user, such as those passed across conversation contexts, they can be used as historical conversation data for reference. If not, the user's request, i.e., the first-call script, can be parsed to identify their intent, and an appropriate opening line or guiding question can be selected accordingly. Based on the initial information input by the user and possible cross-conversation contexts, a suitable first-call script template can be selected to establish an initial connection with the user and guide the conversation in a deeper direction.

[0118] The determination of intermediate scenarios can be based on key information, intent information, and historical session data. For example, key information can be extracted from user input, such as product name, order number, and problem description, and the user's current intent can be obtained through an intent recognition model. This can be combined with historical session data, including all interaction records prior to the current session. Based on key information, intent information, and historical session data, one or more intermediate dialogue configuration templates can be selected to accurately respond to user needs, maintain session fluency, and guide the session towards the target direction.

[0119] The conclusion can be determined based on key information and intent information. Key information refers to information directly related to the conversation outcome; intent information could be the user's intended end to the conversation, such as a closing remark like "Thank you, I understand." Based on this key and intent information, a suitable closing remark template can be selected. This template should summarize the key points of the conversation, provide follow-up support information such as contact information and frequently asked questions, and politely conclude the conversation.

[0120] S105: The platform calls the large language model and pre-trains the large language model based on the target speech configuration template to obtain the response language model.

[0121] Among them, large language models, such as GPT and BERT, can be pre-trained based on specific target speech configuration templates to obtain customized language models for responses.

[0122] Understandably, based on the target dialogue template, relevant dialogue template parameters are collected. These parameters can come from historical dialogue records, user feedback, professional knowledge bases, etc. The collected data is then cleaned to remove noise, duplicates, and irrelevant information to ensure data quality. After processing, the cleaned data is labeled, for example, by establishing a correspondence between inputs such as user questions and outputs such as expected responses. Based on the target dialogue, one or more templates can be designed. These templates can cover common dialogue scenarios and possible user questioning methods. The templates are then applied to the data labeling process to ensure that the inputs and outputs in the dataset conform to the template specifications.

[0123] Optionally, a suitable large language model can be selected as the base, taking into account factors such as model performance, trainability, and resource consumption. Specific criteria for model selection are not limited in this solution. Using the prepared data and templates, the base model is trained in a customized manner. During training, model performance can be optimized by adjusting model parameters, optimizing the algorithm, and changing the number of training epochs. The model performance is then evaluated using a test set, and the model is fine-tuned based on the evaluation results, for example, through multiple iterations of training and parameter adjustments to achieve the best results.

[0124] Specifically, the trained response language model is deployed to the platform and integrated into the platform's dialogue system to achieve automatic response functionality. Simultaneously, the model's performance and user feedback can be monitored for subsequent optimization. On the platform, a large language model is invoked and pre-trained based on a specific target dialogue configuration template to obtain a customized language model specifically for responses.

[0125] S106: The platform calls the response language model to analyze and combine the key information, the intent information, and the target dialogue configuration template to obtain the response information corresponding to the conversation information.

[0126] The response language model is based on a pre-trained large language model. Since it relies on the large language model, it can be understood that key information and user intent are taken as input. The large language model can be used to generate or complete content, fill in placeholders in templates, or generate more natural and fluent response content based on template structure. Furthermore, it can adjust the language style of the response information according to user preferences or dialogue context to make it more in line with the scenario and user expectations.

[0127] Specifically, by combining the output of the template and the response language model, response information is generated. Post-processing work such as grammar checking and spell correction can be performed on the generated response to ensure the accuracy and readability of the response information.

[0128] It's worth noting that dialogue effectiveness data can be collected through user feedback such as satisfaction ratings and direct feedback. Based on user feedback and effectiveness evaluation, the intent recognition model, large language model, response language model, and dialogue configuration templates can be continuously optimized to improve the overall dialogue experience. On the platform side, by calling the large language model and combining it with key information, intent recognition, and target dialogue configuration templates, accurate and fluent response information can be provided to users, thereby improving user satisfaction and the platform's service quality.

[0129] S107: The platform will send the identification results and response information back to the after-sales service department.

[0130] S108: The after-sales service department determines the after-sales business corresponding to the response information based on key information and intent information.

[0131] The identification results are determined by the platform's first intent recognition model based on session information and corresponding historical session data. The identification results include key information and intent information. Key information is used to indicate information in the session information that is related to after-sales business. The response information is determined by the platform's response language model based on key information, intent information, and script configuration template.

[0132] Understandably, after receiving the recognition results from the platform, the back-end sales team can determine the corresponding after-sales service based on these results. The back-end team first parses the recognition results from the platform, including user intent, key information, and any additional conversation context. Based on the parsed recognition results, the back-end team can map user issues to predefined after-sales service types. These service types may include, but are not limited to: product inquiries, fault reporting, return / exchange requests, complaints and suggestions. Specific after-sales services can be set according to different products, and this application does not impose any restrictions.

[0133] Specifically, after determining the business type, the back-end sales team can determine the corresponding processing steps and relevant maintenance personnel based on the established after-sales service strategies and processes. These processes may include: information verification, problem diagnosis, solution development, execution, and follow-up.

[0134] In one possible implementation, the after-sales service department determines the corresponding after-sales service based on key information and intent information, and determines the corresponding business processing module for the after-sales service, wherein different intent information corresponds to different business processing modules; based on the after-sales service, the business processing module, and the response text, an after-sales service plan is determined.

[0135] Optionally, the back-end system can parse the recognition results transmitted from the platform, which includes key information and intent information. Key information may include specific data related to after-sales service, such as fault description and order number, while intent information can be used to indicate the user's desired goal or the type of problem to be solved.

[0136] Understandably, based on key information and intent information, the after-sales service department can determine the type of after-sales service requested by the user. For example, if the intent is "return or exchange request" and the key information can include a valid order number and a reasonable reason for return or exchange, then the after-sales service is determined to be "processing return or exchange request".

[0137] Specifically, different intent information typically corresponds to different business processing modules. A business processing module is a set of components or processes within the after-sales system used to execute specific types of after-sales business. The back-end system can select the appropriate business processing module to further process user requests based on the determined after-sales business type. For example, a return / exchange request can trigger a business processing module that includes multiple steps such as product inspection, inventory verification, and logistics arrangement. Similarly, a "query" request can call the order expediting interface of the work order system to expedite the user's order. The back-end system can combine after-sales business, business processing modules, and response text to determine a complete after-sales service plan.

[0138] S109: The after-sales service department provides feedback and response information to the user, as well as the corresponding after-sales service plan.

[0139] The back-end system can confirm that the user's after-sales request has been received and receive the response information generated by the platform. At the same time, the back-end system can determine the after-sales service plan and send the response information and after-sales service plan back to the user.

[0140] Understandably, after-sales service providers can send response information and after-sales service plans to users through appropriate channels such as email, SMS, APP notifications, and online customer service systems.

[0141] It's important to note that after feedback is sent, the after-sales team can maintain communication with the user, promptly answer their questions, resolve any potential issues, and adjust the after-sales service plan based on the actual situation. Through continuous follow-up and communication, the after-sales team can ensure the smooth implementation of the after-sales service plan, thereby improving user satisfaction and loyalty.

[0142] This application provides an after-sales service management method based on a large language model. The method involves receiving session information from the back-end system via a platform. If historical session data corresponding to the session information exists in the database, a first intent recognition model is used to determine the recognition result based on the session information and the historical session data. The recognition result includes key information and intent information, where the key information indicates information in the session information related to the target after-sales service. Based on the historical session data, key information, and intent information, a corresponding target dialogue configuration template is determined. The large language model is invoked and pre-trained based on the target dialogue configuration template to obtain a response language model. The response language model is then used to analyze and combine the key information, intent information, and target dialogue configuration template to obtain the response information corresponding to the session information. The recognition result and response information are then fed back to the back-end system. The back-end system determines the after-sales service corresponding to the response information based on the recognition result and provides the user with the corresponding after-sales service plan. By calling large language models, the understanding ability of the after-sales service system is improved, the system's corpus model is perfected, complex user queries are better understood, the self-service problem-solving rate is increased, time and resources are saved, and the problems of limited understanding ability, poor flexibility and insufficient sentiment recognition in the existing after-sales service system are solved.

[0143] Figure 3 The process of an after-sales service management method based on a large language model provided in this application embodiment Figure 1 .like Figure 3 As shown, this embodiment is... Figure 2 Based on the embodiments, this paper further explains a possible implementation method for performing intent recognition on session information and historical session data and generating recognition results, with the platform as the implementing entity. This method includes:

[0144] S201: Determine the text information corresponding to the session information and historical session data.

[0145] The process of identifying the text information corresponding to session information and historical session data involves extracting and organizing key content from user interactions with the system or customer service. The session information needs to clearly define the start and end points of each session. This helps divide continuous interaction records into independent session units, facilitating subsequent processing and analysis. For each session unit, text information can be extracted, including user-sent messages, system or customer service replies, and text content from any relevant attachments or links, ensuring that the extracted text information comprehensively reflects the content and context of the session.

[0146] Understandably, if a session is continuous or the user has multiple interaction records, the current session can be linked with historical session data to establish the user's session history, understand user behavior patterns and preferences, and provide personalized services. During the association process, matching can be performed based on information such as user ID, session ID, and timestamp. Based on the analysis results and the associated historical session data, text information corresponding to the current session information and historical session data can be extracted and organized. Through the above operations, the text information corresponding to the session information and historical session data can be determined, providing strong support for subsequent service provision, problem analysis, and user behavior research.

[0147] S202: The first intent recognition model is used to perform intent recognition processing on the text information.

[0148] The first intent recognition model is used to determine the user's intent or purpose from the text or related content entered by the user.

[0149] Understandably, when the text to be recognized is input into a pre-trained intent recognition model, the model can predict the intent category it belongs to based on the input text and output the prediction result, i.e., the intent category corresponding to the text. At the same time, the intent recognition model can be continuously optimized through user feedback to improve the accuracy and robustness of intent recognition.

[0150] S203: Determine whether the first intent recognition model has successfully recognized the intent; if yes, proceed to step S204; if no, proceed to step S205.

[0151] S204: Determine the identification results, which include key information and intent information.

[0152] Step S204 is similar to step S103 above, and will not be described again here.

[0153] S205: Call the large language model and pre-train it based on multiple candidate intent templates to obtain the second intent recognition model.

[0154] Each of the multiple predefined candidate intent templates can represent a possible user intent. The corresponding intent template in the platform-side intent recognition model can be transmitted to the large language model to impose conditional constraints on the large language model. The large language model can then determine the intent information corresponding to the text information to be recognized based on the multiple candidate intent templates.

[0155] Understandably, when an intent recognition model fails to identify the intent of a text, it is appropriate to call upon a more general large language model to perform in-depth language analysis of the text information. Large language models typically have stronger text understanding and generation capabilities, and can capture complex semantics and contextual information in the text, thereby extracting the intent and key information of the text more accurately.

[0156] It should be noted that, based on task requirements, resource constraints, and model performance, a suitable large language model can be selected, and the desired information can be queried, such as "Please analyze the intent of this text." Furthermore, other constraints or guidelines can be added according to specific needs.

[0157] In this embodiment, the corresponding intent template in the platform-side intent recognition model can be transmitted to the large language model to impose conditional constraints on the large language model, thereby obtaining the second intent recognition model.

[0158] S206: Call the second intent recognition model to perform language analysis processing on the text information to obtain the first intent information and key information corresponding to the text information.

[0159] S207: Based on the first intent information, determine the intent information of the session information, and use the intent information and the key information as the identification result.

[0160] Understandably, the conversation information can be parsed to understand its overall content and structure. The parsed conversation information is then mapped to a previously determined target intent template. This can be done through semantic role labeling, dependency parsing, or entity recognition to extract key semantic elements from the conversation information and match them with elements in the target intent template. Based on the mapping results, the intent information of the conversation information can be confirmed. Key information is extracted from the conversation information; this key information refers to information crucial for understanding the conversation intent or necessary for subsequent operations. This information may include entity names, attribute values, product information, etc.

[0161] Specifically, assuming the target intent template is "supplementary information," the first intent information is identified as "supplementing user information for a certain product," and the session information is "I want to supplement my address information; my address is in a certain city." Key elements in the session information, such as "supplement," "I," and "a certain city," are parsed. Because the semantic content of the session information highly matches the target intent template, the intent of the session information is confirmed as supplementary information. Key information, namely "a certain city" and "I," is extracted from the session information. "I" identifies the user who initiated the session. The intent information and key information are then combined to form the recognition result.

[0162] In one possible implementation, if the identification result is determined to be unrecognizable, an identification error message is generated.

[0163] Among these, situations where identification is not possible may include, but are not limited to: unclear intent, the session information being too vague or ambiguous, making it impossible to determine the specific intent; the session information lacking the key information necessary to perform the operation; template mismatch, that is, no existing intent template can match the intent of the session information; system limitations, that is, the current functional or resource limitations of the system prevent the processing of the session information, etc.

[0164] Understandably, the platform can write a clear and user-friendly error message based on the identified error type. The error message can include the following elements: clearly stating the reason that cannot be identified; using easy-to-understand language and avoiding technical jargon or overly complicated explanations; and, if possible, providing suggestions for solving the problem or guiding the user through subsequent operations.

[0165] For example, when the error is identified as an unclear intent, the response could be, "We cannot determine your specific request. Please try to rephrase your intent more clearly."; or, if key information is missing, the response could be, "To complete your request, we need more information. Please provide [the type of missing information], such as [a specific example]."

[0166] This application provides an after-sales service management method based on a large language model. When the intent recognition model fails to identify the query, the large language model is introduced to improve the corpus recognition model, enabling a better understanding of complex user queries, including context, tone, and underlying intent. This allows for more accurate query parsing and the handling of ambiguous or incomplete information, inferring missing parts from context, and reducing misunderstandings and repeated inquiries. The large model allows for rapid retrieval and integration of large amounts of information, providing more comprehensive and timely answers, reducing manual intervention, and extracting key information from large datasets to update and maintain the knowledge base, thereby improving the comprehension capabilities of corpus responses.

[0167] Figure 4The process of an after-sales service management method based on a large language model provided in this application embodiment Figure 2 .like Figure 4 As shown, this embodiment is... Figure 2 Based on the embodiments, the process of the platform confirming historical session data from the database is described in detail. This embodiment provides an after-sales service management method based on a large language model, including:

[0168] S301: Determine whether there is historical interaction data in the database that corresponds to the session information and the interval between the interaction time and the current time is less than the preset duration; if yes, proceed to step S303; if no, proceed to step S302.

[0169] The preset duration can be used to limit a series of session times, thereby improving the correlation between historical session data and session information, and enhancing the accuracy of semantic recognition.

[0170] S302: Based on the session information, use an intent recognition model to determine the recognition result corresponding to the session information.

[0171] In cases where it is determined that there is no historical interaction data in the database corresponding to the session information, and the interval between the interaction time and the current time is less than the preset duration, it is determined that the current session information has no context connection. At this time, based on the session information, the intent recognition model is used to determine the recognition result.

[0172] S303: Determine whether the number of historical interaction data is greater than the preset number; if yes, proceed to step S304; if no, proceed to step S305.

[0173] S304: Use a preset number of historical interaction data points whose interaction time is closest to the current time as historical session data.

[0174] In cases where it is determined that there are historical interaction data in the database that correspond to the session information and the interval between the interaction time and the current time is less than a preset duration, and the number of historical interaction data is greater than a preset number, a preset number of historical interaction data that are associated with the current session information can be filtered out and used as historical session data to prevent information redundancy from affecting the recognition results.

[0175] S305: Use historical interaction data as historical session data.

[0176] Understandably, if the amount of historical interaction data does not exceed the preset amount, the currently retrieved interaction data can be used as historical session data so that semantic recognition of the current session information can be performed based on the current historical session data.

[0177] Optionally, historical session data can also be used to provide more relevant product or service recommendations based on users' shopping history, search records, and other information.

[0178] This application provides an after-sales service management method based on a large language model. By combining session information and historical session data with context, the actual intent of the user can be inferred, so that the large model can better understand the user's query intent and context. Even if the expression is not completely clear or contains grammatical errors, the user's intent can be more accurately parsed, thereby improving the understanding ability of the after-sales service system.

[0179] Figure 5 This is a schematic diagram of the structure of an after-sales service management device based on a large language model, provided for this application. Figure 5 As shown, this application provides an after-sales service management device based on a large language model, applied to a platform. The after-sales service management device 400 based on a large language model includes:

[0180] The receiving module 401 is used to receive session information sent by the after-sales service department;

[0181] The determining module 402 is used to determine the recognition result corresponding to the session information based on the session information and the historical session data when the database contains historical session data corresponding to the session information, using a first intent recognition model. The recognition result includes key information and intent information, wherein the key information is used to indicate information in the session information that is associated with the target after-sales business.

[0182] The determining module 402 is further configured to determine the corresponding target speech configuration template based on the historical conversation data, the key information, and the intent information;

[0183] Module 403 is used to call the large language model and pre-train the large language model based on the target speech configuration template to obtain the response language model;

[0184] The calling module 403 is further configured to call the response language model to analyze and combine the key information, the intent information, and the target speech configuration template to obtain the response information corresponding to the conversation information;

[0185] Feedback module 404 is used to send the identification result and the response information back to the after-sales service department;

[0186] Optionally, the device may also include: an identification module 405;

[0187] The determining module 402 is further configured to determine the text information corresponding to the session information and the historical session data;

[0188] The recognition module 405 is used to perform intent recognition processing on the text information using the first intent recognition model.

[0189] The calling module 403 is also used to call a large language model to perform language analysis processing on the text information in the event that the first intent recognition model fails to recognize the text, so as to obtain the recognition result corresponding to the conversation information.

[0190] Optionally, the calling module 403 is further configured to call a large language model and pre-train the large language model based on multiple candidate intent templates to obtain a second intent recognition model;

[0191] The calling module 403 is further configured to call the second intent recognition model to perform language analysis processing on the text information to obtain the first intent information and key information corresponding to the text information;

[0192] The determining module 402 is further configured to determine the intent information of the session information based on the first intent information, and use the intent information and the key information as the identification result;

[0193] Optionally, the determining module 402 is further configured to determine the session scenario of the session information based on at least one of the key information, the intent information, and the historical session data;

[0194] The determining module 402 is further configured to determine the corresponding target dialogue configuration template from multiple candidate dialogue configuration templates based on the intent information and the conversation scenario;

[0195] Optionally, the device further includes: a determination module 406;

[0196] The determining module 402 is further configured to determine whether there is historical interaction data in the database that corresponds to the session information and the interval between the interaction time and the current time is less than a preset duration;

[0197] The judgment module 406 is used to determine whether the number of historical interaction data is greater than a preset number if the condition is met.

[0198] The determining module 402 is further configured to, when the number of historical interaction data is greater than a preset number, take a preset number of historical interaction data whose interaction time is closest to the current time as the historical session data.

[0199] The determining module 402 is further configured to use the historical interaction data as the historical session data when the number of historical interaction data is not greater than a preset number.

[0200] Optionally, the determining module 402 is further configured to, in the case that the historical session data does not exist in the database, determine the recognition result corresponding to the session information based on the session information using a first intent recognition model.

[0201] Figure 6 A schematic diagram of the structure of an after-sales service management device based on a large language model provided in this application. Figure 2 .like Figure 6 As shown, this application provides an after-sales service management device based on a large language model, applied to the after-sales service department. The after-sales service management device 500 based on a large language model includes:

[0202] Forwarding module 501 is used to forward session information sent by the user terminal to the platform terminal;

[0203] The receiving module 502 is used to receive the recognition result and response information corresponding to the session information fed back by the platform. The recognition result is determined by the first intent recognition model of the platform based on the session information and the corresponding historical session data. The recognition result includes key information and intent information. The key information is used to indicate information related to after-sales business in the session information. The response information is determined by the response language model of the platform based on the key information, the intent information and the script configuration template.

[0204] The determination module 503 is used to determine the after-sales service corresponding to the response information based on the key information and the intent information;

[0205] Feedback module 504 is used to provide the user terminal with the reply information and the after-sales service plan corresponding to the after-sales service.

[0206] Optionally, the determining module 503 is further configured to determine the corresponding after-sales service based on the key information and the intent information, and determine the business processing module corresponding to the after-sales service, wherein different intent information corresponds to different business processing modules;

[0207] The determining module 503 is further configured to determine the after-sales service plan based on the after-sales service, the service processing module, and the reply text.

[0208] Figure 7 This application provides a structural diagram of an after-sales service management device based on a large language model. (For example...) Figure 7 As shown, this application provides an after-sales service management device based on a large language model. The after-sales service management device 600 based on a large language model includes: a receiver 601, a transmitter 602, a processor 603, and a memory 604.

[0209] Receiver 601 is used to receive instructions and data;

[0210] Transmitter 602 is used to send commands and data;

[0211] Memory 604 is used to store instructions executed by the computer;

[0212] Processor 603 is used to execute computer execution instructions stored in memory 604 to implement the various steps of the after-sales service management method based on a large language model in the above embodiments. For details, please refer to the relevant descriptions in the foregoing embodiments of the after-sales service management method based on a large language model.

[0213] Alternatively, the memory 604 can be either standalone or integrated with the processor 603.

[0214] When the memory 604 is set up independently, the electronic device also includes a bus for connecting the memory 604 and the processor 603.

[0215] This application also provides a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the after-sales service management method based on a large language model, as described above, which is executed by the after-sales service management device based on a large language model.

[0216] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0217] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for after-sales service management based on a large language model, characterized in that, Applied to the platform side, including: Receive session information sent by the back-end support team; If historical session data corresponding to the session information exists in the database, the recognition result corresponding to the session information is determined based on the session information and the historical session data using a first intent recognition model. The recognition result includes key information and intent information, wherein the key information is used to indicate information in the session information that is associated with the target after-sales business. Based on the historical conversation data, the key information, and the intent information, the corresponding target dialogue configuration template is determined; The large language model is invoked, and pre-trained based on the target speech configuration template to obtain the response language model; The response language model is invoked to analyze and combine the key information, the intent information, and the target dialogue configuration template to obtain the response information corresponding to the conversation information, and the recognition result and the response information are fed back to the sales backend.

2. The method according to claim 1, characterized in that, The step of determining the recognition result corresponding to the session information using a first intent recognition model based on the session information and the historical session data includes: Determine the text information corresponding to the session information and the historical session data; The first intent recognition model is used to perform intent recognition processing on the text information; If the first intent recognition model fails to recognize the intent, a large language model is invoked to perform language analysis on the text information to obtain the recognition result corresponding to the conversation information.

3. The method according to claim 2, characterized in that, The step of calling a large language model to perform language analysis on the text information to obtain the recognition result corresponding to the conversation information includes: A large language model is invoked, and the large language model is pre-trained based on multiple candidate intent templates to obtain a second intent recognition model; The second intent recognition model is invoked to perform language analysis processing on the text information to obtain the first intent information and key information corresponding to the text information; Based on the first intent information, the intent information of the session information is determined, and the intent information and the key information are used as the identification result.

4. The method according to claim 1, characterized in that, The step of determining the corresponding target script configuration template based on the historical session data, the key information, and the intent information includes: The session scenario of the session information is determined based on at least one of the key information, the intent information, and the historical session data; Based on the intent information and the conversation scenario, the corresponding target dialogue configuration template is determined from multiple candidate dialogue configuration templates.

5. The method according to claim 1, characterized in that, Before determining the recognition result corresponding to the session information using an intent recognition model based on the session information and the historical session data, the method further includes: Determine whether there exists historical interaction data in the database that corresponds to the session information and the interval between the interaction time and the current time is less than a preset duration; If so, determine whether the number of historical interaction data is greater than a preset number; If the number of historical interaction data is greater than a preset number, a preset number of historical interaction data whose interaction time is closest to the current time will be used as the historical session data. If the amount of historical interaction data is not greater than a preset amount, the historical interaction data will be used as the historical session data.

6. The method according to claim 5, characterized in that, The method further includes: If the historical session data is not present in the database, the recognition result corresponding to the session information is determined using a first intent recognition model based on the session information.

7. A method for after-sales service management based on a large language model, characterized in that, Applications in after-sales service, including: Forward session information sent by the user to the platform. The system receives the recognition result and response information corresponding to the session information fed back by the platform. The recognition result is determined by the first intent recognition model of the platform based on the session information and the corresponding historical session data. The recognition result includes key information and intent information. The key information is used to indicate information related to after-sales business in the session information. The response information is determined by the response language model of the platform based on the key information, the intent information and the script configuration template. Based on the key information and the intent information, the after-sales service corresponding to the response information is determined, and the after-sales service plan corresponding to the response information and the after-sales service is fed back to the user terminal.

8. The method according to claim 7, characterized in that, The step of determining the after-sales service corresponding to the response information based on the key information and the intent information includes: Based on the key information and the intent information, the corresponding after-sales service is determined, and the corresponding business processing module for the after-sales service is determined, wherein different intent information corresponds to different business processing modules; Based on the after-sales service, the business processing module, and the response text, the after-sales service plan is determined.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the after-sales service management method based on a large language model as described in any one of claims 1 to 8.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the after-sales service management method based on a large language model according to any one of claims 1 to 8 through the computer program.

Images