Method and system for IT service industry intention recognition based on large model

Through the combination of the JointBERT model training and the intention classification system, the problem of automatically identifying the needs of IT service personnel in IT service management is solved, efficient IT service intention recognition is achieved, and the efficiency and user experience of IT service management are improved.

CN119048288APending Publication Date: 2024-11-29SHANDONG LANGCHAO YUNTOU INFORMATION TECH CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411049424.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The existing technology is difficult to automatically capture the needs of IT service personnel, resulting in inefficient IT service management and poor user experience. Traditional rules-based automation systems have limited recognition capabilities in complex scenarios.

Method used

Using a large model-based intent recognition method, the JointBERT model is trained and fine-tuned, combined with the intention classification system and multiple rounds of interaction strategies, users query and dialogue data in IT service scenarios can be collected and processed to achieve unique intentions for the IT service field. Accurate identification.

Benefits of technology

It significantly improves the efficiency and user experience of IT service management, and greatly improves the accuracy and coverage of IT service intention identification, which can automatically capture the needs of IT service personnel, reduce work burdens, and provide support for the digital transformation of IT services.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048288A_ABST
    Figure CN119048288A_ABST
Patent Text Reader

Abstract

The invention discloses a large-model-based IT service industry-oriented intention recognition method and system, belongs to the technical field of large models and data processing, and aims to solve the technical problems of how to automatically capture the requirements of IT service personnel, reduce the workload of the IT service personnel and improve the IT service management efficiency and user experience. According to the technical scheme, the method comprises the steps of collecting annotation data, wherein various data closely related to IT service are collected in a multi-channel and all-around mode, and a foundation is laid for constructing a high-performance IT service supporting platform; designing prompt expansion for the annotated data; further expanding the annotated data by utilizing a prompt technology to enhance the diversity and robustness of a training set; arranging the data set: converting the format of the collected excel data into a json format; dividing a data set: dividing a test set and a training set according to the time distribution of the samples; consciousness identification: realizing intention identification for the IT service industry; and realizing an API interface, wherein the awareness identification packaging is in an API interface form.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of big models and data processing technology, and in particular to a method and system for intent recognition in the IT service industry based on big models. Background Art

[0002] With the rapid development of information technology, the digital transformation of various industries is being promoted in an all-round way. In the field of IT service management, how to improve work efficiency and user experience through intelligent means has become one of the key issues in the industry.

[0003] Traditional IT service management models usually rely on manual operations and rule-based automation systems. IT service personnel need to manually diagnose equipment failures, adjust system configurations, optimize workflows, etc., which is not only inefficient but also prone to human errors. Although rule-based automation systems can handle some common IT service tasks, their recognition capabilities and flexibility are often limited when faced with complex IT service scenarios. For example, the request data is: query the total amount of data for today's work orders, which can be classified as daily IT service intention recognition. The request data is: query the solution to a certain problem, which can be classified as fault IT service intention recognition. Subsequently, different intention recognitions can be called based on daily IT service intention recognition and fault IT service recognition. Daily IT services call the interface to generate SQL statements to display to IT service personnel, and fault IT services call the knowledge question and answer interface to display to IT service personnel.

[0004] In recent years, with the rapid advancement of artificial intelligence technology, intent recognition methods based on large models have attracted widespread attention. Such methods can automatically capture users' implicit needs and complex intentions through deep learning, providing strong support for the digital transformation of various industries. However, most of the existing large-model intent recognition technologies are oriented to general scenarios and are difficult to meet the unique needs of the IT service field.

[0005] Therefore, how to automatically capture the needs of IT service personnel, reduce their workload, and improve the efficiency of IT service management and user experience is a technical problem that needs to be solved urgently. Summary of the invention

[0006] The technical task of the present invention is to provide a method and system for intent recognition in the IT service industry based on a large model to solve the problem of how to automatically capture the needs of IT service personnel, reduce the workload of IT service personnel, and improve the efficiency of IT service management and user experience.

[0007] The technical task of the present invention is achieved in the following way: a method for intent recognition in the IT service industry based on a large model, which is to collect user query and dialogue data in IT service scenarios, train and fine-tune the JointBERT model based on the user query and dialogue data in the IT service scenarios, so that the JointBERT model has the ability to accurately identify the unique intents in the IT service field; at the same time, an intent classification system for IT service scenarios is designed, covering various common IT service requirements and operation intentions such as equipment management, fault diagnosis, performance optimization, process control, etc., to achieve more fine-grained intent recognition, and introduce a multi-round interaction strategy, through a cyclical dialogue with IT service personnel, continuously refine and correct the preliminary intent recognition results, and improve the reliability of recognition; for example, after preliminarily identifying the intention of equipment failure, you can actively ask for information such as the fault phenomenon and equipment model to obtain a more accurate diagnosis result; the details are as follows:

[0008] Collect and annotate data: Collect all kinds of data closely related to IT services through multiple channels and in an all-round manner, laying the foundation for building a high-performance IT service support platform;

[0009] Design prompt expansion for labeled data: Use prompt technology to further expand labeled data to enhance the diversity and robustness of the training set;

[0010] Organize the data set: convert the collected Excel data format into JSON format;

[0011] Divide the data set: divide the test set and training set according to the time distribution of the samples;

[0012] Awareness recognition: realizing intention recognition for the IT service industry;

[0013] Implement API interface: Consciousness recognition is encapsulated in the form of API interface.

[0014] As a preferred method, the data is obtained in the following manner:

[0015] Collect multi-round request data of historical IT services and provide rich original samples; the multi-round request data of historical IT services includes user ID, request time, request text and final processing result information;

[0016] Online customer service chat records: Enterprises provide online customer service systems to provide technical support to users through chat. Chat records contain a large number of IT service dialogue cases.

[0017] Customer service phone recordings: Companies save recordings of conversations between customers and customer service personnel and convert them into text data through voice transcription; the recordings include conversations related to IT services.

[0018] As a preferred method, the prompt expansion is designed for the labeled data as follows:

[0019] Clarify the goal of prompts: generate similar but original meanings, use natural and fluent language, and avoid using overly stiff or artificial sentences; or refer to the expressions of real users to make the generated samples closer to reality;

[0020] Describe the task scenario: The prompt clearly describes the specific scenario and background of the task, so that the language model can better understand the context information; for IT service scenarios, this includes key information such as user identity, problem description, and expected results;

[0021] Control the length of the prompt: The length of the prompt should not be too long. Generally, it is best to control it within 50-100 characters. Too long prompts may reduce the generation effect of the language model.

[0022] Determine output requirements: Determine the expected output format based on the target task; the output format includes the length and style of the generated text, classification labels, and answer content.

[0023] As a preferred method, the prompt text is constructed as follows:

[0024] Based on the original questions provided, a batch of new questions are constructed, all of which are stored in a json object; the new questions are fine-tuned in the wording of the expression of the questions while maintaining the original questions, so that the new questions are similar in meaning to the original questions but different in form;

[0025] Original questions: {} Please provide at least 10 new questions, and make sure the new questions add variety and novelty while maintaining the original intent.

[0026] As a preference, the JSON format of the data set is as follows:

[0027]

[0028] Among them, text is the request data of the IT service personnel; intent is the intention marked by the request data; slots is the content extracted from the slot.

[0029] Preferably, the consciousness recognition is as follows:

[0030] During the training process, the JointBERT model is trained using a dataset with intent and slot annotations, so that the JointBERT model can learn to simultaneously identify intent and slots from the input text; and optimize the model parameters by minimizing the loss function, which includes the cross entropy loss for intent classification and the sequence annotation loss for slot filling;

[0031] Use the last hidden state corresponding to the [CLS] token to predict the intent of the entire sentence, and use the last hidden states of the sentence tokens for sequence annotation to find the tokens containing slot values. At the same time, define your own intent and slot labels, provide your own data, train your own model, and load the trained model in the JointIntentSlotDetector class to directly predict intent and slot values.

[0032] A system for IT service industry intention recognition based on a large model, the system is used to implement the above-mentioned method for IT service industry intention recognition based on a large model; the system comprises:

[0033] The data collection and annotation module is used to collect various types of data closely related to IT services through multiple channels and in an all-round manner, laying the foundation for building a high-performance IT service support platform;

[0034] The expansion module is used to further expand the labeled data using prompt technology to enhance the diversity and robustness of the training set;

[0035] The data set organization module is used to convert the collected Excel data format into JSON format;

[0036] The data set partitioning module is used to partition the test set and training set according to the time distribution of samples;

[0037] Awareness recognition module, used to realize intention recognition for the IT service industry;

[0038] The interface module is used to encapsulate consciousness recognition into an API interface form.

[0039] Preferably, the expansion module comprises:

[0040] Clarify the submodules to clarify the goal of prompt, that is, to generate similar but original meanings, use natural and fluent language expressions, and avoid using overly stiff or artificial sentences; or refer to the expressions of real users to make the generated samples closer to reality;

[0041] The description submodule is used to describe the task scenario. That is, the prompt clearly describes the specific scenario and background of the task, so that the language model can better understand the context information. For IT service scenarios, it includes key information such as user identity, problem description, and expected results.

[0042] The control submodule is used to control the prompt length. That is, the prompt length should not be too long. Generally, it is better to control it within 50-100 characters. Too long prompts may reduce the generation effect of the language model.

[0043] A determination submodule is used to determine the expected output form according to the target task; wherein the output form includes the length and style of the generated text, the classification label and the answer content;

[0044] The JointBERT model in the consciousness recognition module is trained using a dataset with intent and slot annotations, so that the JointBERT model learns to simultaneously recognize intent and slots from input text; and optimizes the parameters of the model by minimizing a loss function, the loss function including a cross entropy loss for intent classification and a sequence annotation loss for slot filling;

[0045] At the same time, use the last hidden state corresponding to the [CLS] token to predict the intent of the whole sentence, and use the last hidden states of the sentence tokens for sequence annotation to find the tokens containing slot values; define your own intent and slot labels, provide your own data, train your own model, and load the trained model in the JointIntentSlotDetector class to directly predict intent and slot values.

[0046] An electronic device comprising: a memory and at least one processor;

[0047] Wherein, the memory stores computer-executable instructions;

[0048] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above-mentioned method for intent recognition for the IT service industry based on a big model.

[0049] A computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method for intent recognition based on a large model for the IT service industry as described above is implemented.

[0050] The method and system for IT service industry intention recognition based on a large model of the present invention have the following advantages:

[0051] (I) The present invention improves the efficiency of IT service work and user experience by accurately identifying the implicit needs and complex intentions of IT service personnel, and adopts intention recognition as the core technology to identify conversations in IT service scenarios; among them, intention recognition is a key link in the field of natural language processing, and its main goal is to distinguish the intention or goal of the user from the language expression; in short, intention recognition is to understand the meaning of the user's language so as to more effectively respond to the user's questions or provide the required services; in the scope of natural language processing, intention recognition is to achieve the recognition of intention by classifying the sentences input by the user into the established intention category;

[0052] (ii) The present invention has achieved significant improvements in the accuracy, coverage and response speed of intent recognition in IT service scenarios, which is far superior to traditional rule-based or machine learning methods. At the same time, the present invention has also achieved good results in actual IT service applications, greatly improving the work efficiency and satisfaction of IT service personnel;

[0053] (III) The present invention has important application value in improving IT service management level and user experience, and through automated and intelligent intent recognition, it can better understand the needs of IT service personnel and provide strong support for the digital transformation of IT service work;

[0054] (IV) The present invention can automatically capture the needs of IT service personnel, reduce the workload of IT service personnel, and improve the efficiency of IT service management and user experience; at the same time, the present invention can also provide strong support for the digital transformation of IT service work and lay the foundation for intelligent IT service management in the industry;

[0055] (V) The present invention uses intent recognition as the core technology to better understand the demands of IT service personnel; at the same time, the present invention also introduces innovative technologies such as intent classification system and multi-round interaction strategy for IT service scenarios to further improve the accuracy and reliability of recognition;

[0056] (VI) The present invention significantly improves the accuracy and coverage of IT service intent recognition: by adopting the JointBERT model as the core technology, it can simultaneously perform two tasks, intent recognition and slot extraction, so as to better understand the demands of IT service personnel; at the same time, it introduces an intent classification system for IT service scenarios and a modeling method based on the attention mechanism, which further improves the accuracy of recognition; large-scale experimental results show that the intent recognition accuracy of the present invention in IT service scenarios can reach more than 90%, which is far superior to traditional rule-based or machine learning methods;

[0057] (VII) The present invention greatly improves the efficiency of IT service work and user experience: by automatically capturing the needs of IT service personnel, it can greatly reduce the workload of IT service personnel and improve the efficiency of IT service management; at the same time, the introduction of a multi-round interaction strategy can continuously refine and correct the preliminary intent recognition results, further improving the reliability of recognition; in practical applications, it has significantly improved the work efficiency and satisfaction of IT service personnel;

[0058] (VIII) The present invention provides strong support for the digital transformation of IT service work: Compared with the traditional IT service management model, the intelligent intent recognition method based on a large model proposed in the present invention can better understand the needs of IT service personnel and lay the foundation for intelligent IT service management in the industry; and through automated and intelligent intent recognition, it can better support the digital transformation of IT service work and improve the management level and user experience of the entire industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The present invention is further described below in conjunction with the accompanying drawings.

[0060] Attached Figure 1 This is a flowchart of the method for intent recognition in the IT service industry based on a big model. DETAILED DESCRIPTION

[0061] The method and system for IT service industry intent recognition based on a large model of the present invention are described in detail below with reference to the accompanying drawings and specific embodiments of the specification.

[0062] Embodiment 1:

[0063] As attached Figure 1 As shown, this embodiment provides a method for intent recognition in the IT service industry based on a large model. The method is to collect user query and conversation data in IT service scenarios, train and fine-tune the JointBERT model based on the user query and conversation data in IT service scenarios, so that the JointBERT model has the ability to accurately identify the unique intents in the IT service field; at the same time, an intent classification system for IT service scenarios is designed, covering various common IT service requirements and operation intentions such as equipment management, fault diagnosis, performance optimization, and process control, to achieve more fine-grained intent recognition, and introduce a multi-round interaction strategy. Through a cyclical dialogue with IT service personnel, the preliminary intent recognition results are continuously refined and corrected to improve the reliability of recognition; for example, after the preliminary identification of the intention of equipment failure, the fault phenomenon, equipment model and other information can be actively asked to obtain more accurate diagnosis results; the details are as follows:

[0064] S1. Collect annotated data: Collect all kinds of data closely related to IT services through multiple channels and in an all-round manner, laying the foundation for building a high-performance IT service support platform;

[0065] S2. Design prompt expansion for labeled data: Use prompt technology to further expand the labeled data to enhance the diversity and robustness of the training set;

[0066] S3, organize the data set: convert the collected excel data format into json format;

[0067] S4. Divide the data set: divide the test set and training set according to the time distribution of the samples;

[0068] S5. Awareness recognition: realizing intention recognition for the IT service industry;

[0069] S6. Implement API interface: Consciousness recognition is encapsulated in the form of API interface.

[0070] The specific method of acquiring data in step S1 of this embodiment is as follows:

[0071] ① Collect multi-round request data of historical IT services and provide rich original samples; the multi-round request data of historical IT services includes user ID, request time, request text and final processing result information;

[0072] ② Online customer service chat records: The company provides an online customer service system to provide technical support to users through chat. The chat records contain a large number of IT service dialogue cases;

[0073] ③ Customer service phone recordings: Companies save recordings of conversations between customers and customer service personnel and convert them into text data through voice transcription; the recordings include conversations related to IT services.

[0074] The prompt expansion designed for the labeled data in step S2 of this embodiment is specifically as follows:

[0075] S201. Clarify the goal of prompt: generate similar but original meanings, use natural and fluent language, and avoid using overly stiff or artificial sentences; or refer to the expression of real users to make the generated samples closer to reality;

[0076] S202, describe the task scenario: the prompt clearly describes the specific scenario and background of the task, so that the language model can better understand the context information; for IT service scenarios, this includes key information such as user identity, problem description, and expected results;

[0077] S203. Control prompt length: The prompt length should not be too long. Generally, it is best to control it within 50-100 characters. Too long prompts may reduce the generation effect of the language model.

[0078] S204, determine output requirements: determine the desired output format according to the target task; wherein the output format includes the length and style of the generated text, classification labels and answer content.

[0079] The method of constructing the prompt text in step S2 of this embodiment is specifically as follows:

[0080] Based on the original questions provided, a batch of new questions are constructed, all of which are stored in a json object; the new questions are fine-tuned in the wording of the expression of the questions while maintaining the original questions, so that the new questions are similar in meaning to the original questions but different in form;

[0081] Original questions: {} Please provide at least 10 new questions, and make sure the new questions add variety and novelty while maintaining the original intent.

[0082] The json format of the data set in step S3 of this embodiment is as follows:

[0083]

[0084] Among them, text is the request data of the IT service personnel; intent is the intention marked by the request data; slots is the content extracted from the slot.

[0085] The data set division in step S4 of this embodiment is specifically as follows:

[0086] 80% of the samples in the above collected data set are used as training set, and the remaining 20% ​​are used as test set.

[0087] At the same time, the time distribution of samples is considered. Generally speaking, earlier patent samples are used as training sets, and newer patent samples are used as test sets. This can not only better evaluate the generalization performance of the model on new patents, but also reflect the dynamic changes in technological development.

[0088] The consciousness recognition in step S5 of this embodiment is specifically as follows:

[0089] S501. During the training process, the JointBERT model is trained using a data set with intent and slot annotations, so that the JointBERT model learns to simultaneously recognize intent and slots from the input text; and the parameters of the model are optimized by minimizing a loss function, where the loss function includes a cross entropy loss for intent classification and a sequence annotation loss for slot filling;

[0090] S502. Use the last hidden state corresponding to the [CLS] token to predict the intent of the whole sentence, and use the last hidden states of the sentence tokens for sequence annotation to find the tokens containing slot values. At the same time, define your own intent and slot labels, and provide your own data. Train your own model and load the trained model in the JointIntentSlotDetector class to directly predict intent and slot values.

[0091] The line of sight API interface in step S6 of this embodiment is as follows:

[0092] Intent recognition is encapsulated as an interface for other systems to call, and the IP address and interface for external access are set. Other systems can directly call the interface, enter the request question, and directly obtain the category of intent recognition.

[0093] Embodiment 2:

[0094] This embodiment provides a system for IT service industry intention recognition based on a big model, which is used to implement the method for IT service industry intention recognition based on a big model in Embodiment 1; the system includes:

[0095] The data collection and annotation module is used to collect various types of data closely related to IT services through multiple channels and in an all-round manner, laying the foundation for building a high-performance IT service support platform;

[0096] The expansion module is used to further expand the labeled data using prompt technology to enhance the diversity and robustness of the training set;

[0097] The data set organization module is used to convert the collected Excel data format into JSON format;

[0098] The data set partitioning module is used to partition the test set and training set according to the time distribution of samples;

[0099] Awareness recognition module, used to realize intention recognition for the IT service industry;

[0100] The interface module is used to encapsulate consciousness recognition into an API interface form.

[0101] The expansion module in this embodiment includes:

[0102] Clarify the submodules to clarify the goal of prompt, that is, to generate similar but original meanings, use natural and fluent language expressions, and avoid using overly stiff or artificial sentences; or refer to the expressions of real users to make the generated samples closer to reality;

[0103] The description submodule is used to describe the task scenario. That is, the prompt clearly describes the specific scenario and background of the task, so that the language model can better understand the context information. For IT service scenarios, it includes key information such as user identity, problem description, and expected results.

[0104] The control submodule is used to control the prompt length. That is, the prompt length should not be too long. Generally, it is better to control it within 50-100 characters. Too long prompts may reduce the generation effect of the language model.

[0105] The determination submodule is used to determine the expected output form according to the target task; wherein the output form includes the length and style of the generated text, the classification label and the answer content.

[0106] The JointBERT model in the consciousness recognition module in this embodiment is trained using a dataset with intent and slot annotations, so that the JointBERT model learns to simultaneously recognize intent and slots from the input text; and optimizes the parameters of the model by minimizing the loss function, which includes the cross entropy loss of intent classification and the sequence annotation loss of slot filling;

[0107] At the same time, use the last hidden state corresponding to the [CLS] token to predict the intent of the whole sentence, and use the last hidden states of the sentence tokens for sequence annotation to find the tokens containing slot values; define your own intent and slot labels, provide your own data, train your own model, and load the trained model in the JointIntentSlotDetector class to directly predict intent and slot values.

[0108] Embodiment 3:

[0109] This embodiment also provides an electronic device, including: a memory and at least one processor;

[0110] Wherein, the memory stores computer-executable instructions;

[0111] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method for intent recognition based on a large model for the IT service industry as described in any one of the present inventions.

[0112] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor may be a microprocessor or any conventional processor, etc.

[0113] The memory can be used to store computer programs and / or modules. The processor realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory can also include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash memory card, at least one disk storage period, a flash memory device, or other volatile solid-state storage devices.

[0114] Embodiment 4:

[0115] This embodiment also provides a computer-readable storage medium, which stores a plurality of instructions, which are loaded by a processor to enable the processor to execute the method for identifying intent based on a large model for the IT service industry in any embodiment of the present invention. Specifically, a system or device equipped with a storage medium can be provided, on which a software program code that implements the functions of any of the above embodiments is stored, and a computer (or CPU or MPU) of the system or device reads and executes the program code stored in the storage medium.

[0116] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

[0117] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.

[0118] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.

[0119] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or written to a memory provided in an expansion unit connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or the expansion unit is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intent recognition in IT service industry based on a large model, characterized in that: The method is to collect user query and conversation data in IT service scenarios, train and fine-tune the JointBERT model based on the user query and conversation data in IT service scenarios, so that the JointBERT model has the ability to accurately identify the unique intents in the IT service field; at the same time, an intent classification system for IT service scenarios is designed to achieve fine-grained intent recognition, and a multi-round interaction strategy is introduced to continuously refine and correct the preliminary intent recognition results through cyclical dialogues with IT service personnel to improve the reliability of recognition; the details are as follows: Collect and annotate data: Collect all kinds of data closely related to IT services through multiple channels and in an all-round manner, laying the foundation for building a high-performance IT service support platform; Design prompt expansion for labeled data: Use prompt technology to further expand labeled data to enhance the diversity and robustness of the training set; Organize the data set: convert the collected Excel data format into JSON format; Divide the data set: divide the test set and training set according to the time distribution of the samples; Awareness recognition: realizing intention recognition for the IT service industry; Implement API interface: Consciousness recognition is encapsulated in the form of API interface.

2. The method for IT service industry intention recognition based on a large model according to claim 1 is characterized in that: The data is obtained in the following ways: Collect multi-round request data of historical IT services and provide rich original samples; the multi-round request data of historical IT services includes user ID, request time, request text and final processing result information; Online customer service chat records: The company provides an online customer service system to provide technical support to users through chat. The chat records contain IT service dialogue cases. Customer service phone recordings: Companies save recordings of conversations between customers and customer service personnel and convert them into text data through voice transcription; the recordings include conversations related to IT services.

3. The method for IT service industry intention recognition based on a large model according to claim 1 is characterized in that: The prompt expansion designed for labeled data is as follows: Clarify the goal of prompts: generate similar but original meanings, use natural and fluent language, and avoid using overly stiff or artificial sentences; or refer to the expressions of real users to make the generated samples close to reality; Describe the task scenario: The prompt clearly describes the specific scenario and background of the task, so that the language model can better understand the context information; for IT service scenarios, this includes key information such as user identity, problem description, and expected results; Control the prompt length: The prompt length should be controlled within 50-100 characters; Determine output requirements: Determine the expected output format based on the target task; the output format includes the length and style of the generated text, classification labels, and answer content.

4. The method for IT service industry intention recognition based on a large model according to claim 1 is characterized in that: The prompt text is constructed as follows: Based on the original questions provided, a batch of new questions are constructed, all of which are stored in a json object; the new questions are fine-tuned in the wording of the expression of the questions while maintaining the original questions, so that the new questions are similar in meaning to the original questions but different in form; Original questions: {} Please provide at least 10 new questions, and make sure the new questions add variety and novelty while maintaining the original intent.

5. The method for IT service industry intention recognition based on a large model according to claim 1 is characterized in that: The json format of the dataset is as follows: { "text":"Query the total amount of data for today's work orders", "intent":"Daily IT Services", Among them, text is the request data of the IT service personnel; intent is the intention marked by the request data; slots is the content extracted from the slot.

6. The method for IT service industry intention recognition based on a large model according to any one of claims 1 to 5, characterized in that: Awareness recognition is as follows: During the training process, the JointBERT model is trained using a dataset with intent and slot annotations, so that the JointBERT model can learn to simultaneously recognize intent and slots from the input text; The model parameters are optimized by minimizing the loss function, which includes the cross entropy loss for intent classification and the sequence labeling loss for slot filling. Use the last hidden state corresponding to the [CLS] token to predict the intent of the entire sentence, and use the last hidden states of the sentence tokens for sequence annotation to find the tokens containing slot values. At the same time, define your own intent and slot labels, provide your own data, train your own model, and load the trained model in the JointIntentSlotDetector class to directly predict intent and slot values.

7. A system for IT service industry intention recognition based on a large model, characterized in that: The system is used to implement the method for IT service industry intention recognition based on a large model as described in any one of claims 1 to 6; the system includes: The data collection and annotation module is used to collect various types of data closely related to IT services through multiple channels and in an all-round manner, laying the foundation for building a high-performance IT service support platform; The expansion module is used to further expand the labeled data using prompt technology to enhance the diversity and robustness of the training set; The data set organization module is used to convert the collected Excel data format into JSON format; The data set partitioning module is used to partition the test set and training set according to the time distribution of samples; Awareness recognition module, used to realize intention recognition for the IT service industry; The interface module is used to encapsulate consciousness recognition into an API interface form.

8. The system for IT service industry intention recognition based on a large model according to claim 7 is characterized in that: The expansion module includes: Clarify the submodules to clarify the goal of prompt, that is, to generate similar but original meanings, use natural and fluent language expressions, and avoid using overly stiff or artificial sentences; or refer to the expressions of real users to make the generated samples close to reality; The description submodule is used to describe the task scenario. That is, the prompt clearly describes the specific scenario and background of the task, so that the language model can better understand the context information. For IT service scenarios, it includes key information such as user identity, problem description, and expected results. The control submodule is used to control the prompt length, that is, the prompt length is preferably controlled within 50-100 characters; A determination submodule is used to determine the expected output form according to the target task; wherein the output form includes the length and style of the generated text, the classification label and the answer content; The JointBERT model in the consciousness recognition module is trained using a dataset with intent and slot annotations, so that the JointBERT model learns to simultaneously recognize intent and slots from input text; and optimizes the parameters of the model by minimizing a loss function, the loss function including a cross entropy loss for intent classification and a sequence annotation loss for slot filling; At the same time, use the last hidden state corresponding to the [CLS] token to predict the intent of the whole sentence, and use the last hidden states of the sentence tokens for sequence annotation to find the tokens containing slot values; define your own intent and slot labels, provide your own data, train your own model, and load the trained model in the JointIntentSlotDetector class to directly predict intent and slot values.

9. An electronic device, characterized in that: include: memory and at least one processor; Wherein, the memory stores computer-executable instructions; The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method for IT service industry intent recognition based on a large model as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the processor executes the computer-executable instructions, the method for intent recognition based on a large model for the IT service industry as described in any one of claims 1 to 6 is implemented.

Citation Information

Cited By

  • Multi-modal large model-based intent understanding interaction method

    CN119807811A

  • An intention understanding interaction method based on multimodal large model

    CN119807811B