Government affair hotline work order dispatching method and system based on large language model

Abstract

The application discloses a government affair hotline work order distribution method and system based on a large language model, relates to the technical field of intelligent government affairs, and solves the technical problem that the prior art often relies on manual judgment and experience and lacks consideration of whether a to-be-distributed department can handle a work order within an effective time, so that the accuracy and efficiency of a work order distribution system are low; work order data is generated according to hotline data; a distribution target function is constructed according to the work order data, and a distribution suggestion is obtained by solving; the work order data is distributed according to the distribution suggestion; the processing time length of the work order data is acquired in real time, an alarm signal is generated according to the processing time length, and local dialect characteristics are considered, so that the work order distribution system can accurately identify the appeals of different groups of people, a distribution target function is constructed, and multiple factors are comprehensively considered, so that work orders can be timely and effectively distributed; and the accuracy and efficiency of the work order distribution system are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent government affairs, and in particular to a government hotline work order dispatching method and system based on a large language model. BACKGROUND

[0002] Large language models are advanced natural language processing tools based on deep learning technology and trained using vast amounts of text data. They can capture complex patterns of language, understand context, and generate coherent and meaningful text. These models have shown great capabilities in text generation, translation, question answering, and other fields, greatly promoting the development of natural language processing technology. Despite challenges such as data privacy and interpretability, large language models still have broad development prospects.

[0003] The government service hotline system is a comprehensive work platform that integrates multiple system functions and implements "one-stop" centralized handling and service. With the deepening of service-oriented government, the number of government hotline work orders has increased dramatically, putting pressure on manual customer service. Moreover, work order matters are complex and involve many departments, making it difficult to identify difficult work orders and accurately dispatch them, which affects work order handling time and public satisfaction. Existing work order dispatch systems often rely on manual judgment and experience, and lack consideration of whether the department to be dispatched can handle the work order within an effective time, resulting in low accuracy and efficiency of the work order dispatch system. Therefore, the government hotline work order dispatch system still needs further improvement. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application proposes a government hotline work order dispatch method and system based on a large language model to solve the technical problem that existing technology often relies on manual judgment and experience, and lacks consideration of whether the department to be dispatched can handle the work order within an effective time, resulting in low accuracy and efficiency of the work order dispatch system.

[0005] To achieve the above-mentioned purpose, the first aspect of the present application provides a government hotline work order dispatch system based on a large language model, comprising: a data acquisition module, a data analysis module, a warning module, and a database.

[0006] The data acquisition module: acquires hotline data through a data acquisition device;

[0007] The data analysis module: generates work order data based on hotline data; solves the dispatch target function after constructing it based on work order data to obtain a dispatch suggestion; dispatches the work order data based on the dispatch suggestion; and acquires the processing duration of the work order data in real time and generates an alarm signal based on the processing duration;

[0008] The warning module: makes a prompt based on the alarm signal and contacts a management personnel;

[0009] The database is used to store historical data required by the training model.

[0010] The present application incorporates the dialect characteristics of different regions through the above steps, ensures that the work order distribution system can accurately capture and understand the needs from different groups, and constructs a distribution target function that fully integrates multiple consideration factors, aiming to achieve efficient and timely distribution of work orders, and improves the accuracy and operation efficiency of the work order distribution system.

[0011] Further, the work order data generated according to the hotline data comprises:

[0012] Obtaining hotline data; the hotline data comprises text data and voice data;

[0013] Judging whether the hotline data is voice data;

[0014] Yes, inputting the voice data into a voice recognition model to obtain audio-text data and audio-text labels; the voice recognition model is constructed by an artificial intelligence model;

[0015] No, inputting the text data or the audio-text data and the corresponding audio-text labels into a large language model to obtain work order data; the large language model is obtained by fine-tuning in the government field data.

[0016] Further, the voice recognition model is constructed by an artificial intelligence model, comprising:

[0017] Obtaining a plurality of historical voice data and corresponding historical audio-text data and audio-text labels;

[0018] Dividing the plurality of historical voice data and corresponding historical audio-text data and audio-text labels into training data, validation data and test data; performing data preprocessing on the training data, validation data and test data to obtain a training set, a validation set and a test set;

[0019] Selecting an artificial intelligence model as a base model;

[0020] Training the base model through the training set, and adjusting the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0021] Verifying the pre-trained model on the test set to finally obtain a voice recognition model that inputs voice data and outputs audio-text data and audio-text labels.

[0022] Further, the large language model is obtained by fine-tuning in the government field data, comprising:

[0023] Obtaining a pre-trained large language model and government field data; the government field data includes historical text data and corresponding work order data, audio-text data and audio-text labels and corresponding work order data;

[0024] Dividing the historical text data and corresponding work order data, audio-text data and audio-text labels and corresponding work order data into training data, validation data and test data; data preprocessing is performed on the training data, validation data and test data to obtain training set, validation set and test set;

[0025] The pre-trained large language model is trained through the training set, and the learning rate and hyperparameters are adjusted on the validation set;

[0026] The pre-trained large language model is verified on the test set, and finally the input text data or audio-text data and the corresponding audio-text label are obtained, and the output is the large language model of the work order data.

[0027] Further, the dispatching target function is constructed according to the work order data, and then the dispatching suggestion is obtained by solving, comprising:

[0028] Obtaining work order data; the work order data includes consultation time, work order location, deadline duration, work order content, content level and corresponding confidence of the department;

[0029] According to the deadline duration, the content level and the corresponding confidence of the department, a plurality of target quantization functions are generated;

[0030] The dispatching target function FC is constructed by formula , , and are dynamic weight coefficients, ∈( , ), ∈( , ), ∈( , ); the dynamic weight coefficients are generated according to the work order data; , , , , and are constants, and , , , , and ∈(0, 1); ND is a maximum grade quantization function, DD is a unit grade; QT is a minimum deadline ratio quantization function, DT is a unit time ratio; Z is a maximum confidence quantization function, DZ is a unit confidence; BRL is a maximum department task completion rate, DL is a unit completion rate;

[0031] The distribution suggestion is generated according to the distribution target function.

[0032] Further, the target quantization function is generated according to the deadline length, content level, and the department and the corresponding confidence, including:

[0033] The work order waiting distribution time is obtained in real time, and the consultation time, deadline length QS, content level, department and corresponding confidence, and department completion rate BWL are obtained.

[0034] The maximum grade quantization function ND is constructed by the formula represents the content level of the i th work order data.

[0035] The waiting time DS is obtained by subtracting the consultation time from the work order waiting distribution time.

[0036] The minimum deadline ratio quantization function QT is constructed by the formula represents the waiting time of the i th work order data. represents the deadline length of the i th work order data.

[0037] The maximum confidence quantization function Z is constructed by the formula represents the j th confidence in the i th work order data.

[0038] The maximum department task completion rate BRL is constructed by the formula represents the department completion rate of the department corresponding to the j th confidence.

[0039] Further, the dynamic weight coefficient is generated according to the work order data, including:

[0040] The work order data is obtained; the work order data includes deadline length, content level, and department and corresponding confidence;

[0041] The work order data is input into the weight estimation model to obtain the corresponding dynamic weight coefficient;

[0042] The weight estimation model is constructed by a machine learning model, including:

[0043] ​​​​Obtain a plurality of historical work order data and corresponding historical weight coefficients thereof;

[0044] The plurality of historical work order data and corresponding historical weight coefficients thereof are divided into training data, validation data and test data; data preprocessing is performed on the training data, validation data and test data to obtain a training set, a validation set and a test set;

[0045] A machine learning model is selected as a base model;

[0046] The base model is trained through the training set, and the learning rate and hyperparameters are adjusted on the validation set to obtain a pre-trained model;

[0047] The pre-trained model is verified on the test set, and finally a weight estimation model of a plurality of dynamic weight coefficients is obtained.

[0048] Through the above steps, the dynamic weight coefficients are updated in real time, so that the system can adaptively adjust the appropriate weight coefficients under different work order data, and the accuracy and efficiency of the work order dispatching system are improved.

[0049] Further, the generation of the dispatching suggestion according to the dispatching target function comprises:

[0050] Obtaining a dispatching target function;

[0051] The optimal solution of the dispatching target function is obtained by a deep reinforcement learning model, and the dispatching suggestion is generated based on the optimal solution.

[0052] Further, the generation of the alarm signal according to the processing time length comprises:

[0053] Real-time acquisition of the processing time length and the deadline time length of the work order data;

[0054] Determine whether the processing time length is less than the deadline time length;

[0055] Yes, determine whether the processing time length is greater than D times the deadline time length; yes, generate a processing near-expiration warning signal; no, do nothing;

[0056] No, generate a processing expiration warning signal; wherein D is a proportionality coefficient, D∈(0, 1).

[0057] Another aspect of the application provides a government affair hotline work order dispatching method based on a large language model, comprising:

[0058] S0: Obtain hotline data;

[0059] S1: Generate work order data according to the hotline data;

[0060] S2: After constructing a dispatch target function according to the work order data, a dispatch suggestion is obtained by solving; and the work order data is dispatched according to the dispatch suggestion;

[0061] S3: Real-time acquisition of the processing time length of the work order data, and generation of an alarm signal according to the processing time length;

[0062] S4: Making a prompt according to the alarm signal, and contacting a management personnel.

[0063] Compared with the prior art, the beneficial effects of the present application are:

[0064] 1. According to the hot line data, the work order data is generated; after constructing a dispatch target function according to the work order data, a dispatch suggestion is obtained by solving; the work order data is dispatched according to the dispatch suggestion; the processing time length of the work order data is acquired in real time, and an alarm signal is generated according to the processing time length; the characteristics of various local dialects are considered, so that the work order dispatch system can accurately identify the appeals of different groups of people, and at the same time, the dispatch target function is constructed, and multiple factors are comprehensively considered, so that the work order can be dispatched in a timely and effective manner; and the accuracy and efficiency of the work order dispatch system are improved.

[0065] 2. According to the present application, a plurality of target quantitative functions are constructed, and a dispatch target function is constructed based on the target quantitative functions, so that the system comprehensively considers the time, grade and confidence degree of the work order and the department to which the work order belongs, and considers the task completion rate of the department based on the department, so as to comprehensively consider and select the dispatched work order data and the dispatch department, and the intelligent level and dispatch efficiency of the work order dispatch system are improved. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0067] Figure 1 It is a schematic diagram of the government hotline work order dispatch system based on a large language model of the present application;

[0068] Figure 2 It is a flow chart of the government hotline work order dispatch method based on a large language model of the present application. DETAILED DESCRIPTION

[0069] The technical solutions of the present application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0070] Please refer to Figure 1 The first aspect of the present application provides a government hotline work order distribution system based on a large language model, which comprises a data acquisition module, a data analysis module, a warning module and a database.

[0071] The data acquisition module acquires hotline data through a data acquisition device; the data acquisition device comprises various sensors and the like.

[0072] The data analysis module generates work order data according to the hotline data; solves the distribution target function after constructing the distribution target function according to the work order data to obtain distribution suggestions; distributes the work order data according to the distribution suggestions; acquires the processing time length of the work order data in real time, and generates an alarm signal according to the processing time length;

[0073] The warning module gives a prompt according to the alarm signal and contacts a management personnel; the alarm signal comprises a processing about-to-expire warning signal and a processing expired warning signal.

[0074] The database is used for storing historical data required by a training model.

[0075] In the present embodiment, the work order data is generated according to the hotline data, which comprises:

[0076] Acquiring hotline data; the hotline data comprises text data and voice data; the hotline data refers to the appeals of enthusiastic masses acquired through a data acquisition device; the voice data refers to the appeals of enthusiastic masses expressed in a voice manner, and the text data refers to the appeals of enthusiastic masses expressed in a text manner;

[0077] Determining whether the hotline data is voice data;

[0078] Yes, inputting the voice data into a voice recognition model to obtain audio-text data and audio-text labels; the voice recognition model is constructed through an artificial intelligence model; the audio-text data refers to data converted from voice to text, and the audio-text labels refer to dialect labels corresponding to the voice data;

[0079] No, inputting the text data or the audio-text data and the corresponding audio-text labels into a large language model to obtain work order data; the large language model is obtained by fine-tuning in the government field data.

[0080] In the present embodiment, the voice recognition model is constructed through an artificial intelligence model, which comprises:

[0081] Obtain a plurality of historical voice data and corresponding historical audio-text data and audio-text labels thereof;

[0082] Divide the plurality of historical voice data and corresponding historical audio-text data and audio-text labels thereof into training data, validation data and test data; perform data preprocessing on the training data, validation data and test data to obtain a training set, a validation set and a test set; the ratio between the training set, the test set and the validation set is 7:2:1;

[0083] Select an artificial intelligence model as a base model; the artificial intelligence model includes an ASR model and the like;

[0084] Train the base model through the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0085] Verify the pre-trained model on the test set to finally obtain a voice recognition model that inputs voice data and outputs audio-text data and audio-text labels.

[0086] The large language model in the embodiment is obtained by fine-tuning in the government field data, including:

[0087] Obtain a pre-trained large language model and government field data; the government field data includes a plurality of historical text data and corresponding work order data, audio-text data and audio-text labels and corresponding work order data; the work order data includes consultation time, work order location, time limit duration, work order content, content level and department to which it belongs, etc.

[0088] Divide the plurality of historical text data and corresponding work order data, audio-text data and audio-text labels and corresponding work order data into training data, validation data and test data; perform data preprocessing on the training data, validation data and test data to obtain a training set, a validation set and a test set; the ratio between the training set, the test set and the validation set is 7:2:1;

[0089] Train the pre-trained large language model through the training set, and adjust the learning rate and hyperparameters on the validation set; the pre-trained large language model includes a ChatGLM model and the like;

[0090] Verify the pre-trained large language model on the test set to finally obtain a large language model that inputs text data or audio-text data and corresponding audio-text labels and outputs work order data.

[0091] The embodiment constructs a speech recognition model, accurately recognizes different personnel appeals in different regions, and recognizes text data and corresponding dialect characteristics according to the speech, and obtains work order data through a large language model on this basis; wherein the large language model is trained through government affairs field data, which can enhance the accurate recognition of user appeals of the large language model, and provides good data support for subsequent work order distribution.

[0092] The distribution suggestion obtained by solving the distribution target function constructed according to the work order data in the embodiment includes:

[0093] Obtain work order data; the work order data includes consultation time, work order location, deadline duration, work order content, content level and corresponding confidence degree ZD of the department to which the work order belongs;

[0094] Generate a plurality of target quantitative functions according to the deadline duration, the content level and the corresponding confidence degree of the department to which the work order belongs;

[0095] The distribution target function FC is constructed through the formula , wherein, , and are dynamic weight coefficients, ∈( , ), ∈( , ), ∈( , ); the dynamic weight coefficients are generated according to the work order data; , , , , and are constants, and , , , , and ∈(0, 1); the specific values are set according to experience; the setting of the constant is to make the dynamic weight coefficient adjust within a certain range, so as to avoid the situation that the target function corresponding to a certain dynamic weight coefficient does not work; ND is a maximum level quantitative function, DD is a unit level; QT is a minimum deadline ratio quantitative function, DT is a unit time ratio; Z is a maximum confidence quantitative function, DZ is a unit confidence; BRL is a maximum department task completion rate, DL is a unit completion rate;

[0096] Generate a distribution suggestion according to the distribution target function.

[0097] According to the deadline length, content level, and the department to which it belongs and the corresponding confidence, a plurality of target quantification functions are generated in this embodiment, including:

[0098] The time when the work order is to be dispatched is obtained in real time, and the consultation time, deadline length QS, content level, department to which it belongs and the corresponding confidence, and department completion rate BWL are obtained;

[0099] The maximum level quantification function ND is constructed by the formula represents the content level of the i-th work order data;

[0100] The waiting time length DS is obtained by subtracting the consultation time from the time when the work order is to be dispatched;

[0101] The minimum deadline proportion rate quantification function QT is constructed by the formula represents the waiting time length of the i-th work order data; represents the deadline length of the i-th work order data;

[0102] The maximum confidence quantification function Z is constructed by the formula represents the j-th confidence in the i-th work order data, and the goal is to select the most suitable department to which it belongs;

[0103] The maximum department task completion rate BRL is constructed by the formula represents the department completion rate of the department corresponding to the j-th confidence.

[0104] In this embodiment, a plurality of target quantification functions are constructed, and the dispatch target function is integrated based thereon to comprehensively evaluate the time urgency, level importance, and confidence of the department to which the work order belongs. At the same time, the system also comprehensively evaluates the task completion rate of the department based on the consideration of the department to which it belongs, and accurately selects the work order data to be dispatched and the most suitable dispatch department, thereby greatly improving the intelligent degree and dispatch efficiency of the work order dispatch system.

[0105] The dynamic weight coefficient in this embodiment is generated according to the work order data, including:

[0106] The work order data is obtained, including the deadline length, content level, and the department to which it belongs and the corresponding confidence;

[0107] The work order data is input into the weight estimation model to obtain the corresponding dynamic weight coefficient;

[0108] The weight estimation model is constructed by a machine learning model, including: ​​​​

[0109] obtain a plurality of historical work order data and corresponding historical weight coefficients;

[0110] divide the plurality of historical work order data and corresponding historical weight coefficients into training data, validation data and test data; perform data preprocessing on the training data, validation data and test data to obtain a training set, a validation set and a test set; the ratio between the training set, the test set and the validation set is 7:2:1;

[0111] select a machine learning model as a base model; the machine learning model includes a convolutional neural network model and the like;

[0112] train the base model through the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0113] verify the pre-trained model on the test set, and finally obtain a weight estimation model that inputs work order data and outputs a plurality of dynamic weight coefficients.

[0114] Through the above steps, the dynamic weight coefficient is updated in real time, so that the system can adaptively adjust the appropriate weight coefficient under different work order data, and the accuracy and efficiency of the work order distribution system are improved.

[0115] In the embodiment, generating a distribution suggestion according to a distribution target function includes:

[0116] obtain a distribution target function;

[0117] obtain an optimal solution of the distribution target function through a deep reinforcement learning model, and generate a distribution suggestion based on the optimal solution; the deep reinforcement learning model is a pre-trained model, which can effectively obtain the optimal solution of the distribution target function and generate a distribution suggestion based on the optimal solution.

[0118] In the embodiment, generating an alarm signal according to a processing time length includes:

[0119] obtain a processing time length and a deadline time length of the work order data in real time;

[0120] determine whether the processing time length is less than the deadline time length;

[0121] Yes, determine whether the processing time length is greater than D times the deadline time length; yes, generate a processing about-to-expire warning signal; no, do nothing;

[0122] No, generate a processing-expired alarm signal; wherein D is a proportion coefficient, D e (0, 1), and the specific value is set according to experience. In the embodiment, D is set to 0.8.

[0123] Please refer to Figure 2Another aspect of the present application provides a government hotline work order distribution method based on a large language model, comprising:

[0124] S0: obtaining hotline data;

[0125] S1: generating work order data according to the hotline data;

[0126] S2: constructing a distribution target function according to the work order data, then solving to obtain a distribution suggestion; distributing the work order data according to the distribution suggestion;

[0127] S3: obtaining the processing time length of the work order data in real time, and generating an alarm signal according to the processing time length;

[0128] S4: making a prompt according to the alarm signal, and contacting a management personnel.

[0129] Some data in the above formula are calculated by removing the dimension, and the formula is obtained by software simulation of a large amount of collected data to obtain a formula closest to the real situation; the preset parameters and the preset threshold in the formula are set by a person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0130] The working principle of the present application is as follows: obtaining hotline data; generating work order data according to the hotline data; constructing a distribution target function according to the work order data, then solving to obtain a distribution suggestion; distributing the work order data according to the distribution suggestion; obtaining the processing time length of the work order data in real time, and generating an alarm signal according to the processing time length; making a prompt according to the alarm signal, and contacting a management personnel; considering the characteristics of various local dialects, so that the work order distribution system can accurately identify the demands of different groups of people, while constructing a distribution target function, comprehensively considering multiple factors to make the work order timely and effectively distributed; improving the accuracy and efficiency of the work order distribution system, avoiding the problem that the prior art often relies on manual judgment and experience, while lacking consideration of whether the to-be-distributed department can process the work order within an effective time, resulting in low accuracy and efficiency of the work order distribution system.

[0131] The above embodiments are only used to illustrate the technical method of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A government affair hotline work order dispatch system based on a large language model, characterized in that, include: Data collection module, data analysis module, early warning module and database; The data acquisition module is used to acquire hotline data through data acquisition equipment; The data analysis module generates work order data based on the hotline data; constructs a dispatch objective function based on the work order data and solves it to obtain dispatch suggestions; and dispatches the work order data based on the dispatch suggestions; Obtain the processing time of work order data in real time and generate an alarm signal based on the processing time; The early warning module: makes prompts according to the alarm signal and contacts the management personnel; The database is used to store historical data required for training the model; Generating work order data according to hotline data includes: Acquiring hotline data; the hotline data includes text data and voice data; Determine whether the hotline data is voice data; Yes, inputting the speech data into a speech recognition model to obtain speech data and speech labels; the speech recognition model is constructed using an artificial intelligence model; No, the text data or audio data and their corresponding audio-text labels are input into a large language model to obtain work order data; the large language model is obtained by fine-tuning data in the government affairs field; The dispatch objective function is constructed based on the work order data and then solved to obtain the dispatch suggestion, including: Obtain work order data; the work order data includes consultation time, work order location, deadline, work order content, content level, department to which it belongs, and its corresponding confidence level ZD; Generate several target quantitative functions based on the deadline, content level, department and their corresponding confidence levels; by formula constructing a dispatch target function FC; wherein, , and are dynamic weight coefficients, ∈( , ), ∈( , ), ∈( , ); the dynamic weight coefficients are generated according to work order data; , , , , and are constants, and , , , , and ∈(0, 1); ND is a maximum grade quantification function, DD is a unit grade; QT is a minimum deadline proportion quantification function, DT is a unit time proportion; Z is a maximum confidence quantification function, DZ is a unit confidence; BRL is a maximum department task completion rate, DL is a unit completion rate; Generate dispatch recommendations based on the dispatch objective function; The target quantitative functions are generated according to the deadline, content level, department and corresponding confidence level, including: Get real-time information on the time it takes for a work order to be dispatched, as well as the consultation time, deadline length (QS), content level, department to which it belongs, its corresponding confidence level, and department completion rate (BWL); By formula constructing a maximum grade quantization function ND; wherein, denotes the content grade of the i-th work order data; The waiting time DS is obtained by subtracting the work order dispatch time from the consultation time; By formula constructing a minimum deadline proportion ratio quantification function QT; wherein, represents the waiting duration of the i th work order data; represents the deadline duration of the i th work order data; By formula constructing a maximum confidence quantification function Z; wherein, represents the jth confidence in the ith work order data; By formula constructing a maximized department task completion rate BRL; wherein, denotes the department completion rate of the department corresponding to the jth confidence level.

2. The government hotline work order dispatching system based on a large language model according to claim 1, characterized in that, The speech recognition model is constructed using an artificial intelligence model, including: Obtaining a number of historical speech data and their corresponding historical audio-text data and audio-text labels; Divide a number of historical speech data and their corresponding historical audio-text data and audio-text labels into training data, verification data, and test data; perform data preprocessing on the training data, verification data, and test data to obtain a training set, a verification set, and a test set; Select an artificial intelligence model as the base model; Train the basic model using the training set, and adjust the learning rate and hyperparameters on the validation set to obtain the pre-trained model; By verifying the pre-trained model on the test set, we finally obtain a speech recognition model that inputs speech data and outputs audio-text data and audio-text labels.

3. The government hotline work order distribution system based on a large language model according to claim 1, characterized in that, The large language model is obtained by fine-tuning on government data, including: Obtain a pre-trained large language model and government data; the government data includes a number of historical text data and its corresponding work order data, audio and text data, and audio and text labels and their corresponding work order data; A plurality of historical text data and corresponding work order data, audio-text data and audio-text labels and corresponding work order data are divided into training data, validation data and test data; the training data, validation data and test data are preprocessed to obtain a training set, a validation set and a test set; The pre-trained large language model is trained through the training set, and the learning rate and the hyperparameter are adjusted on the validation set; The pre-trained large language model is verified on the test set, and finally the input text data or audio-text data and the corresponding audio-text label are obtained, and the large language model outputs the work order data.

4. The government hotline work order distribution system based on a large language model according to claim 1, characterized in that, The dynamic weight coefficient is generated according to the work order data, comprising: Obtain the work order data; the work order data includes deadline duration, content level, and corresponding confidence of the department; The work order data is input into the weight estimation model to obtain the corresponding dynamic weight coefficient; The weight estimation model is constructed by a machine learning model, comprising: Obtain a plurality of historical work order data and corresponding historical weight coefficients; The plurality of historical work order data and corresponding historical weight coefficients are divided into training data, validation data and test data; the training data, validation data and test data are preprocessed to obtain a training set, a validation set and a test set; Select a machine learning model as a base model; The base model is trained through the training set, and the learning rate and the hyperparameter are adjusted on the validation set to obtain a pre-trained model; The pre-trained model is verified on the test set, and finally the input work order data is obtained, and the weight estimation model outputs a plurality of dynamic weight coefficients.

5. The government affair hotline work order dispatching system based on a large language model according to claim 1, characterized in that, The dispatching suggestion is generated according to the dispatching target function, comprising: Obtain the dispatching target function; The production data and the dispatching target function are input into the deep reinforcement learning model to obtain the optimal solution of the dispatching target function, and the dispatching suggestion is generated based on the optimal solution.

6. The government affair hotline work order dispatching system based on a large language model according to claim 1, characterized in that, The alarm signal is generated according to the processing duration, comprising: Real-time acquisition of the processing duration and the deadline duration of the work order data; Determine whether the processing duration is less than the deadline duration; Yes, determine whether the processing duration is greater than D times the deadline duration; yes, generate a processing near-expiration warning signal; no, do nothing; No, generate a processing expiration warning signal; wherein D is a proportion coefficient, D∈(0, 1).

7. The method according to any one of claims 1-6, applied to the system according to any one of claims 1-6, wherein, Comprising: S0: Obtain hotline data; S1: Generate work order data according to hotline data; The work order data is generated according to the hotline data, comprising: Obtain the hotline data; the hotline data includes text data and voice data; Determine whether the hotline data is voice data; Yes, input the voice data into the speech recognition model to obtain audio-text data and audio-text labels; the speech recognition model is constructed by an artificial intelligence model; No, input the text data or audio-text data and corresponding audio-text labels into the large language model to obtain the work order data; the large language model is obtained by fine-tuning in the government affairs field data; S2: After constructing the dispatching target function according to the work order data, the dispatching suggestion is obtained by solving; the work order data is dispatched according to the dispatching suggestion; after constructing the dispatching target function according to the work order data, the dispatching suggestion is obtained by solving, comprising: Obtain work order data; the work order data includes consultation time, work order location, deadline duration, work order content, content level and the confidence of the corresponding department; Generate a number of target quantitative functions according to the deadline duration, content level and the confidence of the corresponding department; by formula constructing a dispatch target function FC; wherein, , and are dynamic weight coefficients, ∈( , ), ∈( , ), ∈( , ) ; the dynamic weight coefficients are generated according to work order data; , , , , and are constants, and , , , , and ∈(0, 1); ND is a maximum grade quantification function, DD is a unit grade; QT is a maximum deadline proportion quantification function, DT is a unit time proportion; Z is a maximum confidence quantification function, DZ is a unit confidence; BRL is a maximum department task completion rate, DL is a unit completion rate; Generate a distribution suggestion according to the distribution target function; The generation of a number of target quantitative functions according to the deadline duration, content level and the confidence of the corresponding department includes: Real-time acquisition of work order to-be-distributed time, and acquisition of consultation time, deadline duration QS, content level, the confidence of the corresponding department and department completion rate BWL; By formula constructing a maximum grade quantization function ND; wherein, denotes the content grade of the i-th work order data; Obtain the waiting duration DS by subtracting the consultation time from the work order to-be-distributed time; By formula constructing a minimum deadline proportion ratio quantification function QT; wherein, represents the waiting duration of the i-th work order data; represents the deadline duration of the i-th work order data; By formula constructing a maximum confidence quantification function Z; wherein, represents the jth confidence in the ith work order data; By formula constructing a maximized department task completion rate BRL; wherein, denotes the department completion rate of the department corresponding to the jth confidence level; S3: Real-time acquisition of the processing duration of the work order data, and generation of an alarm signal according to the processing duration; S4: Make a prompt according to the alarm signal, and contact the management personnel.

Images