Data processing method and related equipment

Abstract

The invention belongs to the field of artificial intelligence and big data, is applied to the field of smart cities, and relates to a data processing method which comprises the following steps: generating a prompt text template and a sample summary case through a pre-trained teacher model; performing knowledge distillation training processing on a student model based on the prompt text template and the sample summary case to obtain a trained target summary model; and summarizing a to-be-summarized dialogue text through the trained target summarizing model, and outputting a summarizing result of the to-be-summarized dialogue text. The invention further provides a data processing device, computer equipment and a storage medium. The dialogue processing method solves the problems that an existing traditional dialogue processing method is inaccurate in understanding, biased in summarization, sensitive to abnormal input and the like, and customer service staff are often difficult to accurately and quickly understand and summarize customer demands.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence and big data technology, and in particular to a data processing method and related equipment. Background Art

[0002] With the development of the Internet, customer service has become an important part of business operations. However, due to the diversity and complexity of customer needs, traditional dialogue processing systems face problems such as inaccurate understanding, biased summarization, and sensitivity to abnormal input. Customer service personnel often find it difficult to accurately and quickly understand and summarize customer needs. In addition, a large amount of customer feedback information also requires customer service personnel to spend a lot of time and energy to process and analyze. Summary of the invention

[0003] The purpose of the embodiments of the present application is to propose a data processing method and related equipment to solve the problems faced by existing traditional dialogue processing methods, such as inaccurate understanding, biased summarization, and sensitivity to abnormal input. Customer service personnel often find it difficult to accurately and quickly understand and summarize customer needs. Its main purpose is to automatically analyze and summarize customer demands by integrating knowledge distillation and adversarial training, thereby helping customer service personnel improve work efficiency and service quality.

[0004] In order to solve the above technical problems, the present application provides a data processing method, which adopts the following technical solution:

[0005] Generate prompt text templates and sample summary cases through the pre-trained teacher model;

[0006] Based on the prompt text template and the sample summary case, the student model is trained with knowledge distillation to obtain a trained target summary model;

[0007] The dialogue text to be summarized is summarized by using the trained target summary model, and a summary result of the dialogue text to be summarized is output.

[0008] Furthermore, before generating the prompt text template and the sample summary case by the pre-trained teacher model, the method further includes:

[0009] Obtaining sample training corpus, wherein the training corpus is in the form of question-answer pairs;

[0010] The teacher model is trained using the sample training corpus to obtain a pre-trained teacher model.

[0011] Furthermore, the obtaining of sample training corpus includes:

[0012] Get the historical conversation text;

[0013] Performing a first data preprocessing on the historical conversation text to obtain a preprocessed historical conversation text;

[0014] The preprocessed historical conversation text is subjected to a second data processing to obtain a sample training corpus in the form of question-answer pairs.

[0015] Furthermore, the generation of prompt text templates and sample summary cases by the pre-trained teacher model includes:

[0016] Obtaining a reference case in the training corpus of the pre-trained teacher model, inputting the training corpus of the reference case into the prompt text, and generating a prompt text template;

[0017] The prompt text of the reference case is input into the pre-trained teacher model for summary processing, and the sample summary case is output.

[0018] Furthermore, the case input in the training corpus of the pre-trained teacher model is used to generate a prompt text template, including:

[0019] Adopting an adversarial attack algorithm to perform adversarial perturbation on the reference cases in the training corpus of the pre-trained teacher model to obtain adversarial training corpus;

[0020] The adversarial training corpus is input into the prompt text to generate a prompt text template.

[0021] Furthermore, the optimization goal of the adversarial attack algorithm is:

[0022]

[0023] Where x is the input text, x′ is the adversarial input text, and f θ represents the model, θ is the parameter of the model, ∈ Limit the disturbance amplitude, p is often taken as ∞, δ is the disturbance, and y is the predicted result.

[0024] Furthermore, the process of summarizing the dialogue text to be summarized by the trained target summarization model and outputting the summary result of the dialogue text to be summarized includes:

[0025] Generate target prompt text template through teacher model;

[0026] Fill the dialogue text to be summarized into the target prompt text template to obtain the target prompt text;

[0027] The target prompt text is input into the trained target summary model for summary processing, and the summary result of the dialogue text to be summarized is output.

[0028] In order to solve the above technical problems, the present application also provides a data processing device, which adopts the following technical solution:

[0029] A generation module for generating prompt text templates and sample summary cases through a pre-trained teacher model;

[0030] A knowledge distillation training module is used to perform knowledge distillation training on the student model based on the prompt text template and the sample summary case to obtain a trained target summary model;

[0031] The summary processing module is used to perform summary processing on the dialogue text to be summarized through the trained target summary model, and output the summary result of the dialogue text to be summarized.

[0032] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0033] The computer device includes a memory and a processor, the memory stores computer-readable instructions, and the processor implements the steps of any one of the data processing methods described in the embodiments of the present application when executing the computer-readable instructions.

[0034] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0035] The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the data processing method described in any one of the embodiments of the present invention are implemented.

[0036] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects: a prompt text template and a sample summary case are generated through a pre-trained teacher model, and the student model is trained and processed with knowledge distillation through the prompt text template and the sample summary case to obtain a trained target summary model, and the dialogue text to be summarized is summarized through the trained target summary model, and the summary result of the dialogue text to be summarized is output, and by integrating knowledge distillation, automatic analysis and summary of customer demands, it helps customer service personnel improve work efficiency and service quality, and solves the problems faced by existing traditional dialogue processing methods such as inaccurate understanding, biased summary and sensitivity to abnormal input, and the problem that customer service personnel often find it difficult to accurately and quickly understand and summarize customer needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0039] Figure 2 is a flow chart of an embodiment of a data processing method according to the present application;

[0040] Figure 3 yes Figure 2 A flowchart of a specific implementation method before step S201;

[0041] Figure 4 yes Figure 3 A flowchart of a specific implementation of step S301;

[0042] Figure 5 yes Figure 2 A flowchart of a specific implementation method of step S201;

[0043] Figure 6 yes Figure 5 A flowchart of a specific implementation method of step S2011;

[0044] Figure 7 yes Figure 2 A flowchart of a specific implementation method of step S203;

[0045] Figure 8 is a structural framework diagram of the data processing method according to the present application;

[0046] Fig. 9 is a structural diagram of a specific implementation of a data processing device 900 according to the present application;

[0047] Fig.10 yes Fig. 9 A structural diagram of another specific implementation of the data processing device 900;

[0048] Fig.11 yes Fig.10 A schematic diagram of a structure of a specific implementation of the acquisition module 904;

[0049] Fig.12 yes Fig. 9 A schematic diagram of a structure of a specific implementation of the generation module 901;

[0050] Fig.13 yes Fig.12 A schematic diagram of a structure of a specific implementation of the first generation submodule 9011;

[0051] Fig.14 yes Fig. 9 A schematic diagram of a specific implementation of the summary processing module 903;

[0052] Fig.15 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0054] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0055] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0056] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0057] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0058] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV), laptop computers, desktop computers, etc.

[0059] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0060] It should be noted that the data processing method provided in the embodiments of the present application is generally executed by a server / terminal device, and accordingly, the data processing device is generally arranged in the server / terminal device.

[0061] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0062] Continue to refer Figure 2 , shows a flow chart of an embodiment of a data processing method according to the present application. The data processing method comprises the following steps:

[0063] Step S201: Generate prompt text templates and sample summary cases through the pre-trained teacher model.

[0064] In this embodiment, the electronic device (eg, Figure 1 The server / terminal device shown in the figure can receive the data processing request of the terminal device through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0065] The above-mentioned terminal devices may be terminal devices equipped with remote interaction functions such as personal assistants and intelligent customer service.

[0066] The above data processing method can be applied to online file data processing, online file management and other businesses.

[0067] Specifically, the above-mentioned electronic equipment can be used for different industry types such as medical data processing, property insurance data processing, financial data processing, store data processing, etc.

[0068] The teacher model is a data processing tool that can capture complex patterns and information in deep data. The teacher model is used to guide the learning process of the student model. The teacher model can be understood as a large and complex model.

[0069] The pre-trained teacher model is learned based on a large amount of training corpus. The large amount of training corpus can be a training corpus in the form of question-answer pairs obtained by collating and collecting a large amount of historical dialogue texts and performing data processing on the historical dialogue texts. The data processing can be a data processing method such as natural language processing and manual annotation. The question-answer pair format can be understood as an interactive text format between questions and answers.

[0070] The above prompt text template is used to guide the student model to generate prompt text. The prompt text can be a short phrase or sentence that describes the customer's intention in the dialogue text to be summarized and explains the reason.

[0071] The above sample summary cases are a data set consisting of a set of input texts and corresponding output summaries. The sample summary cases are used to guide the student model to learn how to summarize the input prompt text.

[0072] It should be noted that the sample summary cases can provide existing input and output data as references, thereby guiding the student model to learn how to generate summaries.

[0073] Step S202: Perform knowledge distillation training on the student model based on the prompt text template and the sample summary case to obtain a trained target summary model.

[0074] In this embodiment, after obtaining the prompt text template and the sample summary case, the electronic device can use the prompt text template and the sample summary case to perform knowledge distillation training on the student model, thereby obtaining a trained target summary model.

[0075] The student model is a lightweight data processing tool. The student model can be understood as a lightweight student model that aims to learn and imitate the behavior or predict results from the teacher model. The student model is to reduce computational complexity and resource consumption while maintaining high performance, making it more suitable for running in resource-constrained environments.

[0076] The above-mentioned knowledge distillation is a compression and knowledge transfer technology that achieves lightweight and efficient models by transferring the knowledge of the teacher model to the student model. Knowledge distillation uses the soft labels output by the teacher model as supervision signals to guide the training of the student model, so that the student model can better simulate the behavior of the teacher model.

[0077] The above-trained target summary model can capture the key knowledge information in the input data, extract the key knowledge and summarize the input data.

[0078] It should be noted that the student model is trained by the prompt text templates and sample summary cases generated by the pre-trained teacher model, and the prompt text templates and sample summary cases output by the pre-trained teacher model are provided to the student model as soft targets instead of directly providing the original labels. In this way, the student model can learn the knowledge of the teacher model without having to learn directly from the original data. This application extracts key knowledge from a large and complex model and transmits it to a lightweight student model, thereby achieving a significant reduction in computing resource requirements while maintaining high performance.

[0079] Step S203: Summarize the dialogue text to be summarized by using the trained target summary model, and output a summary result of the dialogue text to be summarized.

[0080] In this embodiment, after obtaining the dialogue text to be summarized, the electronic device summarizes the dialogue text to be summarized by using the trained target summarization model, and outputs the summary result of the dialogue text to be summarized.

[0081] The above-mentioned dialogue text to be summarized can be understood as the dialogue content that needs to be summarized or generalized, including one or more dialogue sentences or paragraphs.

[0082] The above-mentioned summarization process can be understood as the process of extracting important information from the dialogue text to be summarized after in-depth analysis, and refining this information into a concise and refined statement. For example, it can be to summarize and explain the customer's intention and the reasons supporting this intention in the dialogue text to be summarized; evaluate and infer the customer's needs, problems, concerns, and specific services that may be required based on the dialogue text.

[0083] Specifically, a target prompt text template is generated through a trained target summary model, the dialogue text to be summarized is input into the target prompt text template to obtain the target prompt text, the target prompt text is summarized through the trained target summary model, and a summary result of the dialogue text to be summarized is output.

[0084] The target prompt text template can describe the main content or core idea of ​​the text to be summarized. The target prompt text template can effectively guide the trained target summary model to accurately summarize the input text.

[0085] In this embodiment, the application uses large model knowledge distillation technology to extract key knowledge from large and complex models and transfer it to lightweight student models, thereby achieving a significant reduction in computing resource requirements while maintaining high performance. Knowledge distillation technology can compress the rich knowledge and powerful capabilities of large pre-trained models into smaller and less computationally expensive student models, maintaining high performance while reducing deployment and operating costs. This application has been specially optimized for industry-specific language and business logic, and can more accurately and automatically extract key information from text conversations, accurately understand customer intent and quickly provide accurate feedback, shorten processing time, reduce human errors, improve customer satisfaction and trust, and enhance the company's competitiveness.

[0086] The present application generates a prompt text template and a sample summary case through a pre-trained teacher model, performs knowledge distillation training on the student model through the prompt text template and the sample summary case, obtains a trained target summary model, and summarizes the dialogue text to be summarized through the trained target summary model, and outputs the summary result of the dialogue text to be summarized, which solves the problems faced by existing traditional dialogue processing methods, such as inaccurate understanding, biased summary, and sensitivity to abnormal input, and the problem that customer service personnel often find it difficult to accurately and quickly understand and summarize customer needs.

[0087] Continue to refer Figure 3 , shown in Figure 2 The data processing method further includes the following steps:

[0088] Step S301: Obtain sample training corpus.

[0089] In this embodiment, the training corpus is in the form of question-answer pairs, which can be understood as an interactive text form between questions and answers.

[0090] The training corpus is a text data set used to train the model, and can be a training corpus in the form of question-answer pairs obtained by collating and collecting a large amount of historical conversation texts and performing data processing on the historical conversation texts. The data processing can be a data processing method such as natural language processing and manual annotation.

[0091] Step S302: Train the teacher model using sample training corpus to obtain a pre-trained teacher model.

[0092] In this embodiment, after obtaining the sample training corpus, the electronic device can train the teacher model using the sample training corpus to obtain a pre-trained teacher model.

[0093] The teacher model is a data processing tool that can capture complex patterns and information in deep data. The teacher model is used to guide the learning process of the student model.

[0094] During the training process, the model will learn the patterns and regularities in a large amount of sample training data, so that it can perform tasks such as summarizing or classifying the training data.

[0095] The above pre-trained teacher model can perform tasks such as summarizing or classifying the input dialogue text.

[0096] This application obtains sample training corpus, trains the teacher model through the sample training corpus, and obtains a pre-trained teacher model. The pre-trained teacher model has strong expressive and generalization capabilities, and can adapt to different application scenarios and task requirements.

[0097] Continue to refer Figure 4 , showing that in Figure 3 Flow chart of a specific implementation of step S301 in FIG. Step S301 specifically includes the following steps:

[0098] Step S3011, obtain historical conversation text.

[0099] In this embodiment, the above historical conversation texts can be understood as a large amount of business-related conversation data, including but not limited to consultation, product comparison, analysis, etc. For example, in the case of insurance business, the historical conversation texts are a large amount of insurance business-related conversation data, including but not limited to consultation, reporting, claims, etc.; for example, in the case of medical business, the historical conversation texts are a large amount of medical business-related conversation data, including but not limited to consultation, diagnosis, treatment, etc.

[0100] Step S3012: Perform a first data preprocessing on the historical conversation text to obtain a preprocessed historical conversation text.

[0101] In this embodiment, the first data preprocessing can be understood as a data processing process of the original data. The data preprocessing can be data cleaning, data labeling, etc. The data cleaning can be understood as a process of removing all unnecessary characters, punctuation marks, numbers and special symbols to ensure the cleanliness and consistency of the data. The data labeling can be understood as classifying the data into specific categories or labels.

[0102] Step S3013: perform a second data processing on the pre-processed historical conversation text to obtain a sample training corpus in the form of question-answer pairs.

[0103] In this embodiment, the second data processing can be understood as the process of collecting, storing, retrieving, processing, transforming and transmitting data, which can be a data processing method such as natural language processing, OCR, image recognition, manual annotation, etc.

[0104] The above-mentioned question-answer pair format is an interactive text format between questions and answers. In the question-answer pair format, two questions answer each other, and both parties communicate with each other; there is usually a conjunction between the question and the answer, such as "whether" or "how", so as to better express the intention of the question and the meaning of the answer.

[0105] It should be noted that by collecting and organizing business-related conversation data, and then performing data processing on the business-related conversation data, it is organized into training corpus in the form of question-answer pairs.

[0106] The present application obtains historical conversation texts, performs a first data preprocessing on the historical conversation texts to obtain preprocessed historical conversation texts, performs a second data processing on the preprocessed historical conversation texts to obtain sample training corpus in the form of question-answer pairs. The sample training corpus in the form of question-answer pairs can be conveniently used in the training process of supervised learning.

[0107] Continue to refer Figure 5 , showing that in Figure 2 Flow chart of a specific implementation of step S201 in FIG. Step S201 specifically includes the following steps:

[0108] Step S2011: obtain reference cases in the training corpus of the pre-trained teacher model, input the training corpus of the reference cases into the prompt text, and generate a prompt text template.

[0109] In this embodiment, a reference case may be selected from all cases in the pre-trained teacher model. The reference case may be one of all cases in the pre-trained teacher model, and may be a verified and instructive classic case.

[0110] The above prompt text is an instruction to guide the model to generate content. It can be a short phrase or sentence that describes the customer's intention in the conversation text to be summarized and explains the reason. For example, if you want the model to summarize the content of a book, you need to give an instruction like "Summarize "XXX"".

[0111] The above prompt text template is used to guide the student model to generate prompt text.

[0112] Step S2012: input the prompt text of the reference case into the pre-trained teacher model for summary processing, and output a sample summary case.

[0113] In this embodiment, the above-mentioned summarizing process can be understood as a process of extracting important information from the prompt text of the reference case after in-depth analysis, and refining the information into a concise and refined statement.

[0114] The sample summary cases above are used to guide the student model to learn how to summarize the input prompt text.

[0115] The present application obtains reference cases from the training corpus of a pre-trained teacher model, inputs the training corpus of the reference cases into a prompt text, generates a prompt text template, and inputs the prompt text of the reference cases into the pre-trained teacher model for summary processing, and outputs sample summary cases, so that the model can better understand the content and structure of the input text, thereby improving the quality of the summary.

[0116] Continue to refer Figure 6 , showing that in Figure 5 Flow chart of a specific implementation of step S2011 in FIG. Step S2011 specifically includes the following steps:

[0117] Step S20111: Use an adversarial attack algorithm to perform adversarial perturbation processing on reference cases in the training corpus of the pre-trained teacher model to obtain adversarial training corpus.

[0118] In this embodiment, the adversarial attack algorithm can be understood as intentionally adding some subtle disturbances to the input data to induce the machine learning model to make wrong predictions or behaviors.

[0119] Adversarial perturbation of the reference cases in the training corpus of the pre-trained teacher model means applying an adversarial attack algorithm on the data of the reference cases to generate challenging input data.

[0120] The above-mentioned adversarial perturbation processing can be understood as the situation when an incorrect judgment is made in the face of input data, and how to deal with this attack legal person processing process.

[0121] The above adversarial training corpus can be understood as a training corpus containing original data and its corresponding adversarial perturbation data.

[0122] It should be noted that by actively injecting noise and constructing adversarial samples, the robustness of the model is significantly enhanced, ensuring that it can still accurately understand and effectively process insurance conversations when facing various interferences, misstatements or non-standard expressions in actual applications, thereby improving the practicality and reliability of the system. The adversarial attack algorithm is used to simulate various noises and abnormal inputs, forcing the model to learn to resist these interferences during the training process, thereby improving the robustness of the model in actual applications when facing non-standard expressions, erroneous inputs or malicious attacks, and ensuring the security and reliability of the dialogue processing system.

[0123] Furthermore, we first construct a generative adversarial network consisting of a generator and a discriminator. The generator is responsible for generating extraction and summary results from the input dialogue text, and the discriminator is used to determine whether the results generated by the generator are real extraction and summary or forged.

[0124] Then, adversarial training is performed. The adversarial training process includes alternately training the generator and the discriminator. The goal of the generator is to generate extraction and summary results that can deceive the discriminator, making it difficult for the discriminator to distinguish between real results and generated results. The goal of the discriminator is to accurately distinguish between real extraction and summary results and results generated by the generator.

[0125] Step S20112: input the adversarial training corpus into the prompt text to generate a prompt text template.

[0126] In this embodiment, the above prompt text can be understood as input text used to guide the model to generate results.

[0127] The above prompt text template can effectively guide the model to generate correct results and prevent the model from being attacked.

[0128] The present application uses an adversarial attack algorithm to perform adversarial perturbation processing on the reference cases in the training corpus of the pre-trained teacher model to obtain adversarial training corpus, and inputs the adversarial training corpus into the prompt text to generate a prompt text template. By using the adversarial attack algorithm to simulate various noises and abnormal inputs, the model is forced to learn to resist these interferences during the training process, thereby improving the robustness of the model in the face of non-standardized expressions, erroneous inputs or malicious attacks in practical applications, and ensuring the security and reliability of the dialogue processing system.

[0129] Optionally, the optimization goal of the adversarial attack algorithm is:

[0130]

[0131] Where x is the input text, x′ is the adversarial input text, and f θ represents the model, θ is the parameter of the model, ∈ Limit the disturbance amplitude, p is often taken as ∞, δ is the disturbance, and y is the predicted result.

[0132] In this embodiment, an adversarial attack algorithm is used to impose a tiny and imperceptible perturbation δ on the conversation text.

[0133] p is often set to ∞ to achieve the L∞ norm constraint and ensure that the disturbance is not easily detected.

[0134] The optimization goal of the adversarial attack algorithm is to generate an adversarial input text through small perturbations so that the model makes the same predictions for the original input and the adversarial input. In other words, by minimally changing the input text, the model is deceived and produces the same output as the original prediction.

[0135] In this embodiment, the optimization goal of the adversarial attack algorithm is:

[0136]

[0137] Where x is the input text, x′ is the adversarial input text, and f θ represents the model, θ is the parameter of the model, ∈ Limit the disturbance amplitude, p is often taken as ∞, δ is the disturbance, y is the prediction result, and the model is optimized by minimizing the adversarial perturbation vector δ, which can make the model difficult to be attacked by adversarial samples, thereby improving the security and robustness of the model.

[0138] Continue to refer Figure 7 , showing Figure 2 Flow chart of a specific implementation manner of step S203 in FIG. Step S203 specifically includes the following steps:

[0139] Step S2031: Generate a target prompt text template through the teacher model.

[0140] In this embodiment, the teacher model is a tool with data processing.

[0141] The above target prompt text template is used to generate the target prompt text.

[0142] Step S2032: Fill the dialogue text to be summarized into the target prompt text template to obtain the target prompt text.

[0143] In this embodiment, the above-mentioned dialogue text to be summarized can be understood as the dialogue content that needs to be summarized or generalized, including one or more dialogue sentences or paragraphs.

[0144] Furthermore, the content in the dialogue text to be summarized is filled into the target prompt text template according to the format and requirements of the target prompt text, thereby obtaining a target prompt text that meets the requirements.

[0145] Step S2033: input the target prompt text into the trained target summary model for summary processing, and output the summary result of the dialogue text to be summarized.

[0146] In this embodiment, the trained target summary model can capture key knowledge information in the input data, extract key knowledge and summarize the input data.

[0147] The above-mentioned summarization process can be understood as the process of extracting important information from the target prompt text after in-depth analysis, and refining this information into a concise and refined statement.

[0148] This application generates a target prompt text template through a teacher model, fills the dialogue text to be summarized into the target prompt text template, obtains the target prompt text, and inputs the target prompt text into a trained target summary model for summary processing, and outputs the summary result of the dialogue text to be summarized. By understanding customer intentions and quickly providing accurate feedback, the processing time is shortened, human errors are reduced, and customer satisfaction and trust in the service are improved.

[0149] like Figure 8 As shown, a structural framework diagram of the data processing method according to the present application is shown. Specifically, it includes: generating a prompt text template and a sample summary case through a pre-trained teacher model, and performing knowledge distillation training on the student model according to the prompt text template and the sample summary case to obtain a trained target summary model, and summarizing the dialogue text to be summarized through the trained target summary model, and outputting the summary result of the dialogue text to be summarized.

[0150] In this embodiment, the pre-trained teacher model is a tool with data processing. The pre-trained teacher model is learned based on a large amount of training corpus. The large amount of training corpus can be a training corpus in the form of question-answer pairs obtained by collating and collecting a large amount of historical dialogue texts and performing data processing on the historical dialogue texts. The data processing can be a data processing method such as natural language processing and manual annotation. The question-answer pair format can be understood as an interactive text format between questions and answers.

[0151] This application introduces an adversarial training mechanism. By adopting an adversarial attack algorithm, the reference cases in the training corpus of the pre-trained teacher model are subjected to adversarial perturbation processing, and an adversarial training step is added to improve the robustness of the model in the face of noise.

[0152] The optimization goal of the adversarial attack algorithm is:

[0153]

[0154] Where x is the input text, x′ is the adversarial input text, and f θ represents the model, θ is the parameter of the model, ∈ Limit the perturbation amplitude, p is often taken as ∞, δ is the perturbation, y is the prediction result, and the model is optimized by minimizing the adversarial perturbation vector δ, which can make the model difficult to be attacked by adversarial samples, thereby improving the security and robustness of the model.

[0155] In this embodiment, the present application simulates various noises and abnormal inputs by introducing an adversarial attack algorithm, forcing the model to learn to resist these interferences during the training process, thereby improving the robustness of the model in the face of non-standardized expressions, erroneous inputs or malicious attacks in practical applications, and ensuring the security and reliability of the dialogue processing system.

[0156] The above prompt text template is used to guide the student model to generate prompt text. The prompt text can be a short phrase or sentence to describe the main content or core idea to be summarized.

[0157] The above sample summary cases are a data set consisting of a set of input texts and corresponding output summaries. The sample summary cases are used to guide the student model to learn how to summarize the input prompt text.

[0158] The student model is a lightweight data processing tool that can learn and imitate the behavior of the teacher model or predict the results.

[0159] The above-mentioned knowledge distillation is a compression and knowledge transfer technology that achieves lightweight and efficient models by transferring the knowledge of the teacher model to the student model. Knowledge distillation uses the soft labels output by the teacher model as supervision signals to guide the training of the student model, so that the student model can better simulate the behavior of the teacher model.

[0160] The above-trained target summary model can capture the key knowledge information in the input data, extract the key knowledge and summarize the input data.

[0161] The above-mentioned dialogue text to be summarized can be understood as the dialogue content that needs to be summarized or generalized, including one or more dialogue sentences or paragraphs.

[0162] The above-mentioned summarization process can be understood as a process of extracting important information from the dialogue text to be summarized after an in-depth analysis, and refining the information into a concise and refined statement.

[0163] In one embodiment, taking insurance business as an example, the input dialogue text of the model is as follows:

[0164] "Seated: Hi, hello

[0165] Seat: Hello

[0166] Customer: Hey, you said

[0167] Agent: Well, for my car, you don't want to buy the commercial insurance, right?

[0168] Customer: Yeah

[0169] Agent: I mean, wait a minute, your car is not for sale yet, right? Customer: Oh

[0170] Customer: Yes, yes, I want to buy compulsory traffic insurance for my car, but the license plate

[0171] Seat: Oh, Xinyang, that’s right, because I see this list is ours

[0172] Agent: Private enterprise Baotong is in xx. I'll ask him to give it to you. Can't we get out?

[0173] Seat: Yes

[0174] Customer: Check it out

[0175] Seat: Mr. Jia, right? May I ask about the water repair xxxxx?

[0176] Customer: Oh, can that be generated?

[0177] Agent: Doctor, is your insurance policy still valid?

[0178] Customer: How can I not care about the parking area?

[0179] Agent: Just a casual remark. Your insurance will expire on XX. If you file a claim before XX, it will be fine.

[0180] Customer: What if I make XX million in six months?

[0181] Agent: Did you pay the money to complete the upgrade in April?

[0182] Customer: Oh, I think I have to. I bought insurance. I'm in the car now. When can I really drive after buying it on the official website?

[0183] Agent: Then you must save the part in April. You can save it from the province until you can save it at any time. Agent: You can come back before the end of XX.

[0184] Customer: Oh, I got a call.

[0185] Customer: Gao Fei, sure.

[0186] Customer: I just borrowed the money, why would I need to remind you of the time?

[0187] Agent: It doesn't matter. If you want to invest, you can do it later.

[0188] Seat: No need to pay now, no impact

[0189] Customer: Ah, no need to guess, I want to tell you that it’s the New Year.”

[0190] The trained target summary model outputs the following input dialogue text:

[0191] "The customer encountered some problems when purchasing compulsory traffic insurance. First, the customer was not sure whether commercial insurance was not necessary. The customer service explained that if the customer did not plan to purchase commercial insurance, he only needed to purchase compulsory traffic insurance. The customer expressed his understanding.

[0192] Next, the customer mentioned his license plate problem. The customer service representative said they would help the customer solve this problem.

[0193] Then the customer asked about the fees. The customer service confirmed that the total fee was XX yuan and told the customer that it could be reimbursed in X month.

[0194] The customer was concerned about the validity of the policy, and the customer service explained that the policy would not expire until XX, and that the upgrade would be sufficient as long as it was completed in X month.

[0195] The customer also asked if there would be a penalty if the upgrade was not completed by X month. The customer service representative said that there might be a penalty if the upgrade was not completed by X month.”

[0196] In this embodiment, the customer demands in the conversation text are analyzed by using the trained target summary model, key information is extracted, and then this information is converted into a concise and clear summary. In this way, the assistant trainer can more conveniently process and analyze customer needs. Next, the assistant trainer will annotate the generated text summary. Through annotation, feedback information can be provided to the model to help the model better understand and summarize customer demands. At the same time, annotation can also be used to evaluate the performance of the model to ensure that the model can accurately identify and summarize customer needs.

[0197] In this embodiment, the application uses large model knowledge distillation technology to extract key knowledge from large and complex models and transfer it to lightweight student models, thereby achieving a significant reduction in computing resource requirements while maintaining high performance. Knowledge distillation technology can compress the rich knowledge and powerful capabilities of large pre-trained models into smaller and less computationally expensive student models, maintaining high performance while reducing deployment and operating costs. This application has been specially optimized for industry-specific language and business logic, and can more accurately and automatically extract key information from text conversations, accurately understand customer intent and quickly provide accurate feedback, shorten processing time, reduce human errors, improve customer satisfaction and trust, and enhance the company's competitiveness.

[0198] In this embodiment, the non-Company software tools or components that appear in the embodiments of this application are merely examples and do not represent actual use.

[0199] This embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0200] The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning. This application can be applied to the fields of artificial intelligence and big data, thereby promoting the construction of smart cities.

[0201] This application belongs to the field of smart cities, and this solution can promote the construction of smart cities.

[0202] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0203] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0204] Further references Fig. 9 , as a response to the above Figure 2 The present application provides an embodiment of a data processing device, and the data processing device embodiment is Figure 2 Corresponding to the data processing method embodiment shown, the data processing device can be specifically applied to various electronic devices.

[0205] like Fig. 9 As shown, the data processing device 900 described in this embodiment includes: a generation module 901, a knowledge distillation training module 902 and a summary processing module 903. Among them:

[0206] A generation module 901, for generating a prompt text template and a sample summary case through a pre-trained teacher model;

[0207] A knowledge distillation training module 902 is used to perform knowledge distillation training on the student model based on the prompt text template and the sample summary case to obtain a trained target summary model;

[0208] The summary processing module 903 is used to perform summary processing on the dialogue text to be summarized by using the trained target summary model, and output the summary result of the dialogue text to be summarized.

[0209] In this embodiment, a prompt text template and a sample summary case are generated through a pre-trained teacher model, and the student model is subjected to knowledge distillation training through the prompt text template and the sample summary case to obtain a trained target summary model, and the dialogue text to be summarized is summarized through the trained target summary model to output a summary result of the dialogue text to be summarized, thereby solving the problems faced by existing traditional dialogue processing methods, such as inaccurate understanding, biased summary, and sensitivity to abnormal input, and the problem that customer service personnel often find it difficult to accurately and quickly understand and summarize customer needs.

[0210] See also Fig.10 ,for Fig. 9 The data processing device 900 is a structural diagram of a specific implementation method, and the data processing device 900 also includes an acquisition module 904 and a training module 905.

[0211] in:

[0212] An acquisition module 904 is used to acquire sample training corpus, wherein the training corpus is a training corpus in the form of question-answer pairs;

[0213] The training module 905 is used to train the teacher model using the sample training corpus to obtain a pre-trained teacher model.

[0214] In this embodiment, by obtaining sample training corpus, the teacher model is trained with the sample training corpus to obtain a pre-trained teacher model. The pre-trained teacher model has strong expression and generalization capabilities and can adapt to different application scenarios and task requirements.

[0215] See also Fig.11 ,for Fig.10The acquisition module 904 is a schematic diagram of a specific implementation scheme, wherein the acquisition module 904 includes an acquisition submodule 9041, a first data preprocessing submodule 9042, and a second data processing submodule 9043.

[0216] The acquisition submodule 9041 is used to acquire the historical conversation text;

[0217] A first data preprocessing submodule 9042, configured to perform a first data preprocessing on the historical conversation text to obtain a preprocessed historical conversation text;

[0218] The second data processing submodule 9043 is used to perform a second data processing on the pre-processed historical dialogue text to obtain a sample training corpus in the form of question-answer pairs.

[0219] In this embodiment, historical conversation texts are obtained, a first data preprocessing is performed on the historical conversation texts to obtain preprocessed historical conversation texts, and a second data processing is performed on the preprocessed historical conversation texts to obtain sample training corpus in the form of question-answer pairs. The sample training corpus in the form of question-answer pairs can be conveniently used in the training process of supervised learning.

[0220] See also Fig.12 ,for Fig. 9 The schematic diagram of the structure of a specific implementation of the generation module 901 in FIG. 9 is a schematic diagram of the structure of a specific implementation of the generation module 901, wherein the generation module 901 includes a first generation submodule 9011 and a first summary processing submodule 9012.

[0221] The first generation submodule 9011 is used to obtain a reference case in the training corpus of the pre-trained teacher model, input the training corpus of the reference case into the prompt text, and generate a prompt text template;

[0222] The first summary processing submodule 9012 is used to input the prompt text of the reference case into the pre-trained teacher model for summary processing and output the sample summary case.

[0223] In this embodiment, by obtaining reference cases from the training corpus of a pre-trained teacher model, inputting the training corpus of the reference cases into the prompt text, generating a prompt text template, and inputting the prompt text of the reference cases into the pre-trained teacher model for summary processing, and outputting sample summary cases, the model can better understand the content and structure of the input text, thereby improving the quality of the summary.

[0224] See also Fig.13 ,for Fig.12 A schematic diagram of a specific implementation of the first generation submodule 9011 in FIG. 9 , wherein the first generation submodule 9011 includes an adversarial perturbation processing unit 90111 and a generation unit 90112.

[0225] An adversarial perturbation processing unit 90111 is used to perform adversarial perturbation processing on reference cases in the training corpus of the pre-trained teacher model using an adversarial attack algorithm to obtain adversarial training corpus;

[0226] The generating unit 90112 is used to input the adversarial training corpus into the prompt text to generate a prompt text template.

[0227] In this embodiment, an adversarial attack algorithm is used to perform adversarial perturbation processing on reference cases in the training corpus of a pre-trained teacher model to obtain adversarial training corpus, and the adversarial training corpus is input into a prompt text to generate a prompt text template. An adversarial attack algorithm is used to simulate various noises and abnormal inputs, forcing the model to learn to resist these interferences during the training process, thereby improving the robustness of the model in the face of non-standardized expressions, erroneous inputs or malicious attacks in practical applications, thereby ensuring the security and reliability of the dialogue processing system.

[0228] Optionally, the optimization goal of the adversarial attack algorithm is:

[0229]

[0230] Where x is the input text, x′ is the adversarial input text, and f θ represents the model, θ is the parameter of the model, ∈ Limit the disturbance amplitude, p is often taken as ∞, δ is the disturbance, and y is the predicted result.

[0231] In this embodiment, the optimization goal of the adversarial attack algorithm is:

[0232]

[0233] Where x is the input text, x′ is the adversarial input text, and f θ represents the model, θ is the parameter of the model, ∈ Limit the disturbance amplitude, p is often taken as ∞, δ is the disturbance, y is the prediction result, and the model is optimized by minimizing the adversarial perturbation vector δ, which can make the model difficult to be attacked by adversarial samples, thereby improving the security and robustness of the model.

[0234] See also Fig.14 ,for Fig. 9 The summary processing module 903 is a schematic diagram of a specific implementation scheme, wherein the summary processing module 903 includes a second generation submodule 9031, a filling submodule 9032 and a second summary processing submodule 9033.

[0235] The second generation submodule 9031 is used to generate a target prompt text template through a teacher model;

[0236] The filling submodule 9032 is used to fill the dialogue text to be summarized into the target prompt text template to obtain the target prompt text;

[0237] The second summary processing submodule 9033 is used to input the target prompt text into the trained target summary model for summary processing, and output the summary result of the dialogue text to be summarized.

[0238] In this embodiment, a target prompt text template is generated through a teacher model, the conversation text to be summarized is filled into the target prompt text template to obtain the target prompt text, and the target prompt text is input into a trained target summary model for summary processing, and the summary result of the conversation text to be summarized is output. By understanding customer intentions and quickly providing accurate feedback, processing time is shortened, human errors are reduced, and customer satisfaction and trust in the service are improved.

[0239] To solve the above technical problems, the present application also provides a computer device. Fig.15 , Fig.15 This is a basic structural block diagram of the computer device in this embodiment.

[0240] The computer device 15 includes a memory 151, a processor 152, and a network interface 153 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 15 with components 151-153, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0241] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0242] The memory 151 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc.

[0243] In some embodiments, the memory 151 may be an internal storage unit of the computer device 15, such as a hard disk or memory of the computer device 15. In other embodiments, the memory 151 may also be an external storage device of the computer device 15, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 15. Of course, the memory 151 may also include both the internal storage unit of the computer device 15 and its external storage device. In this embodiment, the memory 151 is generally used to store the operating system and various application software installed on the computer device 15, such as computer-readable instructions of data processing methods, etc. In addition, the memory 151 may also be used to temporarily store various types of data that have been output or are to be output. The processor 152 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 152 is generally used to control the overall operation of the computer device 15. In this embodiment, the processor 152 is used to execute computer-readable instructions stored in the memory 151 or process data, such as computer-readable instructions for executing the data processing method.

[0244] The network interface 153 may include a wireless network interface or a wired network interface. The network interface 153 is generally used to establish a communication connection between the computer device 15 and other electronic devices.

[0245] In this embodiment, the impact data of the system in the knowledge base can be used to generate data processing affected by the target system to be tested for the user, so that the user can intuitively obtain the impact scope of the target system to be tested. According to the impact scope of the target system to be tested, the impact of the change can be evaluated more accurately and quickly, irrelevant interference points can be eliminated, and regression use cases can be streamlined to avoid omissions of impact points in the early stage of the evaluation, resulting in the subsequent development of various systems. When they are jointly debugged, it is found that some associations have not been processed. In the mildest case, it will take up development time to supplement the logic. In the most serious case, the solution design may need to be overturned and redone, thereby improving the efficiency of project development.

[0246] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the data processing method as described above.

[0247] In this embodiment, the impact data of the system in the knowledge base can be used to generate data processing affected by the target system to be tested for the user, so that the user can intuitively obtain the impact scope of the target system to be tested. According to the impact scope of the target system to be tested, the impact of the change can be evaluated more accurately and quickly, irrelevant interference points can be eliminated, and regression use cases can be streamlined to avoid omissions of impact points in the early stage of the evaluation, resulting in the subsequent development of various systems. When they are jointly debugged, it is found that some associations have not been processed. In the mildest case, it will take up development time to supplement the logic. In the most serious case, the solution design may need to be overturned and redone, thereby improving the efficiency of project development.

[0248] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0249] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A data processing method, characterized in that: The steps include: Generate prompt text templates and sample summary cases through the pre-trained teacher model; Based on the prompt text template and the sample summary case, the student model is trained with knowledge distillation to obtain a trained target summary model; The dialogue text to be summarized is summarized by using the trained target summary model, and a summary result of the dialogue text to be summarized is output.

2. The data processing method according to claim 1, characterized in that: Before generating the prompt text template and the sample summary case by the pre-trained teacher model, the method further includes: Obtaining sample training corpus, wherein the training corpus is in the form of question-answer pairs; The teacher model is trained using the sample training corpus to obtain a pre-trained teacher model.

3. The data processing method according to claim 2, characterized in that The obtaining of sample training corpus includes: Get the historical conversation text; Performing a first data preprocessing on the historical conversation text to obtain a preprocessed historical conversation text; The preprocessed historical conversation text is subjected to a second data processing to obtain a sample training corpus in the form of question-answer pairs.

4. The data processing method according to claim 1, characterized in that: The pre-trained teacher model generates prompt text templates and sample summary cases, including: Obtaining a reference case in the training corpus of the pre-trained teacher model, inputting the training corpus of the reference case into the prompt text, and generating a prompt text template; The prompt text of the reference case is input into the pre-trained teacher model for summary processing, and the sample summary case is output.

5. The data processing method according to claim 4, characterized in that: The example in the training corpus of the pre-trained teacher model is input into the prompt text to generate a prompt text template, including: Adopting an adversarial attack algorithm to perform adversarial perturbation on the reference cases in the training corpus of the pre-trained teacher model to obtain adversarial training corpus; The adversarial training corpus is input into the prompt text to generate a prompt text template.

6. The data processing method according to claim 5, characterized in that: The optimization goal of the adversarial attack algorithm is: Where x is the input text, x′ is the adversarial input text, and f θ represents the model, θ is the parameter of the model, ∈ Limit the disturbance amplitude, p is often taken as ∞, δ is the disturbance, and y is the predicted result.

7. The data processing method according to claim 1, characterized in that: The summarizing process of the dialogue text to be summarized by the trained target summarizing model and outputting the summary result of the dialogue text to be summarized includes: Generate target prompt text template through teacher model; Fill the dialogue text to be summarized into the target prompt text template to obtain the target prompt text; The target prompt text is input into the trained target summary model for summary processing, and the summary result of the dialogue text to be summarized is output.

8. A data processing device, characterized in that: include: A generation module for generating prompt text templates and sample summary cases through a pre-trained teacher model; A knowledge distillation training module is used to perform knowledge distillation training on the student model based on the prompt text template and the sample summary case to obtain a trained target summary model; The summary processing module is used to perform summary processing on the dialogue text to be summarized through the trained target summary model, and output the summary result of the dialogue text to be summarized.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the data processing method according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the data processing method according to any one of claims 1 to 7 are implemented.

Images