Customer service assistance method and device based on large language model, electronic equipment and storage medium

Abstract

The invention discloses a customer service assistance method and device based on a large language model (LLM), electronic equipment and a storage medium, and relates to the technical field of computers. A specific embodiment of the method comprises the following steps: receiving an input from a client device of a user; retrieving candidate outputs of the input in a first database and a second database based on the input, where the first database and the second database are corpora enhanced with LLM; in response to determining that the candidate output is not retrieved, generating a candidate output of the input using the LLM; and providing the candidate output for presentation via a user interface of the client device. According to the embodiment, accurate output of the existing customer service scheme can be realized, and quick response of the unknown customer service scheme can be realized by virtue of the LLM, so that the customer service scheme can be efficiently provided for the user.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more particularly to a method, apparatus, electronic device, and storage medium for customer service assistance based on Large Language Model (LLM). Background Technology

[0002] In human-computer interaction (e.g., interaction between humans and computer devices such as laptops, telephones, and tablets), text input, as a crucial means of information transmission, typically needs to be converted into corresponding instructions so that the computer can understand and execute the user's intentions. With the rapid development of internet technology, text input plays an increasingly important role in numerous application scenarios (e.g., search engine systems, online customer service consultations, website comments and replies, content editing on knowledge-sharing platforms, and various social media applications).

[0003] However, with the increasing richness of online content and the accelerating pace of social interaction, the time cost users spend typing has become increasingly prominent. This is especially true in service scenarios such as online customer service, where customer service representatives often need to type lengthy explanatory text to resolve issues. Without efficient input assistance methods, this directly impacts service response speed and user experience.

[0004] Most existing input assistance products make recommendations based on users' personal input habits, and most are limited to word-level input suggestions. Customer service personnel can only rely on word-level acceleration features when entering answers.

[0005] Input association, a common input assistance technology, helps users save input time and operational costs by predicting their input. A typical intelligent input association system usually includes the following process: First, a corpus is built containing possible user input and related features; then, when the user inputs at the front-end user interface, the input content is captured and features are extracted. These features are then matched with features in the corpus to ultimately obtain and provide the corpus most consistent with the user's current input intent. However, the required corpus and functional requirements vary depending on the business scenario. Existing input association methods suffer from low business adaptability and insufficient practicality. Therefore, for specific business scenarios such as online customer service, dedicated input assistance systems are needed to achieve efficient customer service assistance that fits their business characteristics and supports rapid solution output. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method, apparatus, system, and storage medium for LLM-based customer service assistance. The present invention achieves automated management under different business scenarios through LLM fine-tuning, optimization and expansion of the customer service corpus, and construction of a predefined customer service solution library, providing high-quality and targeted assistance functions for different types of businesses. Simultaneously, for the problem of difficulty in obtaining solutions from the corpus, solutions can be generated in real time using LLM based on a domain knowledge base. Therefore, during customer service, both the accurate output of existing customer service solutions and the rapid response to unknown customer service questions can be achieved with the help of LLM, thus efficiently providing customer service solutions to users.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for LLM-based customer service assistance is provided.

[0008] An embodiment of the present invention provides a method for LLM-based customer service assistance, comprising: receiving input from a user's client device; retrieving candidate outputs of the input from a first database and a second database, wherein the first database and the second database are corpora enhanced with LLM; generating candidate outputs of the input using LLM in response to determining that no candidate outputs are retrieved; and providing the candidate outputs for presentation via a user interface of the client device.

[0009] In one embodiment, the LLM is generated by: performing a first-level fine-tuning on the base LLM using a first dataset to obtain a first preprocessed model (PLM), wherein the first dataset includes at least unlabeled customer service history dialogue data and domain knowledge base data; and performing a second-level fine-tuning on the first PLM using a second dataset to obtain a second PLM as the LLM, wherein the second dataset includes labeled candidate solution data.

[0010] In yet another embodiment, the method for LLM-based customer service assistance further includes: identifying whether the input is a command or a keyword; and performing semantic understanding on the command in response to the input being a command.

[0011] In yet another embodiment, retrieving candidate outputs from an input in a first database and a second database is based on identified instructions and / or keywords.

[0012] In yet another embodiment, determining that no candidate output was retrieved includes any of the following: determining that no candidate output was retrieved in either the first database or the second database; and determining that the relevance metric of the retrieved candidate output to the input is below a threshold.

[0013] In yet another embodiment, the LLM-enhanced corpus includes one or more of the following: first customer service solution data, wherein the first customer service solution data is optimized customer service solution data generated by adjusting customer service solution data using LLM; second customer service solution data, wherein the second customer service solution data is alternative customer service solution data similar to the customer service solution data generated using LLM; and third customer service solution data, wherein the third customer service solution data is predefined customer service solution data for a specific customer service question generated using LLM.

[0014] To achieve the above objectives, according to another aspect of the present invention, an apparatus for LLM-based customer service assistance is provided.

[0015] An apparatus for LLM-based customer service assistance according to the present invention includes: a receiving unit for receiving input from a user's client device; a retrieval unit for retrieving candidate outputs of the input in a first database and a second database based on the input, wherein the first database and the second database are corpora enhanced with LLM; a generating unit for generating candidate outputs of the input using LLM in response to determining that no candidate outputs are retrieved; and a providing unit for providing candidate outputs for presentation via a user interface of the client device.

[0016] In one embodiment, the LLM is generated by: performing a first-level fine-tuning on the base LLM using a first data set to obtain a first PLM, wherein the first data set includes at least unlabeled customer service history dialogue data and domain knowledge base data; and performing a second-level fine-tuning on the first PLM using a second data set to obtain a second PLM as the LLM, wherein the second data set includes labeled candidate solution data.

[0017] In yet another embodiment, the apparatus for LLM-based customer service assistance further includes an identification unit for: identifying whether the input is an instruction or a keyword; and performing semantic understanding on the instruction in response to the input being an instruction.

[0018] In yet another embodiment, retrieving candidate outputs from an input in a first database and a second database is based on identified instructions and / or keywords.

[0019] In yet another embodiment, determining that no candidate output was retrieved includes any of the following: determining that no candidate output was retrieved in either the first database or the second database; and determining that the relevance metric of the retrieved candidate output to the input is below a threshold.

[0020] In yet another embodiment, the LLM-enhanced corpus includes one or more of the following: first customer service solution data, wherein the first customer service solution data is optimized customer service solution data generated by adjusting customer service solution data using LLM; second customer service solution data, wherein the second customer service solution data is alternative customer service solution data similar to the customer service solution data generated using LLM; and third customer service solution data, wherein the third customer service solution data is predefined customer service solution data for a specific customer service question generated using LLM.

[0021] To achieve the above objectives, according to another aspect of the present invention, an electronic device is provided.

[0022] An electronic device according to an embodiment of the present invention includes: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the customer service assistance method for LLM-based systems provided in this embodiment of the present invention.

[0023] To achieve the above objectives, according to another aspect of the present invention, a computer-readable medium is provided.

[0024] An embodiment of the present invention provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method for LLM-based customer service assistance provided in the embodiment of the present invention.

[0025] During customer service, customer service personnel often need to manually type out lengthy solution texts or rely on general input assistance features. However, because general input assistance features are not designed with the specific expression habits and business content of particular business domains in mind, their support is limited and insignificant in customer service scenarios that require frequent input of lengthy, professional responses. Furthermore, because manual input makes it difficult to provide consistent answers to the same or similar questions, this can lead to the omission of crucial information. This situation may result in multiple rounds of repetitive communication.

[0026] While some customer service systems have predefined answers for frequently asked questions, as the number of predefined answers increases, customer service personnel find it difficult to quickly locate the appropriate answer in specific scenarios. Furthermore, customer service personnel may encounter problems they cannot resolve during the service process. They typically need to pause their current service and seek help from others or consult a knowledge base. All of these situations can lead to increased waiting times or unresolved issues.

[0027] The embodiments of this invention can solve the above-mentioned technical problems and have the following advantages or beneficial effects: By constructing an optimized and enhanced solution corpus, known solutions can be quickly matched and input with one click, significantly shortening the input time for customer service personnel. Simultaneously, for unknown problems, leveraging the real-time reasoning capabilities of large models and knowledge bases, high-quality solutions can be quickly generated and recommended, effectively reducing service waiting time and improving overall response speed. Providing recommended content through a structured and standardized solution corpus avoids content omissions or expression differences caused by manual input, ensuring the accuracy and consistency of responses to the same or similar problems, thereby improving problem-solving rates and user satisfaction.

[0028] The embodiments of this invention combine functional modules such as LLM fine-tuning, corpus optimization and enhancement, data management, and service provision to achieve end-to-end assistance from problem identification and solution matching to content generation and output, reducing reliance on human experience and particularly helping customer service personnel work more efficiently. Furthermore, the LLM-based customer service assistance method and device provided in the embodiments of this invention have good business adaptability and scalability. The solution corpus can be customized and optimized according to the characteristics of different customer service businesses, and relying on the generalization capability of LLM, it can continuously adapt to new scenarios and new problems, providing high-quality, targeted input assistance support for diverse customer service businesses.

[0029] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0030] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein: Figure 1 This is an application environment diagram of a customer service assistance method based on LLM according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for LLM-based customer service assistance according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the two-stage fine-tuning of an LLM according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the main units of an apparatus for LLM-based customer service assistance according to an embodiment of the present invention; Figure 5 is a schematic diagram of the structure of a computer system suitable for implementing embodiments of the present invention. Detailed Implementation

[0031] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0032] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0033] This invention provides a method for customer service assistance based on LLM, which can be used in scenarios where customer service personnel provide services.

[0034] Figure 1 This is an application environment diagram for a customer service assistance method based on LLM according to an embodiment of the present invention. Figure 1 The system architecture 100 shown includes multiple terminal devices 101, 102, and 103, a network 105, and a server 104. Network 105 serves as the medium for providing communication links between the terminal devices 101, 102, and 103 and the server 104. Network 105 can include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0035] Users can use terminal devices 101, 102, and 103 to interact with server 104 via network 105 to receive or send messages, etc. Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers. Various client applications, such as customer service applications, can be installed on terminal devices 101, 102, and 103 (for example only).

[0036] Server 104 can be a server that provides various services, such as generating customer service solutions (for example only) based on input (e.g., customer service questions) sent by users using terminal devices 101, 102, and 103. The server can process the received input and feed back the processing results (e.g., candidate customer service solutions - for example only) to the terminal devices.

[0037] It should be noted that the method for LLM-based customer service assistance provided in the embodiments of the present invention is generally executed by server 104, and correspondingly, the device for LLM-based customer service assistance is generally located in server 104.

[0038] It should be understood that Figure 1The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0039] This invention provides a method for customer service assistance based on LLM, which can be implemented by... Figure 1 The server 104 in the system architecture 100 shown executes, as follows: Figure 2 As shown, the method includes the following steps.

[0040] S201: Receive input from the user's client device.

[0041] In one embodiment, the user can be a customer service representative. Input from the user's client device can include the question the user wants to inquire about, such as a customer service representative's inquiry in an online customer service setting. The user can type the text or speak the input via a user interface.

[0042] S202: Based on the input, retrieve candidate outputs of the input in the first database and the second database.

[0043] The first and second databases are corpora enhanced using LLM.

[0044] In one embodiment, the LLM utilized can be generated through two-stage fine-tuning. The data used for fine-tuning the LLM can be collected before the two-stage fine-tuning. This can include customer service history dialogue data, domain knowledge base data, and manually annotated candidate solution data.

[0045] Customer service history dialogue data primarily originates from historically accumulated customer service records. This unlabeled historical dialogue data records genuine user inquiries and customer service responses, covering various business scenarios and language expression habits. These records can be preprocessed (e.g., data cleaning, anonymization, and deduplication), and the preprocessed records can then be used as historical customer service dialogue data.

[0046] Domain knowledge base data may come from multiple sources, such as internal product manuals, standard operating procedures, frequently asked questions, and solution libraries.

[0047] The labeled candidate solution data may originate from high-quality customer service history records and knowledge bases. Each data point includes task data, solution data, and its corresponding manually labeled standard customer service solution. Tags may include question intent, business scenario, keywords, and solutions.

[0048] Figure 3 This is a schematic diagram of the two-stage fine-tuning of an LLM according to an embodiment of the present invention. (Reference) Figure 3As shown in diagram 300, to generate an LLM, the base LLM can first be fine-tuned using a first dataset to obtain a first PLM. The first dataset can include at least the aforementioned unlabeled customer service historical dialogue data and domain knowledge base data. This unlabeled data can be used to perform first-level full-parameter fine-tuning of the base LLM (e.g., ChatGPT, ChatGLM, and JD LLM). The base LLM can be a general, unoptimized starting model. It possesses broad language understanding and generation potential, but its knowledge and style are not specifically aligned to a vertical domain such as customer service. Therefore, first-level fine-tuning is needed to acquire relevant knowledge and capabilities in the customer service domain. First-level fine-tuning can be unsupervised.

[0049] Furthermore, after obtaining the first PLM, a second-level fine-tuning is performed on the first PLM using a second dataset to obtain a second PLM as an LLM. The second dataset includes labeled candidate solution data. The labeled candidate solution data may be used to train the first PLM model in a supervised fine-tuning manner. The second-level fine-tuning can be incremental. The generated second PLM is aligned with the customer service solution capabilities, thus obtaining an LLM (i.e., the second PLM) that can be directly applied to customer service assistance.

[0050] Back Figure 2 In step S202, using the obtained LLM, an enhanced corpus can be generated and stored in the first database and the second database. The LLM-enhanced corpus may include one or more of the following: first customer service plan data, wherein the first customer service plan data is optimized customer service plan data generated by adjusting the customer service plan data using the LLM; second customer service plan data, wherein the second customer service plan data is alternative customer service plan data similar to the customer service plan data generated using the LLM; and third customer service plan data, wherein the third customer service plan data is predefined customer service plan data for a specific customer service question generated using the LLM.

[0051] In one embodiment, a prompt for optimizing customer service solution data and the customer service solution data to be optimized can be input into a two-level fine-tuned LLM. The LLM adjusts the original scripts, processes, and other content in the customer service solution data; this process is a single-input, single-output generation mode. The output optimized customer service solution data is used as the first customer service solution data.

[0052] In another embodiment, the prompts for expanding customer service solution data and the combined customer service solution data can be input into the second-level fine-tuned LLM. The LLM can generate and output one or more alternative customer service solution data, which can be semantically similar variants of the original solution or other reasonable solutions to the same problem, thereby achieving a single-input, multiple-output generation mode. The output alternative customer service solution data is used as the second customer service solution data.

[0053] In a further embodiment, specific customer service questions, such as difficult and / or frequently asked questions, can be identified as input to the LLM model. A preliminary solution is generated from the second-level fine-tuned LLM model combined with a domain knowledge base. Subsequently, the preliminary solution is manually reviewed, and high-quality solutions that meet the standards are used as predefined customer service solution data. This predefined customer service solution data is then used as third-level customer service solution data.

[0054] One or more of the first, second, and third customer service solution data can be stored as an enhanced corpus in a first and a second database for use in online customer service. In one embodiment, the first database can be an Elasticsearch (ES) database. The first database can store and index structured customer service solution data in the customer service domain, and its powerful full-text search capabilities enable fast and accurate keyword-based matching. Meanwhile, the second database can be a FAISS database. The second database can store vector embeddings corresponding to text, and its efficient similarity search algorithm enables semantic-based vectorized retrieval. The first and second databases work together to achieve multi-level intelligent recommendation of customer service solutions from keywords to semantics.

[0055] Understandably, the aforementioned data collection, secondary fine-tuning of the basic LLM, and generation of the enhanced corpus were all carried out offline.

[0056] In one embodiment, in response to receiving input from a user, the input can be identified as either an instruction or a keyword. The input can be parsed in real time to determine whether it is an instruction or a scheme keyword. Instructions can be status instructions, such as "urging payment," "overdue," etc. Scheme keywords can be words or subsequences in candidate scheme data. In one embodiment, a user can explicitly indicate that the input is an instruction by typing a specific character (e.g., #) before typing the text. Accordingly, the user's input can be identified as an instruction based on the specific character, otherwise as a keyword. It should be noted that this identification method is not limited to this, and any other common identification methods in the art can be applied. In a further embodiment, in response to the input being an instruction, semantic understanding can be performed on the instruction. The instruction is mapped to corpus tags based on its semantics. When the input is a scheme keyword, no processing is performed on the input. When the input contains both instructions and keywords, only the instructions are semantically understood and mapped to corpus tags.

[0057] Retrieving candidate outputs from the input in the first and second databases is based on the identified instructions and / or keywords. Candidate outputs are retrieved from the first and second databases based on corpus tags of the instructions and / or scheme keywords. Specifically, if the input is an instruction, a tag retrieval method is used. If the input is a scheme keyword, a keyword retrieval method is used. If the input contains both instructions and keywords, a multi-condition combined retrieval is used. Candidate outputs can be answer schemes for customer service questions.

[0058] S203: In response to determining that no candidate output was found, generate candidate outputs for the input using LLM.

[0059] In one embodiment, determining that no candidate output was retrieved includes any of the following: determining that the candidate output was not retrieved in either the first database or the second database; and determining that the retrieved candidate output's relevance metric to the input is below a threshold. The threshold can be manually set. If there is no customer service response plan in the first database or the second database, or if the retrieved customer service response plan is too low in relevance to the question, an LLM-based customer service response plan is generated in real time using a domain knowledge base. Correspondingly, if a candidate output is retrieved in the first database or the second database, the retrieved candidate output is used as the customer service response plan.

[0060] S204: Provide candidate outputs for presentation via the user interface of the client device.

[0061] Before providing candidate outputs, the candidate outputs obtained in S202-S203 can be sorted. In one embodiment, after recalling multiple customer service response schemes, the multiple customer service response schemes can be sorted according to the customer service chat context and associated status information (order status, order items, and logistics status, etc.), and a certain number (e.g., 3) of the highest-ranked customer service response schemes can be provided for presentation on the client device for customer service personnel to select. The number of customer service response schemes provided can be manually configured.

[0062] In this embodiment of the invention, the LLM-based customer service assistance mechanism consists of offline processing and online service. Offline processing includes data collection, two-level fine-tuning of the LLM, and corpus enhancement. Two-level fine-tuning of the LLM enables the large model to acquire customer service domain-related knowledge and capabilities. The enhanced solution corpus enriches the candidate solution data, increases the solution recall rate for questions, and thus improves the input efficiency of customer service.

[0063] The online service performs semantic understanding, database retrieval, and real-time generation and ranking of customer service solutions using LLM. It enables the retrieval of existing solutions and the instant generation of unknown solutions. This allows for rapid responses to different business scenarios, reducing instances of customer service staff being unable to resolve issues or providing incomplete solutions due to insufficient knowledge.

[0064] To address the problems existing in the prior art, embodiments of the present invention provide a device 400 for customer service assistance based on LLM, such as... Figure 4 As shown, the device 400 includes: The receiving unit 401 is used to receive input from the user's client device; The retrieval unit 402 is used to retrieve candidate outputs of the input in a first database and a second database based on the input, wherein the first database and the second database are corpora enhanced with LLM; Generation unit 403 is configured to generate candidate outputs for the input using LLM in response to determining that no candidate output has been retrieved; and Unit 404 is provided for providing candidate outputs for presentation via a user interface of a client device.

[0065] In one embodiment, the LLM is generated by: performing a first-level fine-tuning on the base LLM using a first data set to obtain a first PLM, wherein the first data set includes at least unlabeled customer service history dialogue data and domain knowledge base data; and performing a second-level fine-tuning on the first PLM using a second data set to obtain a second PLM as the LLM, wherein the second data set includes labeled candidate solution data.

[0066] In yet another embodiment, the apparatus for LLM-based customer service assistance further includes an identification unit for: identifying whether the input is an instruction or a keyword; and performing semantic understanding on the instruction in response to the input being an instruction.

[0067] In yet another embodiment, retrieving candidate outputs from an input in a first database and a second database is based on identified instructions and / or keywords.

[0068] In yet another embodiment, determining that no candidate output was retrieved includes any of the following: determining that no candidate output was retrieved in either the first database or the second database; and determining that the relevance metric of the retrieved candidate output to the input is below a threshold.

[0069] In yet another embodiment, the LLM-enhanced corpus includes one or more of the following: first customer service solution data, wherein the first customer service solution data is optimized customer service solution data generated by adjusting customer service solution data using LLM; second customer service solution data, wherein the second customer service solution data is alternative customer service solution data similar to the customer service solution data generated using LLM; and third customer service solution data, wherein the third customer service solution data is predefined customer service solution data for a specific customer service question generated using LLM.

[0070] It should be understood that the manner in which embodiments of the present invention are implemented is different from the implementation method. Figure 2 The embodiments shown are the same and will not be described again here.

[0071] According to embodiments of the present invention, the present invention also provides an electronic device and a readable storage medium.

[0072] The electronic device of the present invention includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the method for LLM-based customer service assistance provided in embodiments of the present invention.

[0073] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system 500 suitable for implementing embodiments of the present invention. Figure 5 The computer system shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0074] like Figure 5As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage section 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the system 500. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0075] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 510 as needed so that computer programs read from it can be installed into storage section 508 as needed.

[0076] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs the functions defined above in the system of this invention.

[0077] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0078] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a unit, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0079] The units described in the embodiments of the present invention can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor can be described as including an acquisition unit, a receiving unit, a determining unit, and a display unit. The names of these units do not necessarily limit the specific unit; for example, an acquisition unit can also be described as "a unit that performs the function of an acquisition unit."

[0080] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to perform the LLM-based customer service assistance method provided by the present invention.

[0081] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for large language model (LLM) based customer service assistance, characterized in that, include: Receive input from the user's client device; Based on the input, candidate outputs of the input are retrieved from a first database and a second database, wherein the first database and the second database are corpora enhanced with LLM; In response to determining that the candidate output is not found, the candidate output of the input is generated using the LLM; and The candidate output is provided for presentation via the user interface of the client device.

2. The method of claim 1, wherein, The LLM is generated through the following: The basic LLM is fine-tuned using a first dataset to obtain a first preprocessed model (PLM), wherein the first dataset includes at least unlabeled customer service historical dialogue data and domain knowledge base data; and The first PLM is fine-tuned at a second level using a second data set to obtain a second PLM as the LLM, wherein the second data set includes labeled candidate scheme data.

3. The method of claim 1, wherein, Further includes: Identify whether the input is a command or a keyword; as well as In response to the input being the instruction, the instruction is semantically understood.

4. The method of claim 3, wherein, Retrieving candidate outputs from the input in the first and second databases is based on the identified instructions and / or keywords.

5. The method of claim 4, wherein, The candidate output not being retrieved includes any of the following: It was determined that the candidate output was not found in either the first database or the second database; as well as The relevance metric of the retrieved candidate output relative to the input is determined to be below a threshold.

6. The method according to any one of claims 1-5, characterized in that, The corpus enhanced by the LLM includes one or more of the following: First customer service solution data, wherein the first customer service solution data is optimized customer service solution data generated by adjusting the customer service solution data using the LLM; Second customer service solution data, wherein the second customer service solution data is alternative customer service solution data similar to the customer service solution data generated using the LLM; and The third customer service solution data is predefined customer service solution data generated using the LLM for specific customer service issues.

7. A device for customer service assistance based on Large Language Model (LLM), characterized in that, include: The receiving unit is used to receive input from the user's client device; A retrieval unit is configured to retrieve candidate outputs of the input in a first database and a second database based on the input, wherein the first database and the second database are corpora enhanced with LLM; A generation unit, configured to generate the candidate output of the input using the LLM in response to determining that the candidate output has not been retrieved; and A providing unit is configured to provide the candidate output for presentation via the user interface of the client device.

8. The apparatus according to claim 7, characterized in that, The LLM is generated through the following: The basic LLM is fine-tuned using a first dataset to obtain a first preprocessed model (PLM), wherein the first dataset includes at least unlabeled customer service historical dialogue data and domain knowledge base data; and The first PLM is fine-tuned at a second level using a second data set to obtain a second PLM as the LLM, wherein the second data set includes labeled candidate scheme data.

9. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images