Steel quality appeal content classification method and device, electronic equipment and medium

By constructing a classification model for quality appeal data set and text data augmentation mode training, the problem of poor classification effect in online steel quality appeal handling is solved, and fast and accurate classification and response are achieved, which improves processing efficiency.

CN119025676APending Publication Date: 2024-11-26欧冶云商股份有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411011124.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

In the online steel quality complaint handling, it is difficult for the existing technology to achieve real-time classification and processing, resulting in waiting for manual review and poor classification effect.

Method used

Build a quality appeal data set, train the target quality appeal classification model based on text data augmentation mode, determine the appeal category through the received data and models, and improve classification efficiency and accuracy.

Benefits of technology

It realizes fast and accurate classification of quality appeal data, improves processing efficiency, and enables the steel quality objection appeal platform to respond to consumer needs faster.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025676A_ABST
    Figure CN119025676A_ABST
Patent Text Reader

Abstract

The invention provides a steel quality appeal content classification method and device, electronic equipment and a medium, and relates to the technical field of steel industry information. The steel quality appeal content classification method comprises the following steps: constructing a quality appeal data set; based on the quality appeal data set and a preset text data enhancement mode, training to obtain a target quality appeal classification model; and determining a target appeal category corresponding to the target quality appeal data according to the received target quality appeal data, the target quality appeal classification model and preset appeal classification information. By using the target quality appeal classification model, the received quality appeal data can be quickly classified into the corresponding appeal category, the classification effect of the steel quality appeal content is integrally improved, the processing efficiency is also improved, the steel quality objection appeal platform can more quickly respond to the appeal of a steel consumer, and the quality of the steel consumer is improved. And effective measures are taken to solve problems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of information technology for the steel industry, and for example, to a method, device, electronic equipment and medium for classifying steel quality complaints. Background Art

[0002] With the development of science and technology, steel trading has gradually moved from the traditional offline market to a convenient online platform. Each online platform has also established an online customer service system to handle related issues of steel trading after sales. Steel quality objection is a type of after-sales problem. It refers to product quality complaints made by users when they find that steel products do not meet the relevant technical standards, quality requirements in contracts and order agreements, or cannot meet user needs, and the problem is caused by the company.

[0003] In the related technology, when handling complaints about steel quality online, a dedicated person is required to preliminarily classify the complaints and then transfer them to the sub-classification personnel. Although this process is more convenient than the traditional offline processing process, it still needs to wait for manual review and circulation, and cannot handle the complaints online in real time. Some online platforms will pre-set some categories of complaint content, but there are many categories corresponding to the quality objection complaint texts in the steel e-commerce field, resulting in poor classification of steel quality complaints. Summary of the invention

[0004] The present application aims to provide a method, device, electronic equipment and medium for classifying steel quality complaint content.

[0005] According to one aspect of the present application, a method for classifying steel quality complaint content is proposed, including: constructing a quality complaint data set; training a target quality complaint classification model based on the quality complaint data set and a preset text data enhancement mode; determining a target complaint category corresponding to the target quality complaint data based on the received target quality complaint data, the target quality complaint classification model and the preset complaint classification information.

[0006] According to some embodiments, constructing a quality complaint data set includes: analyzing multiple groups of public quality complaint data within a preset historical period to determine quality complaint issues and complaint data features corresponding to the quality complaint issues; setting optional complaint categories based on quality complaint issues, complaint data features and a preset category setting mode; and constructing a quality complaint data set based on multiple groups of public quality complaint data and corresponding optional complaint categories.

[0007] According to some embodiments, multiple groups of public quality complaint data within a preset historical period are analyzed to determine quality complaint issues and complaint data features corresponding to the quality complaint issues, including: for each group of public quality complaint data in the multiple groups of public quality complaint data within the preset historical period, each group of public quality complaint data is analyzed based on a preset character matching pattern to determine whether each group of public quality complaint data has invalid typical features; in the case of invalid typical features, the public quality complaint data corresponding to the invalid typical features are determined as invalid description categories, and the quality complaint issues and complaint data features corresponding to the quality complaint issues are determined based on valid data in the multiple groups of public quality complaint data; wherein the valid data are the multiple groups of public quality complaint data after the public quality complaint data corresponding to the invalid description categories are eliminated; in the case of no invalid typical features, the quality complaint issues and complaint data features corresponding to the quality complaint issues are determined based on the multiple groups of public quality complaint data.

[0008] According to some embodiments, a target quality complaint classification model is trained based on a quality complaint data set and a preset text data enhancement mode, including: dividing the quality complaint data set according to a preset division ratio to determine a quality complaint training set and a quality complaint test set; performing data enhancement on the quality complaint training set using a preset text data enhancement mode; and training a preset quality complaint classification model based on the enhanced quality complaint training set and quality complaint test set to generate a target quality complaint classification model.

[0009] According to some embodiments, the quality complaint data set is divided according to a preset division ratio to determine a quality complaint training set and a quality complaint test set, including: analyzing the quality complaint data set to determine the proportion of sample data corresponding to each optional complaint category in the quality complaint data set; determining a preset division ratio according to the proportion of sample data corresponding to each optional complaint category in the quality complaint data set; and randomly sampling the quality complaint data set according to the preset division ratio and a preset stratified sampling pattern to divide the sample data of each optional complaint category into the quality complaint training set and the quality complaint test set according to the corresponding division ratio.

[0010] According to some embodiments, the quality complaint data set is divided according to a preset division ratio to determine a quality complaint training set and a quality complaint test set, and also includes: randomizing the sample data of each optional complaint category so that the probability of sample data of each optional complaint category being extracted is the same.

[0011] According to some embodiments, a preset text data enhancement mode is used to perform data enhancement on a quality complaint training set, including: performing word segmentation processing on each piece of quality complaint data in the quality complaint training set according to a preset word segmentation mode to determine the split vocabulary corresponding to each piece of quality complaint data; obtaining synonymous vocabulary corresponding to the split vocabulary according to a preset vocabulary similarity; and randomly inserting synonymous vocabulary into each piece of quality complaint data to achieve data enhancement of the quality complaint training set.

[0012] According to some embodiments, a preset quality appeal classification model is trained based on an enhanced quality appeal training set and a quality appeal test set to generate a target quality appeal classification model, including: configuring physical parameters of the preset quality appeal classification model; inputting the enhanced quality appeal training set into the preset quality appeal classification model after the physical parameter configuration for training to determine a quality appeal classification fine-tuning model; inputting the quality appeal test set into the quality appeal classification fine-tuning model for training to generate iterative test results; adjusting the hyperparameters of the quality appeal classification fine-tuning model according to the iterative test results, and determining the target hyperparameters based on the iterative test results of the quality appeal classification fine-tuning model after the hyperparameters are adjusted; generating the target quality appeal classification model based on the target hyperparameters and the quality appeal classification fine-tuning model after the hyperparameters are adjusted.

[0013] According to some embodiments, a target complaint category corresponding to the target quality complaint data is determined based on the received target quality complaint data, a target quality complaint classification model and preset complaint classification information, including: matching the received target quality complaint data with the preset complaint classification information; in a case where historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, determining the historical complaint category corresponding to the historical quality complaint data as the target complaint category; in a case where historical quality complaint data identical to the target quality complaint data is not matched in the preset complaint classification information, inputting the target quality complaint data into the target quality complaint classification model to output the target complaint category.

[0014] According to one aspect of the present application, a device for classifying steel quality complaint content is proposed, including: a data set construction module, used to construct a quality complaint data set; a model training module, used to train a target quality complaint classification model based on the quality complaint data set and a preset text data enhancement mode; a category determination module, used to determine the target complaint category corresponding to the target quality complaint data based on the received target quality complaint data, the target quality complaint classification model and the preset complaint classification information.

[0015] According to one aspect of the present application, an electronic device is provided, which includes: a processor; and a memory storing a computer program, wherein when the computer program is executed by the processor, the processor executes the method for classifying steel quality complaint content as described above.

[0016] According to one aspect of the present application, a non-transitory computer-readable medium is provided, on which readable instructions are stored. When the instructions are executed by a processor, the processor executes the method for classifying steel quality complaint content as described above.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the present application.

[0018] Beneficial effects:

[0019] Through the above-mentioned embodiments provided by this application, a quality complaint data set is constructed, and historical and real-time quality complaint data can be systematically collected, sorted and analyzed, ensuring the authenticity and reliability of the data. The robustness of the target quality complaint classification model training is improved through the preset text data enhancement mode, and it is ensured that it has good performance on diversified data. In practical applications, by using the target quality complaint classification model, the received quality complaint data can be quickly classified into the corresponding complaint category, which improves the overall classification effect of the steel quality complaint content and improves the processing efficiency, so that the steel quality objection complaint platform can respond to the complaints of steel consumers more quickly and take effective measures to solve the problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without exceeding the scope of protection required by the present application.

[0021] Figure 1 A flow chart of a method for classifying steel quality complaints provided in an embodiment of the present application;

[0022] Figure 2 A flowchart of constructing a quality complaint data set in step S10 provided in an embodiment of the present application;

[0023] Figure 3 A flowchart for analyzing multiple groups of public quality complaint data within a preset historical period in step S100 to determine quality complaint issues and complaint data features corresponding to the quality complaint issues provided in an embodiment of the present application;

[0024] Figure 4A flowchart of training a target quality complaint classification model based on a quality complaint dataset and a preset text data enhancement mode in step S11 provided in an embodiment of the present application;

[0025] Figure 5 A flowchart of dividing the quality complaint data set according to a preset division ratio in step S110 to determine a quality complaint training set and a quality complaint test set provided in an embodiment of the present application;

[0026] Figure 6 Another flowchart of the embodiment of the present application, in step S110, dividing the quality complaint data set according to a preset division ratio to determine a quality complaint training set and a quality complaint test set;

[0027] Figure 7 A flowchart of performing data enhancement on a quality complaint training set using a preset text data enhancement mode in step S111 provided in an embodiment of the present application;

[0028] Figure 8 A flowchart of training a preset quality complaint classification model based on the enhanced quality complaint training set and the quality complaint test set in step S112 to generate a target quality complaint classification model is provided for an embodiment of the present application;

[0029] Fig. 9 A flowchart of determining a target complaint category corresponding to the target quality complaint data according to the received target quality complaint data, the target quality complaint classification model and the preset complaint classification information in step S12 provided in an embodiment of the present application;

[0030] Fig.10 A block diagram of a device for classifying steel quality complaint content provided in an embodiment of the present application;

[0031] Fig.11 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this application will be comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0033] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the present application.

[0034] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0035] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0036] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another component. Therefore, the first component discussed below can be referred to as the second component without departing from the teachings of the concepts of the present application. As used herein, the term "and / or" includes any one of the associated listed items and all combinations of one or more.

[0037] For specific implementation methods, please refer to the following embodiments.

[0038] Figure 1 This is a flow chart of a method for classifying steel quality complaints provided in an embodiment of the present application. The method of this embodiment can be applied to a complaint classification server. Figure 1 As shown, the method includes: step S10, step S11 and step S12.

[0039] In step S10, a quality complaint dataset is constructed.

[0040] The quality complaint dataset can be obtained through data publicly available on various online platforms or data involved in offline steel complaints. It can include real quality objection complaint data historically released by steel consumers, such as text, as well as complaint categories determined by historical analysis and data features of the complaint categories.

[0041] According to an example embodiment, real quality objection complaint data published by steel consumers can be collected on various network platforms, such as steel e-commerce platforms. In some implementations, quality objection complaint data can be extracted from comments, which can express the negative evaluation of steel consumers on the size, specifications, appearance, performance, etc. of steel products. Then, based on the common problems and data features in these quality objection complaint data, a quality complaint dataset can be constructed accordingly.

[0042] In step S11, a target quality complaint classification model is trained based on the quality complaint dataset and a preset text data enhancement mode.

[0043] Text data augmentation is a technique that aims to increase the diversity and size of a dataset by modifying raw text data, thereby improving the generalization ability of machine learning models. Preset text data augmentation modes can include synonym replacement, random word insertion, random word deletion, back translation (i.e. translating text into another language and then translating it back to the original language), sentence reorganization, etc.

[0044] The goal of the target quality complaint classification model can be to input new quality complaint data into the model and directly output the corresponding complaint category.

[0045] According to the example embodiment, a suitable enhancement method can be selected in a preset text data enhancement mode according to the data characteristics of the quality complaint data set and data enhancement can be performed. The quality complaint data set after data enhancement is split into a training set and a test set for training to obtain a target quality complaint classification model.

[0046] In step S12, the target complaint category corresponding to the target quality complaint data is determined according to the received target quality complaint data, the target quality complaint classification model and the preset complaint classification information.

[0047] Steel consumers can log in to the complaint platform or system and input the content of the steel quality objection complaint, and the input content can be used as the target quality complaint data. The preset complaint classification information can be used to represent a series of pre-defined complaint categories and related information, which can include the name, description, keywords, etc. of the complaint category.

[0048] According to the example embodiment, the received target quality complaint data can be compared with the preset complaint classification information to exclude abnormal complaint data, and then the target quality complaint data after excluding abnormalities is input into the target quality complaint classification model to directly output the target complaint category corresponding to the target quality complaint data.

[0049] By constructing a quality complaint data set, this application can systematically collect, organize and analyze historical and real-time quality complaint data, ensuring the authenticity and reliability of the data. It also improves the robustness of the target quality complaint classification model training and ensures its good performance on diversified data through the preset text data enhancement mode. In practical applications, by using the target quality complaint classification model, the received quality complaint data can be quickly classified into the corresponding complaint category, which improves the overall classification effect of steel quality complaint content and improves processing efficiency, so that the steel quality objection complaint platform can respond to steel consumers' complaints more quickly and take effective measures to solve problems.

[0050] According to some embodiments, reference Figure 2 In step S10, a quality complaint data set is constructed, which can be specifically implemented through steps S100, S101 and S102.

[0051] In step S100, multiple groups of public quality complaint data within a preset historical period are analyzed to determine quality complaint issues and complaint data features corresponding to the quality complaint issues.

[0052] The preset historical period can be a pre-set period for selecting data required by the quality complaint data set. In some implementations, data within one year before the current moment can be selected. Public quality complaint data can be used to represent quality complaint data disclosed on online platforms or disclosed in offline transactions. These data can be complaint information about product quality issues released by steel consumers.

[0053] Quality complaint issues can be used to indicate product quality issues that appear in complaint data. The content of complaint data can include descriptions of quality such as 1111, periodic friction black spots at the same position of the roll, oil-free and chromium-free passivated products purchased, but actually oil-coated products, and a large number of scratches and wrinkles on the surface. Correspondingly, quality complaint issues can include garbled code issues, quality issues, price issues, etc.

[0054] Complaint data features can be data characteristics or attributes associated with each quality complaint issue. These features may include keywords in the complaint text, sentiment (such as positive, negative or neutral), and the number of complaints, such as invalid descriptions, friction spots, inconsistent surface treatment, price issues, and surface quality issues.

[0055] According to the example embodiment, natural language processing technology can be used to perform operations such as word segmentation, part-of-speech tagging, and named entity recognition on multiple groups of public quality complaint data within a preset period to extract keywords and phrases. Then, the main quality complaint issues are identified, and for each quality complaint issue, the complaint data features related to it are summarized.

[0056] In S101, optional complaint categories are set based on the quality complaint problem, complaint data characteristics and a preset category setting mode.

[0057] The preset category setting mode can be used to indicate the principles, methods or rules followed when setting the complaint category. This can be determined by experts in the steel quality industry based on business needs, industry standards, habits of steel consumers, etc. Optional complaint categories can be used to indicate the category labels of the complaint characteristics corresponding to the quality complaint issues.

[0058] According to the example embodiment, for several quality complaint issues, the complaint quantity characteristics corresponding to each quality complaint issue are used to set the optional complaint categories of the quality complaint issues according to the preset category setting mode. Since some complaint data characteristics corresponding to some quality complaint issues may not be representative, or there are problems with the same meaning, these complaint data characteristics cannot be directly determined as the corresponding optional complaint categories. Therefore, they can be set according to the preset category setting mode.

[0059] In S102, a quality complaint data set is constructed according to multiple groups of public quality complaint data and corresponding optional complaint categories.

[0060] According to an example embodiment, multiple groups of public quality complaint data within a preset historical period and the optional complaint categories corresponding to each quality complaint issue contained therein are stored correspondingly to construct a quality complaint data set.

[0061] Through data analysis, this application can quickly and accurately identify the main quality complaints and provide strong support for subsequent improvement measures. By responding to and resolving steel consumers' quality complaints in a timely manner, the satisfaction of steel consumers can be improved.

[0062] According to some embodiments, reference Figure 3 In step S100, multiple groups of public quality complaint data within a preset historical period are analyzed to determine the quality complaint issues and complaint data characteristics corresponding to the quality complaint issues. This can be specifically achieved through steps S1000, S1001 and S1002.

[0063] In step S1000, for each set of public quality complaint data in a plurality of sets of public quality complaint data within a preset historical period, each set of public quality complaint data is analyzed based on a preset character matching pattern to determine whether each set of public quality complaint data has invalid typical features.

[0064] Each of the multiple sets of public quality complaint data may include a steel quality problem published by a steel consumer and a corresponding description.

[0065] The preset character matching mode may be a pre-set method for character matching, which may include regular expressions, string checks, character sequence comparisons, etc. Invalid typical features may be used to indicate invalid content of each set of public quality complaint data during the steel quality objection complaint process, such as all non-objectionable characters such as numbers and symbols.

[0066] According to the example embodiment, the public quality complaint data can be divided into multiple groups of public quality complaint data according to the release time, release source, etc. of the public quality complaint data within a preset historical period. A preset character matching mode is used to compare with each group of public quality complaint data. If the comparison is successful, it means that the group of public quality complaint data has invalid typical features, and if the comparison is unsuccessful, it means that the group of public quality complaint data does not have invalid typical features.

[0067] In step S1001, in the presence of invalid typical features, the public quality complaint data corresponding to the invalid typical features are determined as invalid description categories, and the quality complaint issues and complaint data features corresponding to the quality complaint issues are determined based on valid data in multiple groups of public quality complaint data; wherein the valid data are multiple groups of public quality complaint data after excluding each group of public quality complaint data corresponding to the invalid description categories.

[0068] The invalid description category can be a special category. Since the invalid typical characteristics are irrelevant to the specific situation of steel quality, the corresponding public quality complaint data cannot be used as complaint data features, and the public quality complaint data of this category needs to be eliminated. After eliminating the public quality complaint data corresponding to the invalid description category, the remaining public quality complaint data can be used as public quality complaint data.

[0069] According to an exemplary embodiment, in the case of invalid typical features, each corresponding set of public quality complaint data is determined as an invalid description category. Then, based on valid data, the quality complaint problem and the complaint data features corresponding to the quality complaint problem can be determined.

[0070] In step S1002, in the absence of invalid typical features, quality complaint issues and complaint data features corresponding to the quality complaint issues are determined based on multiple groups of public quality complaint data.

[0071] According to an example embodiment, in the absence of invalid typical features, quality appeal issues and appeal data features corresponding to the quality appeal issues may be determined directly based on multiple groups of public quality appeal data in the above manner.

[0072] This application uses a preset character matching pattern to identify and eliminate public quality complaint data containing invalid typical features, effectively reducing the noise and interference items in the data, thereby improving the data quality relied upon by subsequent analysis. This data cleaning process also ensures the accuracy and reliability of the analysis results. Excluding data in invalid description categories in advance allows subsequent analysis of quality complaint issues and complaint data characteristics to focus more on valuable data. This not only reduces unnecessary calculations, but also speeds up the analysis process and improves overall work efficiency. Identifying quality complaint issues based on cleaned valid data can more accurately reflect the quality issues truly reported by steel consumers.

[0073] According to some embodiments, reference Figure 4 In step S11, based on the quality complaint dataset and the preset text data enhancement mode, a target quality complaint classification model is trained, which can be specifically implemented through steps S110, S111 and S112.

[0074] In step S110, the quality complaint data set is divided according to a preset division ratio to determine a quality complaint training set and a quality complaint test set.

[0075] The preset division ratio may be a preset division method of the training set and the test set, and in some implementations, 8:2 may be selected.

[0076] According to an example embodiment, the quality complaint data set may be divided randomly according to a preset division ratio or divided according to certain requirements into a quality complaint training set and a quality complaint test set.

[0077] In step S111, the quality complaint training set is enhanced using a preset text data enhancement mode.

[0078] The preset text data enhancement mode can be a process of generating new training samples by performing a series of transformations, expansions or improvements on the original text data. These enhancement methods can increase the diversity and scale of the training data set, thereby improving the robustness and generalization ability of the model. For quality complaint data, the text data enhancement mode can include strategies such as synonym replacement, random insertion, random deletion, random exchange, and back translation. In some implementations, a random insertion text data enhancement mode can be used.

[0079] According to an example embodiment, data enhancement may be achieved by randomly inserting selected synonyms into sentences with a small number of characters in a quality complaint training set in a random insertion manner.

[0080] In step S112, the preset quality complaint classification model is trained based on the enhanced quality complaint training set and the quality complaint test set to generate a target quality complaint classification model.

[0081] According to the example embodiment, the quality complaint training set processed by the text data enhancement mode is loaded into the memory for model training. The quality complaint test set is loaded for subsequent evaluation of model performance. The preset quality complaint classification model that has been initially constructed is loaded, including the model structure, initial parameters, etc. According to the task requirements, appropriate evaluation indicators are defined, such as accuracy, recall rate, F1 score, etc., for evaluating model performance.

[0082] Preprocessing operations can be performed on the training set and test set, such as word segmentation, stop word removal, vectorization, etc. The model is trained using the enhanced training set. During the training process, techniques such as cross-validation and early stopping can be used to prevent overfitting and monitor performance indicators during the training process. The parameters of the model are iteratively adjusted based on the performance indicator feedback during the training process. The adjusted parameters may include learning rate, number of iterations, batch size, regularization term strength, etc. The goal of parameter adjustment is to find a set of optimal parameter combinations so that the model performs best on the test set. After each parameter adjustment, the model is evaluated using the test set to verify the effect of the parameter adjustment. The evaluation results will serve as a reference for the next parameter adjustment.

[0083] This application uses a preset text data enhancement mode to perform data enhancement on the quality complaint training set, effectively increasing the diversity and quantity of training samples. This enhancement can help the model learn richer language patterns and contextual information. This rich and diverse training data enables the model to maintain a high classification accuracy when facing unknown or slightly changing target quality complaint data, significantly improving the generalization ability of the model.

[0084] According to some embodiments, reference Figure 5 In step S110, the quality complaint data set is divided according to a preset division ratio to determine a quality complaint training set and a quality complaint test set, which can be specifically implemented through steps S1100, S1101 and S1102.

[0085] In step S1100, the quality complaint data set is analyzed to determine the proportion of sample data corresponding to each optional complaint category in the quality complaint data set.

[0086] According to an example embodiment, the quality complaint dataset is analyzed to determine the number of sample data corresponding to each optional complaint category, and then the corresponding proportion is calculated based on the number of sample data of each optional complaint category and the total amount of sample data in the quality complaint dataset.

[0087] In step S1101, a preset division ratio is determined according to the proportion of sample data corresponding to each optional complaint category in the quality complaint data set.

[0088] According to the example embodiment, a reasonable division ratio is determined based on the proportion of each optional complaint category, combined with actual conditions and needs. This ratio can ensure that the number of complaints in each category in the training set and the test set is sufficient so that the model can fully learn the characteristics of each category. For example, if the proportion of a certain optional complaint category is very high (such as steel quality issues account for more than 50%), then when dividing the data set, it may be necessary to appropriately increase the proportion of samples in this category in the test set to ensure that the performance of the model in this category can also be fully evaluated.

[0089] In step S1102, the quality complaint data set is randomly sampled according to a preset division ratio and a preset stratified sampling mode, so as to divide the sample data of each optional complaint category into a quality complaint training set and a quality complaint test set according to the corresponding division ratio.

[0090] Stratified sampling is a special sampling method that first divides the population into several sub-populations (or layers), and then performs random sampling independently from each layer. In this application, each layer can correspond to an optional complaint category, and the preset stratified sampling mode is a method of performing random sampling independently in each optional complaint category. Stratified sampling ensures that the sample ratio of each optional complaint category in the training set and test set is consistent with the ratio in the original data set.

[0091] According to an example embodiment, the original data set is divided into multiple layers according to the proportion of complaint categories, and random sampling is performed within each layer according to the corresponding preset division ratio (such as 80% training and 20% testing). Ensure that the sample ratio of each complaint category in the training set and the test set is consistent with the ratio in the original data set. Among them, a random number generator or a random sampling function in a programming library can be used to independently perform random sampling on the quality complaint data set corresponding to each optional complaint category. The sample data obtained by sampling are respectively classified into the quality complaint training set and the quality complaint test set.

[0092] This application analyzes the sample proportion of each complaint category in the quality complaint dataset to ensure that when the training set and test set are subsequently divided, the proportion of these categories in the two datasets can be kept consistent with the original dataset. Through the stratified sampling model, samples of each complaint category are fully represented in the training set and the test set. This means that the model can learn the characteristics of each category during the training process, so that it can more accurately identify and classify unknown samples during testing. Since the sample distribution in the test set is consistent with the original dataset, using the test set to evaluate the model performance can more accurately reflect the performance of the model in actual applications. This accurate evaluation helps developers to promptly identify problems with the model and make improvements, thereby ensuring that the final deployed model has a high classification accuracy.

[0093] According to some embodiments, reference Figure 6 In step S110, the quality complaint data set is divided according to a preset division ratio to determine a quality complaint training set and a quality complaint test set. This can also be achieved by combining step S1100, step S1101, step S1102 with step S1103.

[0094] In step S1103, random processing is performed on the sample data of each optional appeal category so that the probability of sampling of the sample data of each optional appeal category is the same.

[0095] According to an example embodiment, for each optional complaint category in the data set, a random number generator or a randomization function in a programming library is used to disrupt the original order of samples within the category. This can be achieved by randomly sorting the indexes of the samples, or by directly randomly rearranging the samples themselves. It can be ensured that the randomization process is independent, that is, sample data of different categories will not affect each other. The randomization process of each category is performed based on the samples within the category.

[0096] In some implementations, the effect of the randomization process can be verified by statistical methods (such as calculating the distance or correlation between samples), or visual means (such as scatter plots or histograms) can be used to visually check whether the data distribution after randomization is uniform.

[0097] According to some embodiments, reference Figure 7 In step S111, the quality complaint training set is enhanced using a preset text data enhancement mode, which can be specifically achieved through steps S1110, S1111 and S1112.

[0098] In step S1110, each piece of quality complaint data in the quality complaint training set is segmented according to a preset segmentation mode to determine the split vocabulary corresponding to each piece of quality complaint data.

[0099] The preset segmentation mode can be set according to the actual situation of steel quality.

[0100] According to the embodiment, in each piece of quality complaint data in the quality complaint training set, a continuous sequence of characters is cut into single or multiple words according to a preset word segmentation pattern, and the obtained words can be used as corresponding split words.

[0101] In step S1111, synonymous words corresponding to the split words are obtained according to the preset word similarity.

[0102] The preset word similarity can be used to indicate the degree to which two words are semantically close or similar. The preset word similarity can be calculated based on a certain metric or algorithm, such as cosine similarity, Jaccard similarity, edit distance, etc., and combined with specific word resources (such as dictionaries, word vector models). This preset value can be a fixed threshold value used to determine whether two words are similar enough to be considered as synonyms.

[0103] According to an example embodiment, the split words may be matched with a preset word library to determine the words with the highest word similarity as synonyms.

[0104] In step S1112, synonymous words are randomly inserted into each piece of quality complaint data to achieve data enhancement of the quality complaint training set.

[0105] According to an example embodiment, for each piece of quality complaint data in the training set, its text content is traversed. For each word in the text, it is checked whether the word has a corresponding synonymous word set in the synonym dictionary. If so, a synonymous word in the set is randomly selected as a candidate replacement word and inserted into the piece of quality complaint data.

[0106] The data augmentation in this application significantly increases the size and diversity of the training set by introducing new variant data that maintains the original meaning of the complaint. This enables the trained model to better cope with various situations that may arise in practical applications and improves the generalization ability of the model. In the process of data augmentation, the random insertion of synonymous words not only changes the specific expression of the text, but also forces the model to learn the semantic associations and equivalence between words. This training method makes the model insensitive to slight changes in vocabulary and enhances the robustness of the model.

[0107] According to some embodiments, reference Figure 8 In step S112, the preset quality complaint classification model is trained based on the enhanced quality complaint training set and the quality complaint test set to generate a target quality complaint classification model, which can be specifically achieved through steps S1120, S1121, S1122 and S1123.

[0108] In step S1120, physical parameters of a preset quality appeal classification model are configured; the enhanced quality appeal training set is input into the preset quality appeal classification model after the physical parameter configuration for training to determine a quality appeal classification fine-tuning model.

[0109] Physical parameters can characterize specific numerical settings that affect the model training process and performance, including but not limited to the learning rate (a hyperparameter that controls the step size of parameter updates), batch size (the number of samples used to calculate gradients in each iteration), number of iterations (the number of times the training process is repeated), optimization algorithm, etc.

[0110] According to the example embodiment, the physical parameters of the preset quality complaint classification model can be set according to the task requirements. For example, the learning rate is set to 0.001, the batch size is set according to the hardware resources, and the number of iterations is determined based on the number of samples in the training set. The enhanced quality complaint training set is input into the preset quality complaint classification model with the configured physical parameters to generate a quality complaint classification fine-tuning model to calculate the corresponding prediction results.

[0111] In step S1121, the quality appeal test set is input into the quality appeal classification fine-tuning model for training to generate iterative test results.

[0112] The iterative test results can be a collection of results of the quality complaint classification fine-tuning model performing classification prediction on samples in the quality complaint test set during multiple iterative tests. Through iterative testing, the performance changes of the model at different stages and the generalization ability of the model on the test set data can be evaluated.

[0113] According to the exemplary embodiment, the quality appeal test set is input into the quality appeal classification fine-tuning model, and the prediction result obtained in step S1120 is calculated with the actual corresponding real result in the quality appeal test set to determine the performance index, such as accuracy, precision, recall, etc. Then, iterative testing can be performed under conditions such as simulating data in different time periods and test sets with different data volumes, so that the quality appeal classification fine-tuning model can meet the expected requirements and generate iterative test results.

[0114] In step S1122, the hyperparameters of the quality complaint classification fine-tuning model are adjusted according to the iterative test results, and the target hyperparameters are determined based on the iterative test results of the quality complaint classification fine-tuning model after the hyperparameter adjustment.

[0115] Hyperparameters can be used to characterize the parameters that need to be set before training the quality complaint classification fine-tuning model. They can determine the model training process and final effect, but are different from the parameters obtained through learning within the model (such as weights and biases). They can include learning rate, regularization coefficient, etc.

[0116] According to the example embodiment, various hyperparameter adjustment strategies can be preset, such as grid search, random search, and Bayesian optimization, etc. After the hyperparameter adjustment strategy is selected, it can be adjusted one by one or at the same time. After each adjustment, the performance of the model can be re-evaluated using the same quality complaint test set and the results can be recorded.

[0117] Repeat the adjustment process until the model performance no longer improves significantly or reaches the pre-set number of iterations. Then, based on the results of multiple iterations, the hyperparameter configuration with the best performance can be selected as the target hyperparameter.

[0118] In step S1123, a target quality complaint classification model is generated based on the target hyperparameters and the quality complaint classification fine-tuning model after the hyperparameters are adjusted.

[0119] According to an example embodiment, the quality complaint classification fine-tuning model can be reinitialized using the target hyperparameters, and the training set can be input into the model to start the training process. During the training process, the model can automatically adjust its internal parameters (such as weights and biases) according to the target hyperparameters and training data to minimize the loss function and improve the classification performance. After sufficient training, the final model is saved as the target quality complaint classification model.

[0120] In some implementations, after training is completed, an independent validation set can be used to perform a final performance evaluation of the model. The validation set should be independent of the training set to ensure the objectivity and accuracy of the evaluation results.

[0121] This application further improves the performance of the model by continuously optimizing the internal parameters and configuration of the model through iterative testing and hyperparameter adjustment. The model is trained and evaluated using an independent test set to ensure that the model can also perform well on unseen data. This independent test set evaluation method helps to identify and correct possible overfitting problems in the model, thereby enhancing the generalization ability of the model.

[0122] According to some embodiments, reference Fig. 9 In step S12, the target complaint category corresponding to the target quality complaint data is determined according to the received target quality complaint data, the target quality complaint classification model and the preset complaint classification information, which can be specifically implemented through steps S120 and S121.

[0123] In step S120, the received target quality complaint data is matched with preset complaint classification information. When historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, the historical complaint category corresponding to the historical quality complaint data is determined as the target complaint category.

[0124] The preset complaint classification information may be a predefined quality complaint classification standard or complaint category information that already exists in the historical data set. The preset complaint classification information may include historical quality complaint data and corresponding historical complaint categories, which may be complaint categories that often appear in the process of handling steel quality objection complaints. These historical quality complaint data are verified, so the direct matching result has a higher accuracy.

[0125] According to an example embodiment, the received target quality complaint data is matched with preset complaint classification information. If historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, the corresponding historical complaint category can be directly determined as the target complaint category.

[0126] In step S121, when no historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, the target quality complaint data is input into a target quality complaint classification model to output a target complaint category.

[0127] According to an example embodiment, if no historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, the target quality complaint data may be input into a target quality complaint classification model to directly output a target complaint category.

[0128] In this embodiment, when there is historical data that is identical or highly similar to the target quality complaint data in the preset complaint classification information, the target complaint category can be quickly determined by direct matching without going through a complex model calculation process, thereby improving classification efficiency. For those target quality complaint data that cannot find a match in the preset complaint classification information, by inputting it into the target quality complaint classification model for prediction, the generalization ability of the machine learning model can be fully utilized to handle newly emerging complaint categories or variants. It can ensure that even in the face of unknown or rare quality complaints, reasonable classification results can be given.

[0129] The following describes an apparatus embodiment of the present application, which can be used to execute the method embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, reference can be made to the method embodiment of the present application.

[0130] Fig.10 This is a block diagram of a device for classifying steel quality complaints provided in an embodiment of the present application. Fig.10 As shown, the device 1000 for classifying steel quality complaint content includes a data set construction module 1001 , a model training module 1002 and a category determination module 1003 .

[0131] The model training module 1002 trains a target quality complaint classification model based on the quality complaint data set constructed by the data set construction module 1001 and the preset text data enhancement mode. The category determination module 1003 determines the target complaint category corresponding to the target quality complaint data according to the received target quality complaint data, the target quality complaint classification model and the preset complaint classification information.

[0132] Optionally, the data set building block 1001 can specifically analyze multiple sets of public quality complaint data within a preset historical period to determine the quality complaint problem and the complaint data features corresponding to the quality complaint problem. Then, based on the quality complaint problem, the complaint data features and the preset category setting mode, the optional complaint category is set. According to the multiple sets of public quality complaint data and the corresponding optional complaint categories, a quality complaint data set is constructed.

[0133] Optionally, when analyzing multiple groups of public quality complaint data within a preset historical period to determine quality complaint issues and complaint data features corresponding to the quality complaint issues, the data set building block 1001 can specifically: for each group of public quality complaint data in the multiple groups of public quality complaint data within the preset historical period, analyze each group of public quality complaint data based on a preset character matching pattern to determine whether each group of public quality complaint data has invalid typical features; in the case of invalid typical features, determine the public quality complaint data corresponding to the invalid typical features as invalid description categories, and determine the quality complaint issues and complaint data features corresponding to the quality complaint issues based on valid data in the multiple groups of public quality complaint data. Among them, the valid data are multiple groups of public quality complaint data after eliminating each group of public quality complaint data corresponding to the invalid description category. In the case of no valid typical features, determine the quality complaint issues and complaint data features corresponding to the quality complaint issues based on multiple groups of public quality complaint data.

[0134] Optionally, the model training module 1002 can divide the quality complaint data set according to a preset division ratio to determine a quality complaint training set and a quality complaint test set. The quality complaint training set is data enhanced using a preset text data enhancement mode. Based on the enhanced quality complaint training set and quality complaint test set, a preset quality complaint classification model is trained to generate a target quality complaint classification model.

[0135] Optionally, when the model training module 1002 divides the quality complaint data set according to a preset division ratio to determine the quality complaint training set and the quality complaint test set, the model training module 1002 can specifically: analyze the quality complaint data set to determine the proportion of sample data corresponding to each optional complaint category in the quality complaint data set. Determine the preset division ratio based on the proportion of sample data corresponding to each optional complaint category in the quality complaint data set. Randomly sample the quality complaint data set based on the preset division ratio and the preset stratified sampling mode to divide the sample data of each optional complaint category into the quality complaint training set and the quality complaint test set according to the corresponding division ratio.

[0136] Optionally, when the model training module 1002 divides the quality complaint data set according to a preset division ratio to determine a quality complaint training set and a quality complaint test set, it can also specifically: randomize the sample data of each optional complaint category so that the probability of sample data of each optional complaint category being extracted is the same.

[0137] Optionally, when the model training module 1002 uses a preset text data enhancement mode to perform data enhancement on the quality complaint training set, it can specifically perform word segmentation processing on each piece of quality complaint data in the quality complaint training set according to a preset word segmentation mode to determine the split vocabulary corresponding to each piece of quality complaint data. According to the preset vocabulary similarity, the synonymous vocabulary corresponding to the split vocabulary is obtained. The synonymous vocabulary is randomly inserted into each piece of quality complaint data to achieve data enhancement of the quality complaint training set.

[0138] Optionally, when the model training module 1002 trains the preset quality appeal classification model based on the enhanced quality appeal training set and the quality appeal test set to generate a target quality appeal classification model, the model training module 1002 can specifically: configure the physical parameters of the preset preliminary quality appeal classification model. Input the enhanced quality appeal training set into the preset quality appeal classification model after the physical parameter configuration for training to determine the preliminary quality appeal classification fine-tuning model. Input the quality appeal test set into the preliminary quality appeal classification fine-tuning model quality appeal classification fine-tuning model for training to generate iterative test results. Adjust the hyperparameters of the preliminary quality appeal classification fine-tuning model quality appeal classification fine-tuning model according to the iterative test results, and determine the target hyperparameters based on the iterative test results of the preliminary quality appeal classification fine-tuning model quality appeal classification fine-tuning model after the hyperparameter adjustment. Generate a target quality appeal classification model based on the target hyperparameters and the preliminary quality appeal classification fine-tuning model quality appeal classification fine-tuning model after the hyperparameter adjustment.

[0139] Optionally, the category determination module 1003 may match the received target quality complaint data with preset complaint classification information. In the case where the historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, the historical complaint category corresponding to the historical quality complaint data is determined as the target complaint category. In the case where the historical quality complaint data identical to the target quality complaint data is not matched in the preset complaint classification information, the target quality complaint data is input into the target quality complaint classification model to output the target complaint category.

[0140] The device performs functions similar to the method provided above. For other functions, please refer to the previous description and will not be repeated here.

[0141] Fig.11 A schematic diagram of the structure of an electronic device according to an exemplary embodiment is shown. Fig.11 As shown, the electronic device 1100 of this embodiment may include: a memory 1101 and a processor 1102 .

[0142] The memory 1101 stores a computer program. When the computer program is executed by the processor 1102, the processor 1102 executes the method in the above embodiment.

[0143] The processor 1102 and the memory 1101 are connected, for example, via a bus.

[0144] Optionally, the electronic device 1100 may further include a transceiver. It should be noted that in actual applications, the number of transceivers is not limited to one, and the structure of the electronic device 1100 does not constitute a limitation on the embodiments of the present application.

[0145] Processor 1102 may be a CPU (Central Processing Unit), a general purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1102 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0146] The bus may include a path to transmit information between the above components. The bus may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0147] The memory 1101 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0148] The memory 1101 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 1102. The processor 1102 is used to execute the application code stored in the memory 1101 to implement the contents shown in the above method embodiment.

[0149] The electronic devices include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and fixed terminals such as digital TVs, desktop computers, etc. It can also be a server, etc. Fig.11 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0150] The electronic device of this embodiment can be used to execute the method of any of the above embodiments, and its implementation principle and technical effect are similar, which will not be described in detail here.

[0151] The present application also provides a non-transitory computer-readable storage medium having computer-readable instructions stored thereon. When the aforementioned instructions are executed by a processor, the processor executes the method in the above embodiment.

[0152] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a non-transient computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0153] The embodiments of the present application are described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present application. At the same time, changes or deformations made by those skilled in the art based on the ideas of the present application, the specific implementation methods and the scope of application of the present application, all belong to the scope of protection of the present application. In summary, the content of this specification should not be construed as a limitation on the present application.

Claims

1. A method for classifying steel quality complaints, characterized in that: include: Constructing a quality complaint dataset; Based on the quality complaint data set and the preset text data enhancement mode, a target quality complaint classification model is trained; According to the received target quality complaint data, the target quality complaint classification model and preset complaint classification information, a target complaint category corresponding to the target quality complaint data is determined.

2. The method according to claim 1, characterized in that The construction quality complaint dataset includes: Analyze multiple sets of public quality complaint data within a preset historical period to determine quality complaint issues and complaint data characteristics corresponding to the quality complaint issues; Setting optional complaint categories based on the quality complaint problem, the complaint data characteristics and a preset category setting mode; The quality complaint data set is constructed according to the multiple groups of public quality complaint data and the corresponding optional complaint categories.

3. The method according to claim 2, characterized in that The analyzing of multiple groups of public quality complaint data within a preset historical period to determine quality complaint issues and complaint data features corresponding to the quality complaint issues includes: For each set of public quality complaint data in the multiple sets of public quality complaint data within the preset historical period, analyzing each set of public quality complaint data based on a preset character matching pattern to determine whether each set of public quality complaint data has invalid typical features; In the case where the invalid typical features exist, the public quality complaint data corresponding to the invalid typical features are determined as invalid description categories, and the quality complaint issues and complaint data features corresponding to the quality complaint issues are determined based on valid data in the multiple groups of public quality complaint data; wherein the valid data are multiple groups of public quality complaint data after excluding the public quality complaint data corresponding to the invalid description categories; In the absence of the invalid typical features, quality complaint issues and complaint data features corresponding to the quality complaint issues are determined based on the multiple groups of public quality complaint data.

4. The method according to claim 1, characterized in that The training of a target quality complaint classification model based on the quality complaint dataset and a preset text data enhancement mode includes: Dividing the quality complaint data set according to a preset division ratio to determine a quality complaint training set and a quality complaint test set; Performing data enhancement on the quality complaint training set using a preset text data enhancement mode; Based on the enhanced quality complaint training set and the quality complaint test set, a preset quality complaint classification model is trained to generate the target quality complaint classification model.

5. The method according to claim 4, characterized in that The step of dividing the quality complaint data set according to a preset division ratio to determine a quality complaint training set and a quality complaint test set includes: Analyze the quality complaint data set to determine the proportion of sample data corresponding to each optional complaint category in the quality complaint data set; Determine a preset division ratio according to the proportion of sample data corresponding to each optional complaint category in the quality complaint data set; According to the preset division ratio and the preset stratified sampling mode, the quality complaint data set is randomly sampled to divide the sample data of each optional complaint category into the quality complaint training set and the quality complaint test set according to the corresponding division ratio.

6. The method according to claim 5, characterized in that The method further comprises dividing the quality complaint data set according to a preset division ratio to determine a quality complaint training set and a quality complaint test set: The sample data of each optional appeal category is randomized so that the probability of the sample data of each optional appeal category being extracted is the same.

7. The method according to claim 4, characterized in that The method of performing data enhancement on the quality complaint training set using a preset text data enhancement mode includes: Performing word segmentation processing on each piece of quality complaint data in the quality complaint training set according to a preset word segmentation mode to determine the split vocabulary corresponding to each piece of quality complaint data; According to the preset vocabulary similarity, obtaining synonymous words corresponding to the split words; The synonymous words are randomly inserted into each piece of quality complaint data to achieve data enhancement of the quality complaint training set.

8. The method according to claim 4, characterized in that The step of training a preset quality complaint classification model based on the enhanced quality complaint training set and the quality complaint test set to generate the target quality complaint classification model includes: Configuring physical parameters of the preset quality complaint classification model; Inputting the enhanced quality complaint training set into the preset quality complaint classification model after physical parameter configuration for training to determine a quality complaint classification fine-tuning model; Inputting the quality complaint test set into the quality complaint classification fine-tuning model for training to generate iterative test results; Adjusting the hyperparameters of the quality complaint classification fine-tuning model according to the iterative test results, and determining target hyperparameters based on the iterative test results of the quality complaint classification fine-tuning model after the hyperparameter adjustment; Based on the target hyperparameters and the quality complaint classification fine-tuning model after the hyperparameters are adjusted, the target quality complaint classification model is generated.

9. The method according to any one of claims 1 to 8, characterized in that: The determining, according to the received target quality complaint data, the target quality complaint classification model and the preset complaint classification information, the target complaint category corresponding to the target quality complaint data comprises: Matching the received target quality complaint data with the preset complaint classification information; When historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, the historical complaint category corresponding to the historical quality complaint data is determined as the target complaint category; When no historical quality complaint data identical to the target quality complaint data is matched in the preset complaint classification information, the target quality complaint data is input into the target quality complaint classification model to output the target complaint category.

10. A device for classifying steel quality complaints, characterized in that: include: Dataset construction module, used to construct quality complaint dataset; A model training module, used for training a target quality complaint classification model based on the quality complaint data set and a preset text data enhancement mode; The category determination module is used to determine the target complaint category corresponding to the target quality complaint data according to the received target quality complaint data, the target quality complaint classification model and preset complaint classification information.

11. An electronic device, characterized in that: include: processor; A memory storing a computer program, which, when executed by the processor, enables the processor to execute the method for classifying steel quality complaint content according to any one of claims 1 to 9.

12. A non-transitory computer-readable storage medium, characterized in that: Computer-readable instructions are stored thereon, and when the instructions are executed by a processor, the processor executes the method for classifying steel quality complaint content as described in any one of claims 1-9.