Intelligent customer service platform development method based on large model and computer equipment

Through large-scale model technology, custom prompt words and preset examples are used to classify and process intentions, which solves the problem that a large number of professional talents is required for the development of intelligent customer service system, and realizes an intelligent system suitable for diversified business scenarios, reducing the complexity of development and cost.

CN119088960AInactive Publication Date: 2024-12-06联通(山西)产业互联网有限公司
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411051767.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing intelligent customer service system has technical barriers in maintenance and deployment, which makes it difficult for subsidiaries to customize systems that meet their own business needs and requires a large number of professional talents to develop.

Method used

Using the intelligent customer service platform development method based on the big model, through customized prompt word engineering and preset input and output examples, the big model can learn the problem context, identify intentions and classify them into unknown intentions, simple intentions and complex intentions, and adopt different methods. The processing methods include automatic conversion to manual services, similarity matching and multi-round dialogue technology, providing customized solutions.

Benefits of technology

It significantly reduces the complexity and expense of customized development, solves the contradiction between insufficient technical capabilities of subsidiaries and diversified business needs, and provides an intelligent system suitable for various segmented scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088960A_ABST
    Figure CN119088960A_ABST
Patent Text Reader

Abstract

The invention discloses an intelligent customer service platform development method based on a large model, and the method comprises the following steps: S1, enabling the large model to learn a problem context through providing a customized prompt word project and a preset input and output example; s2, performing intention recognition on a question input by a user by the large model, classifying the question into an unknown intention, a simple intention and a complex intention according to the complexity of the intention, and forming an obtained classification result into a mapping of'intention-intention classification '; s3, the large model adopts different processing modes according to different intention classifications; and S4, ending the session. According to the invention, the intelligent system suitable for various subdivided scenes is constructed, so that the complexity and cost of customized development are remarkably reduced, and the contradiction between insufficient technical capability and diversified business requirements of sub-companies is effectively solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent customer service question and answer systems, and in particular to a method for developing an intelligent customer service platform based on a large model and computer equipment. Background Art

[0002] The current customer service system mainly relies on language models, knowledge graphs, semantic matching and other technologies when handling professional consultations. These technologies not only require a lot of time and resources to develop and customize, but also require professional talents in related fields.

[0003] Take operators as an example. Their companies are usually divided into group headquarters and city branches. The head office usually has an independent technical development department that can customize systems for the group. These customized systems often can only achieve some general functions because they need to consider the applicability of the entire company. As the front line of the company's business, subsidiaries face a variety of business needs and need to customize systems based on their actual business conditions. Since subsidiaries are at the bottom of the company's structure, they often do not have independent technical development departments, which makes it difficult for subsidiaries to customize systems that meet their own business needs based on their own business characteristics. Therefore, the existing intelligent customer service system has certain technical barriers in maintenance and deployment, which restricts the expansion of the customer service system in terms of application scope. Summary of the invention

[0004] The present invention provides a large-model-based intelligent customer service platform development method and computer equipment that effectively improve the universality and usability of the customer service system, which can solve at least one of the above-mentioned technical problems.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A method for developing an intelligent customer service platform based on a large model, comprising the following steps:

[0007] S1: Allow the large model to learn the context of the question by providing customized prompt word engineering and preset input and output examples;

[0008] S2: The large model recognizes the intent of the user input question and classifies it into unknown intent, simple intent, and complex intent according to its complexity. Figure 3 Classify the classification results and form a "intent: intent classification" mapping;

[0009] S3: Depending on the intent classification, the large model adopts different processing methods, further including:

[0010] S31: For unknown intent, automatically transfer to manual service;

[0011] S32: For simple intents, the user input question is matched with the question data in the read knowledge base for similarity, the most similar question to the user input question is determined, the corresponding solution is retrieved in the knowledge base based on the most similar question, and the corresponding answer or guidance is provided to the user;

[0012] S33: Further segment the user input question for complex intents, enter the corresponding segmented question scenario according to different segmented questions, and output corresponding solutions;

[0013] S4: The session ends.

[0014] Furthermore, S2 includes two intent classification tasks, namely:

[0015] S21: First intent classification task: classify the question raised by the user into one of unknown intent, simple intent or complex intent;

[0016] S22: The second intent classification task: further subdivide the complex intent.

[0017] Furthermore, in S32, the large model uses prompt learning technology and context learning technology to convert the downstream task into a text generation task;

[0018] Prompt learning technology introduces "prompt information" in the input field to understand the different expressions of different users and give accurate responses without significantly changing the structure and parameters of the pre-trained language model;

[0019] Contextual learning techniques work by providing a language model with a “hint” consisting of input-output pairs for a specific task.

[0020] Furthermore, in S32, the knowledge base includes a plurality of different problems N and their corresponding solutions N.

[0021] Furthermore, in S33, the large model uses keyword extraction technology and multi-round dialogue technology to continuously clarify key information through word slots in the form of questions and answers, further including:

[0022] S331: For the question categorized as complex intent input by the user, the large model performs question segmentation and enters the segmented question scenario corresponding to the segmented question. Different segmented question scenarios have different key information that needs to be clarified.

[0023] S332: The large model gradually guides users to clarify key information through preset question statements for different problem scenarios, conducts multiple rounds of dialogue, and asks users targeted questions to guide them to provide detailed information to fill in the word slots;

[0024] S333: The large model uses named entity recognition technology to find the sub-nodes driven by the current segmentation problem and output the corresponding solutions.

[0025] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above-mentioned method for developing an intelligent customer service platform based on a large model.

[0026] The beneficial effects of the present invention are embodied in:

[0027] This invention introduces big model technology and builds an intelligent customer service development platform, aiming to solve the problem that the development of traditional customer service systems requires a large number of professional talents, while subsidiaries do not have relevant talents. By providing customized prompt word engineering and presetting a large number of input and output examples, the big model learns the question context and completes the task of classifying the intent of the questions raised by users, distinguishing them into unknown intent, simple intent, and complex intent. Figure 3 There are large categories, and each category is intended to correspond to different question-and-answer scenarios and processing flows. In this way, there is no need to master the large model language fine-tuning technology to build an intelligent system suitable for various segmented scenarios, thereby significantly reducing the complexity and cost of customized development, and effectively resolving the contradiction between the subsidiary's insufficient technical capabilities and diversified business needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0029] Figure 1 It is a schematic diagram of the overall process of the method for developing an intelligent customer service platform based on a large model according to an embodiment of the present invention.

[0030] Figure 2 It is a specific flowchart diagram of an embodiment of the present invention that adopts different processing methods according to different intent classifications.

[0031] Figure 3 It is an overall flow chart of the method for developing an intelligent customer service platform based on a large model according to an embodiment of the present invention.

[0032] Figure 4 It is an overall flow chart of the implementation of the embodiment of the present invention in a real case.

[0033] Figure 5 It is a structural block diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0035] It should be noted that the meaning of "and / or" appearing in the full text includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or solutions that A and B meet at the same time. In addition, "multiple" refers to more than two. In addition, it can be understood by ordinary technicians in this field that all or part of the steps implemented in the embodiments of the present invention can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)).

[0036] See also Figure 1-2 The embodiment of the present invention provides a method for developing an intelligent customer service platform based on a large model, comprising the following steps:

[0037] S1: Allow the large model to learn the context of the question by providing customized prompt word engineering and preset input and output examples;

[0038] S2: The large model recognizes the intent of the user input question and classifies it into unknown intent, simple intent, and complex intent according to its complexity. Figure 3Classify the classification results and form a "intent: intent classification" mapping;

[0039] S3: Depending on the intent classification, the large model adopts different processing methods, further including:

[0040] S31: For unknown intent, automatically transfer to manual service;

[0041] S32: For simple intents, the user input question is matched with the question data in the read knowledge base for similarity, the most similar question to the user input question is determined, the corresponding solution is retrieved in the knowledge base based on the most similar question, and the corresponding answer or guidance is provided to the user;

[0042] S33: For complex intents, further subdivide the user input question, and enter the corresponding subdivided question scenario according to different subdivided questions. In this embodiment, fault classification is taken as an example, and fault classification is performed on the user input question to confirm the fault entity. According to different fault types, the corresponding fault scenario is entered, and the user is guided to clarify key information according to the current fault scenario, and a corresponding solution is output;

[0043] S4: The session ends.

[0044] The present invention provides a customized prompt word project and presets a large number of input and output examples, allowing the large model to learn the question context and complete the task of classifying the intent of the questions raised by the user, distinguishing them into unknown intent, simple intent and complex intent. Figure 3 Large categories, each category intention corresponds to different question-and-answer scenarios and processing flows. This simplifies the traditional customer service system's high development and maintenance costs of text recognition, semantic classification, semantic matching, knowledge graphs and other technologies. There is no need to master large model language fine-tuning technology to build an intelligent system suitable for various segmented scenarios, thereby significantly reducing the complexity and cost of customized development, and effectively resolving the contradiction between the subsidiary's insufficient technical capabilities and diversified business needs.

[0045] See also Figure 3 In this embodiment, S2 includes two intent classification tasks, namely:

[0046] S21: The first intent classification task: classify the questions raised by the user into unknown questions, simple questions, or complex questions. At this stage, based on the premise of learning the question context of the big model, the questions raised by the user will be divided into the above three categories in order to perform different processing procedures later;

[0047] S22: The second intention classification task: further subdivide the complex intent. This embodiment takes fault classification as an example to classify complex problems. At this stage, based on the premise that the user input problem is classified as a complex problem based on the large model, the fault category identified in the processing flow enters different fault scenarios, with the aim of further classifying the user input problem.

[0048] Through the above two classifications, the large model can more accurately understand the questions raised by users and provide more targeted solutions for the intelligent customer service system.

[0049] See also Figure 3 In this embodiment, in S32, the large model uses prompt learning technology and context learning technology to convert the downstream task into a text generation task;

[0050] Prompt learning technology introduces "prompt information" in the input field to understand the different expressions of different users and give accurate responses without significantly changing the structure and parameters of the pre-trained language model;

[0051] Contextual learning techniques work by providing a language model with a “hint” consisting of input-output pairs for a specific task.

[0052] In traditional customer service question-and-answer systems, semantic matching technologies such as Bert are generally used. However, different users have different ways of expression, which can cause problems with semantic matching. Therefore, in the present invention, developers input prompt word projects and some preset examples into the big model, allowing the big model to learn a large amount of question context data sets. Under this premise, the big model uses prompt learning technology and context learning technology to convert and learn user input questions. While answering questions, it is also learning new question context data, constantly expanding the model's own data set.

[0053] See also Figure 3 In this embodiment, in S32, the knowledge base includes a plurality of different problems N and their corresponding solutions N, for example, problems A, B, C, D..., respectively corresponding to solutions A, B, C, D...;

[0054] The "prompt information" introduced by the prompt learning technology allows the large model to understand different user input questions;

[0055] The input of the input-output pair in the "prompt" provided by the contextual learning technology is the input example question, and the output is the most similar question that matches the input example question in the knowledge base, forming a mapping of "input question: matching the most similar question in the knowledge base", allowing the large model to learn the corresponding answer methods for different user input questions.

[0056] See also Figure 2-3 In this embodiment, in S33, the large model uses keyword extraction technology and multi-round dialogue technology to continuously clarify key information through word slots in the form of questions and answers, further including:

[0057] S331: For the question categorized as complex intent input by the user, the big model performs question segmentation and enters the segmented question scenario corresponding to the segmented question. In the following, this example takes fault classification as an example, the big model performs fault classification and enters the fault scenario flow chart corresponding to the fault category. Different fault scenarios have different key information that needs to be clarified.

[0058] S332: The large model gradually guides users to clarify key information through preset question statements for different fault scenarios, conducts multiple rounds of dialogue, and asks users targeted questions to guide them to provide detailed information to fill in the word slots;

[0059] S333: The large model uses named entity recognition technology to find the sub-nodes driven by the current fault flowchart and output the corresponding solutions.

[0060] Under the premise that the user input question is classified as a complex intent, it is necessary to first classify the complex question raised by the current user to determine which type of fault the current question belongs to, because different types of faults correspond to different fault scenarios, and different fault scenarios require different key information that users need to clarify, so as to more accurately understand the customer's intention. In multiple rounds of dialogue, this intelligent customer service system gradually guides customers to clarify key information based on existing knowledge nodes, so as to continuously fill the word slots, and finally output solutions using named entity recognition technology, such as NER. For example, when the user inputs "My network is faulty", the large model guides the user to state that the current device is a "Windows computer" or "mobile phone" based on the existing NER data. If the customer answers "The faulty device I use here is a Windows computer", the intelligent customer service system first identifies "Windows computer" through NER, and then drives into the "Windows computer" sub-node according to the process, that is, network fault diagnosis tools and network settings and connection management. If the customer answers "The faulty device I use here is a mobile phone", the intelligent customer service system first identifies "mobile phone" through NER, and then drives into the "mobile phone" sub-node according to the process, that is, mobile data settings and restarting the device. This clarification of key information helps to locate the problem more precisely and provide more effective solutions.

[0061] See also Figure 3In this embodiment, in S33, when processing complex intents, the large model subdivides multiple fault types, including at least network faults, signal faults, and software faults. Taking the complex question "My broadband cannot access the Internet" as an example, when the user inputs "My broadband cannot access the Internet", the large model gives the prediction result "Network Failure" based on the prompt word engineering and examples.

[0062] See also Figure 4 In the daily customer service consultation of operators, network problems occur frequently. This embodiment takes the business scenario that a user calls the operator's intelligent customer service to reflect that the home broadband network cannot be connected. This intelligent customer service system quickly identifies user problems and provides common troubleshooting solutions related to network connections. The large model system provides a reference solution, and the specific implementation steps are as follows:

[0063] Step 1: The user submits “My home broadband cannot access the Internet.”

[0064] Step 2: Developers use preset prompts, such as "I am a network engineer. Please classify the input sentence as one of [unknown intent, simple intent, complex intent].", and examples to let the big model learn problem scenarios;

[0065] When the user inputs "My broadband can't access the Internet", the big model determines that the user input question is a "complex intent" based on the prompt word engineering and examples. Finally, the big model outputs the intent and complex intent classification mapping: "My broadband can't access the Internet => complex intent". This example enters the complex intent.

[0066] Step 3: Based on the classification of intent, perform the following operations:

[0067] Step 3.1: When the intent is classified as unknown, it is transferred to manual service.

[0068] Step 3.2: When the intent is classified as a simple intent, enter the simple intent processing flow:

[0069] Step 3.2.1: Developers use preset prompt words, i.e., "I am a network engineer. Please find out what question best matches the input question?", and examples to let the big model learn the problem scenario;

[0070] Step 3.2.2: Take the simple question "How to pay for broadband" as an example. When the customer enters "How to pay for broadband", the big model gives the prediction result "The question that best matches the input question is (question A or question B)" based on the prompt word engineering and examples.

[0071] Step 3.2.3: The intelligent customer service system finds the solution to the corresponding problem in the knowledge base based on the prediction results of the large model.

[0072] Step 3.3: When the intent is classified as a complex intent, enter the complex intent processing flow:

[0073] Step 3.3.1: First, the input problem needs to be classified to determine which type of fault the current problem belongs to. Secondly, different scenarios are entered according to different fault types. Then the user is guided to clarify key information and finally the corresponding solution is output. The developer uses preset prompt words, i.e. "I am a network engineer. Determine whether the current problem belongs to network fault, signal fault or software fault", as well as examples to let the big model learn the problem scenario.

[0074] Step 3.3.2: Take the complex problem "My broadband can't access the Internet" as an example. When the user enters "My broadband can't access the Internet", the large model gives the prediction result "network failure" based on the prompt word engineering and examples.

[0075] Step 3.3.3: The large model is driven by the process and enters the "network failure" process. The user is also required to clarify whether he is using a Windows computer or a mobile phone. The intelligent customer service system asks the user "Are you using a Windows computer or a mobile phone?";

[0076] Step 3.3.4: Developers use preset prompt words, i.e., "I am a network engineer, please extract the device entity in the customer's input sentence" and examples to let the big model learn the problem context;

[0077] Step 3.3.5: Taking the user feedback "Windows computer cannot access the Internet" obtained by the system as an example, the large model uses NER technology to identify "Windows computer" as a device entity;

[0078] Step 3.3.6: The system process driver provides a corresponding solution by identifying the device entity "Windows computer".

[0079] Step 4: The session ends.

[0080] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the above method.

[0081] See also Figure 5 An embodiment of the present invention further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0082] An embodiment of the present invention also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the methods for developing an intelligent customer service platform based on a large model in the above-mentioned embodiments.

[0083] It is understandable that the system, storage medium, device and program product provided by the embodiments of the present invention correspond to the method provided by the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts of the above methods.

[0084] In summary, the present invention introduces big model technology and builds an intelligent customer service development platform, aiming to solve the problem that the development of traditional customer service systems requires a large number of professional talents, while subsidiaries do not have relevant talents. By providing customized prompt word engineering and presetting a large number of input and output examples, the big model learns the question context and completes the task of classifying the intent of the questions raised by users, distinguishing them into unknown intent, simple intent, and complex intent. Figure 3 There are large categories, and each category is intended to correspond to different question-and-answer scenarios and processing flows. In this way, there is no need to master the large model language fine-tuning technology to build an intelligent system suitable for various segmented scenarios, thereby significantly reducing the complexity and cost of customized development, and effectively resolving the contradiction between the subsidiary's insufficient technical capabilities and diversified business needs.

[0085] It should be understood that the examples and implementation modes described herein are for illustrative purposes only and are not intended to limit the present invention. Those skilled in the art may make various modifications or changes based on the examples and implementation modes. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for developing an intelligent customer service platform based on a large model, characterized in that: The following steps are involved: S1: Allow the large model to learn the context of the question by providing customized prompt word engineering and preset input and output examples; S2: The large model identifies the intent of the user input question and classifies it into three categories according to the complexity of the intent: unknown intent, simple intent, and complex intent. The classification results are then mapped into an "intent: intent classification" map. S3: Depending on the intent classification, the large model adopts different processing methods, further including: S31: For unknown intent, automatically transfer to manual service; S32: For simple intents, the user input question is matched with the question data in the read knowledge base for similarity, the most similar question to the user input question is determined, the corresponding solution is retrieved in the knowledge base based on the most similar question, and the corresponding answer or guidance is provided to the user; S33: Further segment the user input question for complex intents, enter the corresponding segmented question scenario according to different segmented questions, and output corresponding solutions; S4: The session ends.

2. The method for developing an intelligent customer service platform based on a large model as claimed in claim 1, characterized in that: S2 includes two intent classification tasks, namely: S21: First intent classification task: classify the question raised by the user into one of unknown intent, simple intent or complex intent; S22: The second intent classification task: further subdivide the complex intent.

3. The method for developing an intelligent customer service platform based on a large model as claimed in claim 1, characterized in that: In S32, the large model uses prompt learning technology and context learning technology to convert the downstream task into a text generation task; Prompt learning technology introduces "prompt information" in the input field, without significantly changing the structure and parameters of the pre-trained language model, to understand the different expressions of different users and give accurate responses; Contextual learning techniques work by providing a language model with a "hint" consisting of input-output pairs for a specific task.

4. The method for developing an intelligent customer service platform based on a large model as claimed in claim 3, characterized in that: In S32, the knowledge base includes a plurality of different problems N and their corresponding solutions N.

5. The method for developing an intelligent customer service platform based on a large model as claimed in claim 1, characterized in that: In S33, the large model uses keyword extraction technology and multi-round dialogue technology to continuously clarify key information through word slots in the form of questions and answers, further including: S331: For the question input by the user and classified as a complex intent, the large model performs question segmentation and enters the segmented question scenario corresponding to the segmented question. Different segmented question scenarios have different key information that needs to be clarified. S332: The large model gradually guides users to clarify key information through preset question statements for different problem scenarios, conducts multiple rounds of dialogue, and asks users targeted questions to guide them to provide detailed information to fill in the word slots; S333: The large model uses named entity recognition technology to find the sub-nodes driven by the current segmentation problem and output the corresponding solutions.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 5.

Citation Information

Patent Citations

  • Intelligent customer service method and device, server and storage medium

    CN117743557A

  • Intelligent problem optimization dialogue method and system based on intention recognition of structural information

    CN117933271A

  • Slot information supplementing method, device and equipment based on multi-agent interaction

    CN118378658A

  • Systems and methods for ensembling soft prompts in few-shot fine-tuning of language models

    US20240070394A1