Gas operation customer service management system and method based on AI

Abstract

The invention relates to the technical field of gas customer service, in particular to an AI-based gas operation customer service management system and method.The AI-based gas operation customer service management system comprises a data acquisition module, a data judgment module, a first processing module, a second processing module and a data processing module, according to the method, JSON data are acquired, action data corresponding to the JSON data and sequence positions corresponding to the action data are recognized, execution processes corresponding to the action data are sequentially called for execution according to the sequence positions corresponding to the action data, and whether the corresponding action data are correctly recognized or not is judged according to execution results. And when all the action data are correctly recognized, identifying that the user input is recognized successfully, and otherwise, identifying that the user input is recognized unsuccessfully.

Description

Technical Field

[0001] The present invention relates to the field of gas customer service technology, and in particular to an AI-based gas operation customer service management system and method. Background Art

[0002] Gas operation customer service plays a vital role in ensuring safe, efficient and satisfactory gas service.

[0003] The current gas operation customer service system has significant limitations in handling user feedback, which are mainly reflected in the following aspects:

[0004] Existing systems are usually limited to reading text input and cannot process other types of user data (such as pictures, voice, etc.), which limits the way users express their problems and reduces the flexibility and efficiency of interaction.

[0005] For text input, existing systems often rely on keyword matching or simple natural language processing technology, which makes it difficult to accurately understand complex contexts and intentions. Especially when faced with non-standard expressions or ambiguous descriptions, it is easy to make misjudgments or fail to parse.

[0006] Based on this, there is an urgent need for an AI-based gas operation customer service management system and method that can achieve multi-dimensional and accurate recognition of user input and improve the system's recognition accuracy and diversity. Summary of the invention

[0007] One of the purposes of the present invention is to provide an AI-based gas operation customer service management system and method, which can realize multi-dimensional and accurate recognition of user input and improve the recognition accuracy and diversity of the system.

[0008] In order to achieve the above objectives, an AI-based gas operation customer service management system is provided, including:

[0009] A data acquisition module is used to acquire user input data fed back by users;

[0010] The data judgment module is used to judge whether the user input data is image data according to the obtained user input data. If not, the user input data is stored in the history library and the corresponding business AI conversion model is retrieved from the model library; if so, the user input data is stored in the backup library and the image AI recognition model is retrieved from the model library;

[0011] The first processing module is used to input the user input data in the historical database as input data into the retrieved business AI conversion model according to the retrieved business AI conversion model, and parse the corresponding JSON data;

[0012] The second processing module is used to input the user input data in the backup library as input data into the retrieved image AI recognition model according to the retrieved image AI recognition model, output corresponding recognition result data, and parse the recognition result data to parse out corresponding JSON data;

[0013] The data processing module is used to identify the action data corresponding to the JSON data and the sequence corresponding to each action data according to the corresponding JSON data parsed out, and to call up the execution process corresponding to each action data in sequence according to the sequence corresponding to each action data for execution, and to judge whether the corresponding action data is correctly identified based on the execution result. When all the action data are correctly identified, the user input is marked as successfully identified, otherwise the user input is marked as failed to be identified.

[0014] Technical principles and effects of this solution: In this solution, the user input data is first processed, and then the obtained user input data is preliminarily classified to determine whether it is image data, and the appropriate processing path is selected according to the result. Among them, if the user input data is non-image data, the corresponding user input data is input into the business AI conversion model in the model library to output structured JSON data. If it is image data, the user input data is input into the image AI recognition model in the model library to recognize the image data, and the recognition result data is parsed to parse the corresponding JSON data.

[0015] Then, based on the parsed JSON data, determine the specific business steps and perform the corresponding operations in sequence. Check the execution results after each action is completed to confirm whether it is correctly recognized. If all actions are successful, the user input is marked as "recognition successful"; otherwise, it is marked as "recognition failed".

[0016] In this solution, after collecting the user's input data, the type of the user's input data is judged immediately, and the corresponding model is called according to the type of input data, which includes the processing of image data and non-image data. For the processing of non-image data, a dialogue state tracking mechanism is introduced to ensure that the system can maintain its understanding of the context in continuous dialogues and provide a consistent service experience. Users can clarify questions step by step until they get a satisfactory answer. Users can express questions freely in natural language without following a fixed format or using specific keywords, which improves the user experience.

[0017] The processing of image data supports users to upload pictures to describe the problem, making the problem feedback more intuitive. For example, users can take photos of the equipment failure point and send them directly to the customer service system. This not only simplifies the user's operation, but also improves the accuracy of the problem description.

[0018] The system can not only process traditional text input, but also accurately analyze image content to provide users with all-round support. This diversified input support method greatly enriches the means for users to express their problems and improves the flexibility and efficiency of services. It can achieve multi-dimensional and accurate recognition of user input, improving the system's recognition accuracy and recognition diversity.

[0019] Furthermore, it also includes a business execution module, which is used to determine and execute the business steps corresponding to the JSON data based on the corresponding JSON data when the judgment result indicates that the user input is successfully recognized.

[0020] Beneficial effects: Automated business processes reduce the need for manual intervention, lower the probability of errors, and speed up response. User problems can be solved faster and more accurately.

[0021] Furthermore, it also includes a model training module, which is used to build and train the business AI conversion model based on historical user input data in the historical database, and store the business AI conversion model in the corresponding model library after the training is completed.

[0022] Furthermore, the model training module includes:

[0023] A collection module is used to collect a number of historical user input data from a historical database;

[0024] A cleaning module is used to clean and pre-process the collected historical user input data;

[0025] The annotation module is used to annotate each collected historical user input data, and the annotation content is the JSON annotation result data corresponding to the historical user input data;

[0026] A partitioning module is used to partition the annotated historical user data into a training set and a test set;

[0027] Building modules for building business AI transformation models;

[0028] A parameter determination module is used to determine the model thresholds corresponding to the various model parameters corresponding to the business AI conversion model according to the constructed business AI conversion model, and form model intervals corresponding to the corresponding model parameters;

[0029] The training module is used to train the business AI conversion model according to the model interval corresponding to each model parameter, each historical user input data in the training set, and the constructed business AI conversion model based on the preset model training strategy until the training is completed, and input the corresponding trained business AI conversion model;

[0030] The verification model is used to input the historical input data in the test set as input data into the trained business AI conversion model according to the trained business AI conversion model, output the corresponding verification results, and based on the preset verification strategy, determine whether the corresponding trained business AI conversion model is feasible. If so, input the corresponding business AI conversion model, otherwise, re-execute the training module.

[0031] Beneficial effects: Several historical user input data are collected from the historical database. By widely collecting user input in different scenarios, the diversity and representativeness of the training data are ensured, and the generalization ability of the model is improved. Then the collected historical user input data is cleaned and preprocessed. Operations such as removing noise, correcting incorrect formats, and unifying standards improve the accuracy and consistency of the data. Eliminating unnecessary information makes the model training more focused on key features and improves the parsing accuracy. The collected historical user input data is then annotated, and the annotated content is the corresponding JSON annotation result data. Through a professional annotation team or tool, ensure that each input data is correctly parsed and mapped to a structured JSON format, thereby enhancing the model's understanding ability. The unified annotation standard ensures that data from different sources has a consistent structure when entering the model, simplifying the subsequent processing logic.

[0032] Furthermore, the preset model training strategy is:

[0033] S1. According to the model interval corresponding to each model parameter, determine the intermediate value P of the model interval corresponding to each model parameter Z , and based on the intermediate value, divide the corresponding model interval into two model sub-intervals;

[0034] S2, randomly select a parameter value P1 and P2 in the two model sub-intervals respectively, and calculate the value of P1, P2 and P2 according to the corresponding P1, P2 and P2 of each model parameter. Z , select a parameter value from each model parameter, and combine all model parameters to form multiple model parameter groups;

[0035] S3. Use the historical user input data in the training set as input data, input it into the business AI conversion model corresponding to each model parameter group, and output the corresponding training result data;

[0036] S4. Evaluate the model reliability of the business AI conversion model corresponding to each model parameter group based on the training result data corresponding to each model parameter group and the corresponding annotation content;

[0037] S5. According to the model reliability corresponding to each model parameter group, a model parameter group with the largest model reliability is selected, and the model parameter group is used as the first group;

[0038] S6. According to the selected first group, based on the corresponding model interval, the model interval is divided into two model sub-intervals again, and parameter values ​​corresponding to each model parameter are randomly selected in the two model sub-intervals, and randomly combined into a new model parameter group;

[0039] S7, re-execute S3 to S5 according to the new model parameter group, select the second group corresponding to the new model parameter group, and determine whether the model reliability corresponding to the second group is greater than the model reliability corresponding to the first group. If so, the second group is the optimal group, otherwise the first group is still the optimal group;

[0040] S8. After determining the optimal group, re-execute S6 until the preset conditions are met, and output the corresponding optimal group as the final model parameter group corresponding to the business AI conversion model.

[0041] Beneficial effects: In this solution, according to the model intervals corresponding to each model parameter, the middle value of each interval is calculated, and the interval is divided into two model sub-intervals. By subdividing the model parameter interval, the subsequent parameter selection is more refined, and the efficiency of searching for the optimal parameter combination is improved. Based on the middle value division, the extreme situations that may occur during random selection are avoided, and the uncertainty of the search space is reduced.

[0042] Then, a parameter value is randomly selected in each of the two model subintervals and combined with the intermediate value to form multiple model parameter groups. By randomly selecting different parameter values, the diversity of parameter combinations is ensured, increasing the possibility of finding the optimal solution.

[0043] Afterwards, the historical user input data in the training set is used as input data and input into the business AI conversion model corresponding to each model parameter group, and the corresponding training result data is output. The performance of each model parameter group is verified by actual data, providing an immediate basis for performance evaluation. Training based on real data ensures that the model can adapt to actual application scenarios, improving the practicality and reliability of the model.

[0044] Then, based on the training result data and annotation content, the model reliability of the business AI conversion model corresponding to each model parameter group is evaluated. Objective evaluation indicators are used to ensure the fairness and transparency of the evaluation process. Based on the evaluation results, the model parameters are continuously adjusted to gradually improve the model performance. According to the evaluation results, the model parameter group with the highest model reliability is selected as the first group. It provides a benchmark for subsequent iterations and clarifies the direction of optimization. Based on the model interval of the first group, it is divided into two model sub-intervals again, and parameter values ​​are randomly selected to form a new model parameter group. Through multiple subdivisions, the optimal parameter combination is gradually approached, which improves the search accuracy. The search range is dynamically adjusted according to the current optimal parameter group to ensure that each iteration can bring performance improvement. Re-execute S3 to S5 according to the new model parameter group, select the second group, and determine whether its model reliability is greater than the first group. Through multiple rounds of iteration and comparison, ensure that the final selected parameter combination is the best. Through repeated verification, the model is prevented from overfitting the training data and the generalization ability is improved.

[0045] Furthermore, the preset verification strategy is:

[0046] According to the annotation content corresponding to each historical input data in the test set and the corresponding verification results, the verification accuracy number corresponding to the model parameter group is counted, and the corresponding verification accuracy is calculated;

[0047] The verification accuracy calculation formula is:

[0048]

[0049] In the formula, Z is the verification accuracy, y k is the annotation content corresponding to the k-th historical input data in the validation set, is the verification result corresponding to the kth historical input data in the verification set, and m is the total number of historical input data in the verification set;

[0050] If the verification accuracy is greater than the preset accuracy threshold, the business AI conversion model corresponding to the model parameter group is judged to be a feasible model; otherwise, the corresponding business AI conversion model is judged to be an infeasible model and the training module is re-executed.

[0051] Beneficial effects: In this solution, verification accuracy is used as the core indicator to measure model performance, providing a clear and quantifiable evaluation standard to facilitate comparison of the performance of different models. For models that fail verification, a retraining mechanism is automatically triggered to form a closed loop of continuous optimization to ensure continuous improvement of the system. Through the automated verification and training cycle, manual intervention is reduced and the operating efficiency of the entire system is improved.

[0052] The present invention also provides an AI-based gas operation customer service management method, using the above-mentioned AI-based gas operation customer service management system. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a logical block diagram of the AI-based gas operation customer service management system in Example 1 of the present invention.

[0054] Figure 2 This is a flowchart corresponding to the AI-based gas operation customer service management system in Example 1 of the present invention. DETAILED DESCRIPTION

[0055] The following is further described in detail through specific implementation methods:

[0056] Embodiment 1

[0057] An AI-based gas operation customer service management system, basically Figure 1 and Figure 2 As shown, including:

[0058] A data acquisition module is used to acquire user input data fed back by users;

[0059] The data judgment module is used to judge whether the user input data is image data based on the obtained user input data. If not, the user input data is stored in the history library and the corresponding business AI conversion model is retrieved from the model library; if so, the user input data is stored in the backup library and the image AI recognition model is retrieved from the model library; in this embodiment, the image AI recognition model is constructed using the existing convolutional neural network.

[0060] The first processing module is used to input the user input data in the historical database as input data into the retrieved business AI conversion model according to the retrieved business AI conversion model, and parse the corresponding JSON data;

[0061] The second processing module is used to input the user input data in the backup library as input data into the retrieved image AI recognition model according to the retrieved image AI recognition model, output corresponding recognition result data, and parse the recognition result data to parse out corresponding JSON data;

[0062] In this embodiment, the JSON data must include the following three parameters:

[0063] Instruction (action), object (targetObj) and parameter (param). Instruction (action): contains actions, such as start, end, edit, save, get, etc. Object (targetObj): contains basic business objects. Parameter (param): contains system configuration parameters.

[0064] For example [{"action":"get","targetObj":"security","param":[{"gasCode":"123"}]}]

[0065] The data processing module is used to identify the action data corresponding to the JSON data and the sequence corresponding to each action data according to the corresponding JSON data parsed out, and to call up the execution process corresponding to each action data in sequence according to the sequence corresponding to each action data for execution, and to judge whether the corresponding action data is correctly identified based on the execution result. When all the action data are correctly identified, the user input is marked as successfully identified, otherwise the user input is marked as failed to be identified.

[0066] The business execution module is used to determine and execute the business steps corresponding to the JSON data according to the corresponding JSON data when the judgment result indicates that the user input is successfully recognized.

[0067] It also includes a model training module, which is used to build and train the business AI conversion model based on the historical user input data in the historical database, and store the business AI conversion model in the corresponding model library after the training is completed.

[0068] The model training module includes:

[0069] A collection module is used to collect a number of historical user input data from a historical database;

[0070] A cleaning module is used to clean and pre-process the collected historical user input data;

[0071] The annotation module is used to annotate each collected historical user input data, and the annotation content is the JSON annotation result data corresponding to the historical user input data;

[0072] A partitioning module is used to partition the annotated historical user data into a training set and a test set;

[0073] A construction module is used to construct a business AI conversion model; in this embodiment, the business AI conversion model is constructed using BP neural network technology. BP neural networks usually use Sigmoid differentiable functions and linear functions as the network's excitation functions. This paper selects the S-type tangent function tansig as the excitation function of the hidden layer neurons. The prediction model selects the S-type logarithmic function tansig as the excitation function of the output layer neurons.

[0074] A parameter determination module is used to determine the model thresholds corresponding to the various model parameters corresponding to the business AI conversion model according to the constructed business AI conversion model, and form model intervals corresponding to the corresponding model parameters;

[0075] The training module is used to train the business AI conversion model according to the model interval corresponding to each model parameter, each historical user input data in the training set, and the constructed business AI conversion model based on the preset model training strategy until the training is completed, and input the corresponding trained business AI conversion model;

[0076] The preset model training strategy is:

[0077] S1. According to the model interval corresponding to each model parameter, determine the intermediate value P of the model interval corresponding to each model parameter Z , and based on the intermediate value, divide the corresponding model interval into two model sub-intervals;

[0078] S2, randomly select a parameter value P1 and P2 in the two model sub-intervals respectively, and calculate the value of P1, P2 and P2 according to the corresponding P1, P2 and P2 of each model parameter. Z , select a parameter value from each model parameter, and combine all model parameters to form multiple model parameter groups;

[0079] S3. Use the historical user input data in the training set as input data, input it into the business AI conversion model corresponding to each model parameter group, and output the corresponding training result data;

[0080] S4. According to the training result data corresponding to each model parameter group and the corresponding annotation content, the model reliability corresponding to the business AI conversion model corresponding to each model parameter group is evaluated; in this embodiment, the model reliability is associated with the total number of model trainings, the content overlap between the training result data and the annotation content, and the number of training result data that are the same as the annotation content.

[0081] S5. According to the model reliability corresponding to each model parameter group, a model parameter group with the largest model reliability is selected, and the model parameter group is used as the first group;

[0082] S6. According to the selected first group, based on the corresponding model interval, the model interval is divided into two model sub-intervals again, and parameter values ​​corresponding to each model parameter are randomly selected in the two model sub-intervals, and randomly combined into a new model parameter group;

[0083] S7, re-execute S3 to S5 according to the new model parameter group, select the second group corresponding to the new model parameter group, and determine whether the model reliability corresponding to the second group is greater than the model reliability corresponding to the first group. If so, the second group is the optimal group, otherwise the first group is still the optimal group;

[0084] S8. After determining the optimal group, re-execute S6 until the preset conditions are met, and output the corresponding optimal group as the final model parameter group corresponding to the business AI conversion model.

[0085] For example, the model thresholds corresponding to each model parameter are: model parameter A: [A_1, A_2], model parameter B: [B_1, B_2], model parameter C: [C_1, C_2], and according to the intermediate values ​​corresponding to each model parameter, it is divided into two model sub-intervals, for example, sub-interval 1: [A_1, P_(Z_A)], sub-interval 2: [P_(Z_A), A_2].

[0086] Then, according to the sub-intervals and intermediate values ​​corresponding to each model parameter, a random team is formed to form multiple model parameter groups. For example, a certain parameter group is [P_{1_A}, P_{1_B}, P_{1_C}]. The historical user input data in the training set is used as input data and input into the business AI conversion model corresponding to each model parameter group, and the corresponding training result data is output. According to the training result data and the annotation content, the model reliability of the business AI conversion model corresponding to each model parameter group is evaluated, and the model parameter group with the largest model reliability is selected. Based on the position of the model parameter group in the model interval corresponding to each model parameter, a new sub-interval is re-formed, and parameter values ​​are randomly selected to form a new model parameter group. Finally, S3 to S5 are re-executed according to the new model parameter group, and the second group corresponding to the new model parameter group is selected, and its model reliability is judged to be greater than the first group. If the accuracy of the newly selected parameter combination is higher than 0.90, it is updated to the new optimal group; otherwise, the first group is kept as the optimal group.

[0087] The verification model is used to input the historical input data in the test set as input data into the trained business AI conversion model according to the trained business AI conversion model, output the corresponding verification results, and based on the preset verification strategy, determine whether the corresponding trained business AI conversion model is feasible. If so, input the corresponding business AI conversion model, otherwise, re-execute the training module.

[0088] The preset verification strategy is:

[0089] According to the annotation content corresponding to each historical input data in the test set and the corresponding verification results, the verification accuracy number corresponding to the model parameter group is counted, and the corresponding verification accuracy is calculated;

[0090] The verification accuracy calculation formula is:

[0091]

[0092] In the formula, Z is the verification accuracy, y k is the annotation content corresponding to the k-th historical input data in the validation set, is the verification result corresponding to the kth historical input data in the verification set, and m is the total number of historical input data in the verification set;

[0093] If the verification accuracy is greater than the preset accuracy threshold, the business AI conversion model corresponding to the model parameter group is judged to be a feasible model; otherwise, the corresponding business AI conversion model is judged to be an infeasible model and the training module is re-executed.

[0094] This embodiment also discloses an AI-based gas operation customer service management method, which uses the above-mentioned AI-based gas operation customer service management system.

[0095] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is described too much here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can know all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A gas operation customer service management system based on AI, characterized by: include: A data acquisition module is used to acquire user input data fed back by users; The data judgment module is used to judge whether the user input data is image data according to the obtained user input data. If not, the user input data is stored in the history library and the corresponding business AI conversion model is retrieved from the model library; if so, the user input data is stored in the backup library and the image AI recognition model is retrieved from the model library; The first processing module is used to input the user input data in the historical database as input data into the retrieved business AI conversion model according to the retrieved business AI conversion model, and parse the corresponding JSON data; The second processing module is used to input the user input data in the backup library as input data into the retrieved image AI recognition model according to the retrieved image AI recognition model, output corresponding recognition result data, and parse the recognition result data to parse out corresponding JSON data; The data processing module is used to identify the action data corresponding to the JSON data and the sequence corresponding to each action data according to the corresponding JSON data parsed out, and to call up the execution process corresponding to each action data in sequence according to the sequence corresponding to each action data for execution, and to judge whether the corresponding action data is correctly identified based on the execution result. When all the action data are correctly identified, the user input is marked as successfully identified, otherwise the user input is marked as failed to be identified.

2. According to claim 1, the AI-based gas operation customer service management system is characterized by: It also includes a business execution module, which is used to determine and execute the business steps corresponding to the JSON data based on the corresponding JSON data when the judgment result indicates that the user input is successfully recognized.

3. The AI-based gas operation customer service management system according to claim 2, characterized in that: It also includes a model training module, which is used to build and train the business AI conversion model based on the historical user input data in the historical database, and store the business AI conversion model in the corresponding model library after the training is completed.

4. The AI-based gas operation customer service management system according to claim 3 is characterized by: The model training module includes: A collection module is used to collect a number of historical user input data from a historical database; A cleaning module is used to clean and pre-process the collected historical user input data; The annotation module is used to annotate each collected historical user input data, and the annotation content is the JSON annotation result data corresponding to the historical user input data; A partitioning module is used to partition the annotated historical user data into a training set and a test set; Building modules for building business AI transformation models; A parameter determination module is used to determine the model thresholds corresponding to the various model parameters corresponding to the business AI conversion model according to the constructed business AI conversion model, and form model intervals corresponding to the corresponding model parameters; The training module is used to train the business AI conversion model according to the model interval corresponding to each model parameter, each historical user input data in the training set, and the constructed business AI conversion model based on the preset model training strategy until the training is completed, and input the corresponding trained business AI conversion model; The verification model is used to input the historical input data in the test set as input data into the trained business AI conversion model according to the trained business AI conversion model, output the corresponding verification results, and based on the preset verification strategy, determine whether the corresponding trained business AI conversion model is feasible. If so, input the corresponding business AI conversion model, otherwise, re-execute the training module.

5. The AI-based gas operation customer service management system according to claim 4 is characterized by: The preset model training strategy is: S1. According to the model interval corresponding to each model parameter, determine the intermediate value P of the model interval corresponding to each model parameter Z , and based on the intermediate value, divide the corresponding model interval into two model sub-intervals; S2, randomly select a parameter value P1 and P2 in the two model sub-intervals respectively, and calculate the value of P1, P2 and P2 according to the corresponding P1, P2 and P2 of each model parameter. Z , select a parameter value from each model parameter, and combine all model parameters to form multiple model parameter groups; S3. Use the historical user input data in the training set as input data, input it into the business AI conversion model corresponding to each model parameter group, and output the corresponding training result data; S4. Evaluate the model reliability of the business AI conversion model corresponding to each model parameter group based on the training result data corresponding to each model parameter group and the corresponding annotation content; S5. According to the model reliability corresponding to each model parameter group, a model parameter group with the largest model reliability is selected, and the model parameter group is used as the first group; S6. According to the selected first group, based on the corresponding model interval, the model interval is divided into two model sub-intervals again, and parameter values ​​corresponding to each model parameter are randomly selected in the two model sub-intervals, and randomly combined into a new model parameter group; S7, re-execute S3 to S5 according to the new model parameter group, select the second group corresponding to the new model parameter group, and determine whether the model reliability corresponding to the second group is greater than the model reliability corresponding to the first group. If so, the second group is the optimal group, otherwise the first group is still the optimal group; S8. After determining the optimal group, re-execute S6 until the preset conditions are met, and output the corresponding optimal group as the final model parameter group corresponding to the business AI conversion model.

6. The AI-based gas operation customer service management system according to claim 5, characterized in that: The preset verification strategy is: According to the annotation content corresponding to each historical input data in the test set and the corresponding verification results, the verification accuracy number corresponding to the model parameter group is counted, and the corresponding verification accuracy is calculated; The verification accuracy calculation formula is: In the formula, Z is the verification accuracy, y k is the annotation content corresponding to the k-th historical input data in the validation set, is the verification result corresponding to the kth historical input data in the verification set, and m is the total number of historical input data in the verification set; If the verification accuracy is greater than the preset accuracy threshold, the business AI conversion model corresponding to the model parameter group is judged to be a feasible model; otherwise, the corresponding business AI conversion model is judged to be an infeasible model and the training module is re-executed.

7. A gas operation customer service management method based on AI, characterized by: An AI-based gas operation customer service management system using any one of claims 1 to 6 above.

Images