Data generation method based on power grid customer service large model and related device
Through the data generation method of the power grid customer service large model, the problems of inaccurate and biased data generation are solved, the robustness and generalization ability of the model are improved, the data acquisition cost is reduced, and efficient data expansion and model training are achieved.
Patent Information
- Application Number
- CN202510784820.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-12
- Publication Date
- 2025-09-19
AI Technical Summary
Existing technologies have problems with inaccurate and biased data generation in large power grid customer service models, especially when the number of data sets is small, resulting in poor model training results.
By preprocessing, data enhancement, semantic constraints and indicator constraints on the original sentences, deep neural networks are used to screen out high-quality new sentences and generate data that meets actual application needs.
It improves the robustness and generalization ability of the model, reduces the cost of data acquisition, and enhances the stability and efficiency of the model in practical applications.
Smart Images

Figure CN120670849A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of data generation, and in particular relates to a data generation method based on a power grid customer service large model and related devices. Background Art
[0002] With the rapid development of big models and artificial intelligence technologies, enterprise digital transformation has become an irreversible trend. Traditional customer service models rely on manual online conversations to provide one-on-one support to resolve customer inquiries, but the customer service industry remains highly reliant on manpower. In recent years, breakthroughs in big model technology have driven profound changes in natural language processing, reshaped the technical architecture of the customer service industry, and profoundly impacted its future development. The massive parameter size and complex network structure of big language models enable them to demonstrate exceptional capabilities in tasks such as conversational speech, but their deployment in everyday applications remains challenging due to computing resource constraints. Customized training for specific industries can significantly improve the model's applicability. In the field of intelligent customer service, companies can leverage their own business data to optimize big models, creating more accurate and efficient customer service systems that address the shortcomings of traditional customer service in terms of efficiency and knowledge coverage.
[0003] The distribution and quality of datasets have a significant impact on fine-tuning large language models. Well-distributed, high-quality datasets can accelerate model training and improve performance. However, currently, for specific scenarios, customized enterprise training often faces challenges with limited datasets, or even insufficient datasets to meet the requirements for model fine-tuning. Therefore, data generation technology can effectively address the issue of limited data volumes and increase data diversity and complexity.
[0004] Although current data generation technology can effectively expand the data set, the generated data may be inaccurate or biased, especially when the model fails to fully capture the complexity of real data. These generated samples may not truly reflect the data distribution in actual applications, resulting in the newly generated data distribution being different from the original data distribution, leading to inaccurate and biased data generation problems. Summary of the Invention
[0005] The purpose of the present invention is to provide a data generation method and related devices based on a large power grid customer service model to solve the problems of inaccuracy and deviation in data generation in the prior art.
[0006] To achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a data generation method based on a power grid customer service large model, characterized by comprising: Obtain the original sentence as a data set, preprocess the data set, and obtain a preprocessed data set; Perform data enhancement on the preprocessed data set to obtain a data-enhanced data set; Generate text sentences from the enhanced dataset through semantic constraints and indicator constraints; The text sentence and the original sentence are used as the input of the deep neural network, the similarity between the text sentence and the original sentence is output, and new sentences are filtered out as new data of the dataset based on the similarity.
[0007] Furthermore, the original sentence is obtained as a data set, and the data set is preprocessed to obtain a preprocessed data set, including: Use data processing tools to detect duplicate values in the data set; scan each item in the data set one by one, and fill in null if missing; use re-encoding and filtering illegal characters to handle text anomalies such as garbled text and inconsistent formatting to obtain the preprocessed data set.
[0008] Furthermore, performing data enhancement on the preprocessed data set to obtain a data-enhanced data set includes: Using context-level augmentation operations, a semantic rewriting method based on synonym replacement is used to replace words and randomize the word order to generate new sentences. Using non-contextual enhancement operations, a semantic rewriting method of noise injection and word order randomization and deletion is used to generate new sentences; All the new sentences generated above are mixed into the original dataset in a completely random manner, covering different semantic levels, to obtain a data-enhanced dataset.
[0009] Furthermore, the context-level enhancement operation specifically includes: Synonym replacement: randomly select n non-stop words in a sentence and replace them with synonyms based on the contextual semantics; Randomly select two words in the sentence and swap their positions, repeat n times, and get a new sentence; Randomly select a non-stop word from the newly obtained sentence, convert it into a synonym and insert it into a random position, repeating n times.
[0010] Furthermore, the non-contextual enhancement operation specifically includes: Randomly delete words in the sentence with probability p, and repeat the number of times to be one tenth of the number of words in the sentence; For text data, two forms of noise injection are used: spelling errors of the same word and word replacement. Word replacement refers to randomly replacing a character with the letter of the adjacent word on the keyboard to simulate accidental keyboard presses.
[0011] Furthermore, generating text sentences from the enhanced dataset through semantic constraints and indicator constraints includes: Build a pre-trained language model as a semantic constraint to represent and filter the text semantics that meet the conditions; Set two indicators: similarity threshold and semantic coverage. New sentences that meet the conditions are retained as subsequent input data.
[0012] Furthermore, the method of using the text sentence and the original sentence as inputs of the deep neural network, outputting the similarity between the text sentence and the original sentence, and screening out new sentences as new data of the data set based on the similarity includes: Build and deploy a deep neural network, taking text sentences and original sentences as input. The output of the deep neural network is the similarity between the newly generated input sentences and the original sentences. Based on the similarity, suitable new sentences are selected as new data in the dataset. Optimize the training process of the deep neural network, adjust the training parameters, use the cross entropy loss function, and use the gradient descent method to complete the adjustment and update of experimental parameters.
[0013] In a second aspect, the present invention provides a data generation system based on a power grid customer service model, comprising: The data acquisition module is used to obtain the original sentence as a data set, preprocess the data set, and obtain the preprocessed data set; The data enhancement module is used to perform data enhancement on the preprocessed data set to obtain a data-enhanced data set; The text generation module is used to generate text sentences from the enhanced dataset through semantic constraints and indicator constraints; The filtering output module is used to take the text sentences and original sentences as input of the deep neural network, output the similarity between the text sentences and the original sentences, and filter out new sentences as new data of the dataset based on the similarity.
[0014] Furthermore, in the data acquisition module, the original sentence is acquired as a data set, and the data set is preprocessed to obtain a preprocessed data set, including: Use data processing tools to detect duplicate values in the data set; scan each item in the data set one by one, and fill in null if missing; use re-encoding and filtering illegal characters to handle text anomalies such as garbled text and inconsistent formatting to obtain the preprocessed data set.
[0015] Furthermore, in the data enhancement module, data enhancement is performed on the preprocessed data set to obtain a data-enhanced data set, including: Using context-level augmentation operations, a semantic rewriting method based on synonym replacement is used to replace words and randomize the word order to generate new sentences. Using non-contextual enhancement operations, a semantic rewriting method of noise injection and word order randomization and deletion is used to generate new sentences; All the new sentences generated above are mixed into the original dataset in a completely random manner, covering different semantic levels, to obtain a data-enhanced dataset.
[0016] Furthermore, the context-level enhancement operation specifically includes: Synonym replacement: randomly select n non-stop words in a sentence and replace them with synonyms based on the contextual semantics; Randomly select two words in the sentence and swap their positions, repeat n times, and get a new sentence; Randomly select a non-stop word from the newly obtained sentence, convert it into a synonym and insert it into a random position, repeating n times.
[0017] Furthermore, the non-contextual enhancement operation specifically includes: Randomly delete words in the sentence with probability p, and repeat the number of times to be one tenth of the number of words in the sentence; For text data, two forms of noise injection are used: spelling errors of the same word and word replacement. Word replacement refers to randomly replacing a character with the letter of the adjacent word on the keyboard to simulate accidental keyboard presses.
[0018] Furthermore, in the text generation module, the enhanced data set is used to generate text sentences through semantic constraints and indicator constraints, including: Build a pre-trained language model as a semantic constraint to represent and filter the text semantics that meet the conditions; Set two indicators: similarity threshold and semantic coverage. New sentences that meet the conditions are retained as subsequent input data.
[0019] Furthermore, in the screening output module, the text sentence and the original sentence are used as inputs of the deep neural network, the similarity between the text sentence and the original sentence is output, and new sentences are screened out as new data of the data set based on the similarity, including: Build and deploy a deep neural network, taking text sentences and original sentences as input. The output of the deep neural network is the similarity between the newly generated input sentences and the original sentences. Based on the similarity, suitable new sentences are selected as new data in the dataset. Optimize the training process of the deep neural network, adjust the training parameters, use the cross entropy loss function, and use the gradient descent method to complete the adjustment and update of experimental parameters.
[0020] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the data generation method based on the power grid customer service large model when executing the computer program.
[0021] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the data generation method based on the power grid customer service model.
[0022] Compared with the prior art, the present invention has the following technical effects: This invention generates high-quality data that meets the needs of practical applications, reduces data acquisition costs, improves the robustness and generalization capabilities of the model, and thus enhances its stability and efficiency in practical applications. In scenarios where large language models with customer service functions are used and the number of datasets is small or insufficient to meet training requirements, data augmentation and diversity constraints address the problem of insufficient data volume, effectively expanding the training dataset and improving the robustness and generalization capabilities of the model. In addition, for datasets that require annotation, reliance on large-scale manually annotated data is reduced, reducing data acquisition costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 It is a flow chart of the present invention.
[0024] Figure 2 This is a logic block diagram of the present invention. DETAILED DESCRIPTION
[0025] The present invention is further described below with reference to the accompanying drawings: Example 1, please refer to Figure 1 The present invention provides a data generation method based on a power grid customer service large model, comprising: Obtain the original sentence as a data set, preprocess the data set, and obtain a preprocessed data set; Perform data enhancement on the preprocessed data set to obtain a data-enhanced data set; Generate text sentences from the enhanced dataset through semantic constraints and indicator constraints; The text sentence and the original sentence are used as the input of the deep neural network, the similarity between the text sentence and the original sentence is output, and new sentences are filtered out as new data of the dataset based on the similarity.
[0026] This paper proposes a data generation method based on a large-scale model system for power grid customer service. This method is suitable for scenarios where a large language model with customer service functions is used and the dataset is small or insufficient to meet model training requirements. This method uses data generation technology to effectively expand the training dataset based on specific characteristics and requirements, improving data diversity and coverage, and enhancing the model's adaptability to different scenarios.
[0027] Example 2, please refer to Figure 2The present invention provides a data generation method based on a large power grid customer service model, specifically comprising: The present invention is used to overcome the above technical difficulties to generate high-quality data that meets the needs of actual applications, reduce data acquisition costs, improve the robustness and generalization ability of the model, and thus enhance its stability and efficiency in practical applications.
[0028] First, obtain the input target text and preprocess the data set to remove duplicates, missing values, and outliers; Then, we perform multi-faceted data augmentation on the pre-processed data, including context-level (such as synonym replacement, word order randomization perturbation, etc.) and non-context-level (such as noise injection, word order randomization deletion, etc.) augmentation operations, and mix the new sentences into the original dataset. Then, a diversity constraint mechanism is designed to dynamically optimize text diversity through pre-trained language models and related indicators; Then, a deep neural network is constructed and deployed, and suitable new sentences are selected as new data based on their output similarity. Finally, the deep neural network training process is optimized and the parameters are adjusted to complete the experimental parameter update.
[0029] Specifically: Step 1: Obtain the input target text and preprocess the data set. Remove duplicate values, missing values, and data containing outliers to ensure data quality. Step 1.1: Use the duplicated() method in the data processing tool pandas to detect duplicate values in the data. The items with a true test result are duplicates. By default, the first duplicate item is retained and the subsequent duplicate items are deleted. Step 1.2: Scan each item in the data set one by one. If any missing data is found, fill it with null to facilitate subsequent data reading and generation. Step 1.3: Processing abnormal data in the dataset. Since the power grid customer service dataset only contains text data, we use re-encoding and filtering of illegal characters to handle text anomalies such as garbled text and inconsistent formatting. Step 2: Perform multi-angle data enhancement on the preliminarily processed data, including context-level enhancement operations and non-context-level enhancement operations, to achieve preliminary data generation; Step 2.1: Use context-level augmentation to rewrite the sentence by replacing synonyms and randomly perturbing the order of words in the sentence, disrupting the original order and generating a new sentence. Step 2.1.1: Synonym replacement: randomly select n non-stop words in the sentence and replace them with synonyms based on the contextual semantics; Step 2.1.2: Randomly select two words in the sentence and swap their positions. Repeat n times to get a new sentence. Step 2.1.3: Randomly select a non-stop word from the newly obtained sentence, convert it into a synonym and insert it into a random position, repeating n times, where the number of repetitions n in steps 2.1.1, 2.1.2 and 2.1.3 depends on the number of independent words in the sentence; Step 2.2: Use non-contextual enhancement operations to generate new sentences using a semantic rewriting method that uses noise injection and randomized word order deletion; Step 2.2.1: Randomly delete words in the sentence with probability p, repeating the number of times to be one tenth of the number of words in the sentence. The default probability is 0.3. If the amount of data to be generated is large, the probability can be increased appropriately, and vice versa. Step 2.2.2: For text data, two noise injection methods are used: misspelling of the same word and word replacement. Word replacement refers to randomly replacing a character with the letter of an adjacent word on the keyboard to simulate accidental keyboard presses. Step 2.3: All new sentences generated in steps 2.1 and 2.2 are mixed into the original dataset in a completely random manner to cover different semantic levels; Step 3: Design a diversity constraint mechanism to dynamically optimize the diversity of generated texts through relevant indicators; Step 3.1: Build a pre-trained language model (LM) based on the open-source ChatGLM-6B. The LM is a semantic constraint module used to represent the semantics of text that meets the requirements. Step 3.2: Set the similarity threshold and semantic coverage indicators. New sentences that meet the conditions can be retained as subsequent input data; Step 4: Build and deploy a deep neural network θ based on a generative adversarial network. Take the sentence processed by the above operation as input. The output of the deep neural network θ is the similarity between the generated new sentence and the original sentence. The similarity requirement is retained, thereby screening out new texts with high similarity as new data in the dataset. Step 5: Optimize the training process of the deep neural network θ. The loss function is cross entropy, and the gradient descent method is used to adjust and update the experimental parameters.
[0030] In yet another embodiment of the present invention, a data generation system based on a large power grid customer service model is provided, which can be used to implement the above-mentioned data generation method based on a large power grid customer service model. Specifically, the system includes: The data acquisition module is used to obtain the original sentence as a data set, preprocess the data set, and obtain the preprocessed data set; The data enhancement module is used to perform data enhancement on the preprocessed data set to obtain a data-enhanced data set; The text generation module is used to generate text sentences from the enhanced dataset through semantic constraints and indicator constraints; The filtering output module is used to take the text sentences and original sentences as input of the deep neural network, output the similarity between the text sentences and the original sentences, and filter out new sentences as new data of the dataset based on the similarity.
[0031] The module division in the embodiments of the present invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in various embodiments of the present invention may be integrated into a single processor, exist physically as separate modules, or two or more modules may be integrated into a single module. The integrated modules may be implemented in either hardware or software functional modules.
[0032] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory, wherein the memory is used to store a computer program, wherein the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of a data generation method based on a large power grid customer service model.
[0033] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device, used to store programs and data. It is understood that the computer-readable storage medium herein may include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides storage space, which stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for being loaded and executed by a processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be high-speed RAM memory or non-volatile memory, such as at least one disk storage device. The processor may load and execute the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the data generation method based on a large power grid customer service model in the above-mentioned embodiment.
[0034] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0035] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0036] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0037] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.
Claims
1. A data generation method based on a large power grid customer service model, characterized in that: include: Obtain the original sentence as a data set, preprocess the data set, and obtain a preprocessed data set; Perform data enhancement on the preprocessed data set to obtain a data-enhanced data set; Generate text sentences from the enhanced dataset through semantic constraints and indicator constraints; The text sentence and the original sentence are used as the input of the deep neural network, the similarity between the text sentence and the original sentence is output, and new sentences are filtered out as new data of the dataset based on the similarity.
2. The data generation method based on the power grid customer service large model according to claim 1 is characterized in that: The process of obtaining the original sentence as a data set and preprocessing the data set to obtain the preprocessed data set includes: Use data processing tools to detect duplicate values in the data set; scan each item in the data set one by one, and fill in null if missing; use re-encoding and filtering illegal characters to handle text anomalies such as garbled text and inconsistent formatting to obtain the preprocessed data set.
3. The data generation method based on the power grid customer service large model according to claim 1 is characterized in that: The data enhancement is performed on the preprocessed data set to obtain a data-enhanced data set, including: Using context-level augmentation operations, a semantic rewriting method based on synonym replacement is used to replace words and randomize the word order to generate new sentences. Using non-contextual enhancement operations, a semantic rewriting method of noise injection and word order randomization and deletion is used to generate new sentences; All the new sentences generated above are mixed into the original dataset in a completely random manner, covering different semantic levels, to obtain a data-enhanced dataset.
4. The data generation method based on the power grid customer service large model according to claim 3 is characterized in that: The context-level enhancement operations specifically include: Synonym replacement: randomly select n non-stop words in a sentence and replace them with synonyms based on the contextual semantics; Randomly select two words in the sentence and swap their positions, repeat n times, and get a new sentence; Randomly select a non-stop word from the newly obtained sentence, convert it into a synonym and insert it into a random position, repeating n times.
5. The data generation method based on the power grid customer service large model according to claim 3 is characterized in that: The non-contextual enhancement operations specifically include: Randomly delete words in the sentence with probability p, and repeat the number of times to be one tenth of the number of words in the sentence; For text data, two forms of noise injection are used: spelling errors of the same word and word replacement. Word replacement refers to randomly replacing a character with the letter of the adjacent word on the keyboard to simulate accidental keyboard presses.
6. The data generation method based on the power grid customer service large model according to claim 1 is characterized in that: The enhanced dataset is used to generate text sentences through semantic constraints and indicator constraints, including: Build a pre-trained language model as a semantic constraint to represent and filter the text semantics that meet the conditions; Set two indicators: similarity threshold and semantic coverage. New sentences that meet the conditions are retained as subsequent input data.
7. The data generation method based on the power grid customer service model according to claim 1 is characterized in that: The method uses the text sentence and the original sentence as inputs of the deep neural network, outputs the similarity between the text sentence and the original sentence, and selects new sentences as new data of the dataset based on the similarity, including: Build and deploy a deep neural network, taking text sentences and original sentences as input. The output of the deep neural network is the similarity between the newly generated input sentences and the original sentences. Based on the similarity, suitable new sentences are selected as new data in the dataset. Optimize the training process of the deep neural network, adjust the training parameters, use the cross entropy loss function, and use the gradient descent method to complete the adjustment and update of experimental parameters.
8. A data generation system based on a large power grid customer service model, characterized in that: include: The data acquisition module is used to obtain the original sentence as a data set, preprocess the data set, and obtain the preprocessed data set; The data enhancement module is used to perform data enhancement on the preprocessed data set to obtain a data-enhanced data set; The text generation module is used to generate text sentences from the enhanced dataset through semantic constraints and indicator constraints; The filtering output module is used to take the text sentences and original sentences as input of the deep neural network, output the similarity between the text sentences and the original sentences, and filter out new sentences as new data of the dataset based on the similarity.
9. The data generation system based on the power grid customer service model according to claim 8, characterized in that: In the data acquisition module, original sentences are acquired as a data set, and the data set is preprocessed to obtain a preprocessed data set, including: Use data processing tools to detect duplicate values in the data set; scan each item in the data set one by one, and fill in null if missing; use re-encoding and filtering illegal characters to handle text anomalies such as garbled text and inconsistent formatting to obtain the preprocessed data set.
10. The data generation system based on the power grid customer service model according to claim 8, characterized in that: In the data enhancement module, data enhancement is performed on the preprocessed data set to obtain a data-enhanced data set, including: Using context-level augmentation operations, a semantic rewriting method based on synonym replacement is used to replace words and randomize the word order to generate new sentences. Using non-contextual enhancement operations, a semantic rewriting method of noise injection and word order randomization and deletion is used to generate new sentences; All the new sentences generated above are mixed into the original dataset in a completely random manner, covering different semantic levels, to obtain a data-enhanced dataset.
11. The data generation system based on the power grid customer service big model according to claim 10, characterized in that: The context-level enhancement operations specifically include: Synonym replacement: randomly select n non-stop words in a sentence and replace them with synonyms based on the contextual semantics; Randomly select two words in the sentence and swap their positions, repeat n times, and get a new sentence; Randomly select a non-stop word from the newly obtained sentence, convert it into a synonym and insert it into a random position, repeating n times.
12. The data generation system based on the power grid customer service big model according to claim 10, characterized in that: The non-contextual enhancement operations specifically include: Randomly delete words in the sentence with probability p, and repeat the number of times to be one tenth of the number of words in the sentence; For text data, two forms of noise injection are used: spelling errors of the same word and word replacement. Word replacement refers to randomly replacing a character with the letter of the adjacent word on the keyboard to simulate accidental keyboard presses.
13. The data generation system based on the power grid customer service model according to claim 8, characterized in that: In the text generation module, the enhanced dataset is used to generate text sentences through semantic constraints and indicator constraints, including: Build a pre-trained language model as a semantic constraint to represent and filter the text semantics that meet the conditions; Set two indicators: similarity threshold and semantic coverage. New sentences that meet the conditions are retained as subsequent input data.
14. The data generation method based on the power grid customer service large model according to claim 8, characterized in that: In the screening output module, the text sentence and the original sentence are used as inputs of the deep neural network, the similarity between the text sentence and the original sentence is output, and new sentences are screened out as new data of the data set based on the similarity, including: Build and deploy a deep neural network, taking text sentences and original sentences as input. The output of the deep neural network is the similarity between the newly generated input sentences and the original sentences. Based on the similarity, suitable new sentences are selected as new data in the dataset. Optimize the training process of the deep neural network, adjust the training parameters, use the cross entropy loss function, and use the gradient descent method to complete the adjustment and update of experimental parameters.
15. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the data generation method based on the power grid customer service model as described in any one of claims 1 to 7 are implemented.
16. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the data generation method based on the power grid customer service large model as described in any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Data enhancement method and device, equipment and storage medium
CN116383644A
Text enhancement quality evaluation method, electronic equipment and storage medium
CN116629238A
Text enhancement method, electronic equipment and computer readable storage medium
CN116860955A
Open domain dialog reply method and system based on thematic enhancement
US20240062006A1
