Complaint text classification method and device, computer device and storage medium

Abstract

The application relates to a customer complaint text classification method and device, computer equipment and a storage medium. The method comprises the following steps: preprocessing text data to be processed to obtain a plurality of word segmentation results; performing vectorization processing on the plurality of word segmentation results to obtain first vector values of the word segmentation results; inputting the plurality of word segmentation results into a pre-trained topic analysis model to obtain second vector values of each customer complaint topic to which the text data to be processed belongs and third vector values of keywords of the text data to be processed; performing splicing processing on the first vector values, the second vector values and the third vector values to obtain splicing features, and taking the splicing features as features of the text data to be processed; and inputting the features of the text data to be processed into a topic classification model and a perception classification model of a pre-trained classification model respectively to obtain a customer complaint topic category and a customer complaint perception category to which the text data to be processed belongs. The method can deeply analyze specific customer complaint contents under a field category.

Description

Technical Field

[0001] This application relates to the field of text classification technology, and in particular to a method, apparatus, computer device and storage medium for classifying customer complaint texts. Background Technology

[0002] A customer complaint refers to a written or oral objection, protest, claim, or demand for a solution made by a customer due to dissatisfaction with the quality of a company's products or services. The customer complaint handling process includes the following steps: Step 1, compiling the entire customer complaint into a written complaint document, including details such as the complainant, the time of the complaint, the target of the complaint, and the complaint request; Step 2, identifying the specific reasons for the customer complaint; Step 3, proposing a solution.

[0003] With the development of the internet, text classification of customer complaints helps provide foundational data and a range of auxiliary functions for subsequent business applications. Existing text classification methods primarily determine the topic category of the text data to be classified through a text classification model, and then derive the topic distribution of the customer complaints based on these categories. However, existing text classification methods cannot deeply analyze the specific customer complaint content within each topic category, leading to low accuracy in the classification results. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer device, and storage medium for classifying customer complaint texts that can improve the accuracy of text classification, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a method for classifying customer complaint texts. The method includes:

[0006] Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results;

[0007] The multiple word segmentation results are vectorized to obtain the first vector value of each word segmentation result;

[0008] Multiple word segmentation results are input into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0009] The first vector value, the second vector value, and the third vector value are concatenated to obtain the concatenated feature, which is then used as the feature of the text data to be processed.

[0010] The features of the text data to be processed are input into the pre-trained classification models of the topic classification model and the perception classification model to obtain the customer complaint topic category and the customer complaint perception category to which the text data to be processed belongs.

[0011] In one embodiment, the method further includes:

[0012] Obtain historical customer complaint data;

[0013] Based on a pre-built two-dimensional labeling system, the historical customer complaint data is labeled with themes and perceptions to obtain first labeled data with theme labels and second labeled data with perception labels.

[0014] The first labeled data is input into the topic classification model of the classification model to obtain the topic prediction data of the first labeled data;

[0015] The second labeled data is input into the perceptual classification model of the classification model to obtain the perceptual prediction data of the second labeled data;

[0016] Based on the error values ​​between the topic prediction data and the first labeled data, as well as the error values ​​between the perception prediction data and the second labeled data, the parameters of the topic classification model and the perception classification model are adjusted to complete one iteration.

[0017] If the error value is less than the preset value, the iteration stops, and the trained classification model is obtained.

[0018] In one embodiment, the method further includes:

[0019] Preprocessing historical customer complaint data yields word segmentation results corresponding to the historical customer complaint data;

[0020] The word segmentation results are used to extract themes and keywords to obtain the basic themes and keywords corresponding to each of the historical customer complaint data.

[0021] Each basic topic and each basic keyword is tagged to obtain three-level topic tags for each basic topic and three-level perception tags for each basic keyword;

[0022] Clustering is performed on the third-level topic tags and third-level perception tags respectively to obtain multiple topic cluster sets and perception cluster sets. Labeling is then performed on the topic cluster sets and perception cluster sets respectively to obtain the second-level topic tags corresponding to each topic cluster set and the second-level perception tags corresponding to each perception cluster set.

[0023] Clustering is performed on the secondary topic tags and secondary perception tags respectively to obtain multiple topic cluster subsets and perception cluster subsets. Labeling is then performed on the topic cluster subsets and perception cluster subsets respectively to obtain the primary topic tags corresponding to each topic cluster subset and the primary perception tags corresponding to each perception cluster subset.

[0024] A two-dimensional tagging system is determined based on three-level topic tags, three-level perception tags, two-level topic tags, two-level perception tags, one-level topic tags, and one-level perception tags.

[0025] In one embodiment, historical customer complaint data is labeled with themes and perceptions based on a pre-built two-dimensional labeling system to obtain first labeled data with theme labels and second labeled data with perception labels, including:

[0026] Vectorize the basic themes and keywords corresponding to the current historical customer complaint data to obtain the theme vector value and keyword vector value of the current historical customer complaint data;

[0027] Based on the topic vector value and keyword vector value of the current historical customer complaint data, the third-level topic tag and third-level perception tag corresponding to the current historical customer complaint data are determined, as well as the first labeled data with the third-level topic tag and the second labeled data with the third-level perception tag.

[0028] In one embodiment, historical customer complaint data is labeled with themes and perception based on a pre-built two-dimensional labeling system to obtain first labeled data with theme labels and second labeled data with perception labels, and the method further includes:

[0029] Based on the three-level topic tag, three-level perception tag and two-dimensional tag system, determine the two-level topic tag and two-level perception tag corresponding to the current historical customer complaint data, as well as the first labeled data with the three-level topic tag and the second labeled data with the three-level perception tag and the second labeled data.

[0030] In one embodiment, historical customer complaint data is labeled with themes and perception based on a pre-built two-dimensional labeling system to obtain first labeled data with theme labels and second labeled data with perception labels, and the method further includes:

[0031] Based on the secondary topic tags, secondary perception tags, and two-dimensional tag system, determine the primary topic tags and primary perception tags corresponding to the current historical customer complaint data, as well as the first labeled data with tertiary topic tags, secondary topic tags, and primary topic tags, and the second labeled data with tertiary perception tags, secondary perception tags, and primary perception tags.

[0032] In one embodiment, the text data to be processed is acquired, and the text data to be processed is preprocessed to obtain multiple word segmentation results, including:

[0033] Obtain the text data to be processed, and perform duplicate word removal and stop word removal on the text data to obtain the initial text;

[0034] The initial text is segmented into words to obtain multiple segmentation results.

[0035] Secondly, this application also provides an apparatus for classifying customer complaint texts. The apparatus includes:

[0036] The word segmentation module is used to acquire the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results;

[0037] The first vectorization module is used to vectorize multiple word segmentation results to obtain the first vector value of each word segmentation result.

[0038] The second vectorization module is used to input multiple word segmentation results into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0039] The feature concatenation module is used to concatenate the first vector value, the second vector value, and the third vector value to obtain concatenated features, which are then used as features of the text data to be processed.

[0040] The classification module is used to input the features of the text data to be processed into the pre-trained classification model of the topic classification model and the perception classification model to obtain the customer complaint topic category and customer complaint perception category to which the text data to be processed belongs.

[0041] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0042] Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results;

[0043] The multiple word segmentation results are vectorized to obtain the first vector value of each word segmentation result;

[0044] Multiple word segmentation results are input into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0045] The first vector value, the second vector value, and the third vector value are concatenated to obtain the concatenated feature, which is then used as the feature of the text data to be processed.

[0046] The features of the text data to be processed are input into the pre-trained classification models of the topic classification model and the perception classification model to obtain the customer complaint topic category and the customer complaint perception category to which the text data to be processed belongs.

[0047] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0048] Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results;

[0049] The multiple word segmentation results are vectorized to obtain the first vector value of each word segmentation result;

[0050] Multiple word segmentation results are input into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0051] The first vector value, the second vector value, and the third vector value are concatenated to obtain the concatenated feature, which is then used as the feature of the text data to be processed.

[0052] The features of the text data to be processed are input into the pre-trained classification models of the topic classification model and the perception classification model to obtain the customer complaint topic category and the customer complaint perception category to which the text data to be processed belongs.

[0053] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0054] Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results;

[0055] The multiple word segmentation results are vectorized to obtain the first vector value of each word segmentation result;

[0056] Multiple word segmentation results are input into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0057] The first vector value, the second vector value, and the third vector value are concatenated to obtain the concatenated feature, which is then used as the feature of the text data to be processed.

[0058] The features of the text data to be processed are input into the pre-trained classification models of the topic classification model and the perception classification model to obtain the customer complaint topic category and the customer complaint perception category to which the text data to be processed belongs.

[0059] The aforementioned customer complaint text classification method, apparatus, computer equipment, and storage medium concatenate the first vector values ​​of each word segmentation result of the text data to be processed, the second vector values ​​of the text data belonging to each customer complaint topic, and the third vector values ​​of the keywords in the text data to be processed to obtain concatenated features. These concatenated features are then input as features of the text data to be processed into a pre-trained classification model's domain classification model and perceptual classification model to obtain the domain category and perceptual category to which the text data to be processed belongs. This method not only enhances the influence of topics and keywords on the classification model, thereby improving the accuracy of the classification model, but also divides the text data to be processed into specific topic dimensions and the corresponding questions for each topic dimension, deeply analyzing the specific customer complaint content under the domain category and improving the accuracy of the classification results. Attached Figure Description

[0060] Figure 1 This is a diagram illustrating the application environment of a customer complaint text classification method in one embodiment.

[0061] Figure 2 This is a flowchart illustrating a customer complaint text classification method in one embodiment;

[0062] Figure 3 This is a flowchart illustrating the process of determining a two-dimensional labeling system in one embodiment;

[0063] Figure 4 This is a structural diagram of a two-dimensional label system in another embodiment;

[0064] Figure 5 This is a flowchart illustrating the training process of a classification model in one embodiment.

[0065] Figure 6 This is a structural diagram of a customer complaint text classification device in one embodiment;

[0066] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0068] The customer complaint text classification method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 obtains the text data to be processed from server 104, preprocesses the text data to obtain multiple word segmentation results; vectorizes the multiple word segmentation results to obtain the first vector value of each word segmentation result; inputs the multiple word segmentation results into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed; concatenates the first vector value, the second vector value, and the third vector value to obtain the concatenated feature, and uses the concatenated feature as the feature of the text data to be processed; input the feature of the text data to be processed into the topic classification model and the perception classification model of the pre-trained classification model respectively to obtain the customer complaint topic category and the customer complaint perception category to which the text data to be processed belongs. Terminal 102 stores the customer complaint topic category and the customer complaint perception category to which the text data to be processed belongs in the server. The data storage system can store the historical customer complaint data of server 104, the customer complaint topic category to which the text data to be processed belongs, and the customer complaint perception category. The data storage system can be integrated on server 104, or it can be placed in the cloud or other network servers. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include intelligent voice robots, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0069] In one embodiment, such as Figure 2 As shown, a customer complaint text classification method is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps:

[0070] S202: Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results.

[0071] The text data to be processed can include information such as the complainant, complaint time, target of the complaint, and complaint request, or it can be dialogue data recorded by an intelligent voice robot. The text data to be processed can be in Word or Excel format.

[0072] The preprocessing of the text data to be processed mainly includes data cleaning and word segmentation. Data cleaning is to remove the invalid information in the text data to be processed and obtain the valid information. Word segmentation is to split the valid information into the domain word library of the text data to be processed and the nouns of specific products in this domain, and obtain multiple word segmentation results. The word segmentation results include the key information of the text data to be processed. For example, the word segmentation results include one or more of the information such as the theme, keywords, complainant, complaint problem, and complaint object.

[0073] Specifically, obtain the text data to be processed, perform data cleaning on the text data to be processed to obtain the valid information, and perform word segmentation on the valid information to obtain multiple word segmentation results.

[0074] In one embodiment, since the text data to be processed includes the communication content between the account manager and the customer, therefore, when analyzing, it is necessary to extract the key content from the communication content. In order to improve the accuracy of extracting the key content from the communication content, this embodiment specifically includes the following steps:

[0075] Step 1, obtain the text data to be processed, perform duplicate word removal and stop word removal on the text data to be processed, and obtain the initial text.

[0076] Among them, data cleaning includes duplicate word removal, stop word removal, etc. Duplicate word removal is mainly to remove the duplicate fields in the text data to be processed. Stop word removal is to remove the words without actual meaning in the sentence, such as "ah", "um", "of", etc.

[0077] Step 2, perform word segmentation on the initial text to obtain multiple word segmentation results.

[0078] Among them, the initial text includes multiple conversations. Therefore, the initial text is segmented into sentences with the period as the node to obtain the conversation sentence set, and the conversation sentence set is segmented by using a word segmentation tool or algorithm to obtain multiple word segmentation results.

[0079] For example, use JIEBA word segmentation to perform word segmentation on the valid information of the text data to be processed, and obtain the specific names of the fields and products to which the text data to be processed belongs.

[0080] S204, perform vectorization processing on the multiple word segmentation results to obtain the first vector value of each word segmentation result.

[0081] Among them, the purpose of vectorization processing is to determine the occurrence frequency of each word segmentation result in the text data to be processed. The first vector value represents the probability value of the occurrence of each word segmentation result. The multiple word segmentation results can be vectorized respectively through the pre-constructed CountVector vectorization model to obtain the first vector value of each word segmentation result.

[0082] Specifically, the terminal uses the CountVector vectorization model to convert each word segmentation result into a word frequency matrix, calculates the number of times each word appears, and determines the first vector value of each word segmentation result based on the number of times each word appears.

[0083] S206. Input multiple word segmentation results into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0084] The second vector value represents the probability that the text data to be processed belongs to each customer complaint topic; the third vector value represents the probability distribution of the keywords in the text data to be processed. The topic analysis model can identify topics in a document and mine hidden information in the corpus. Specifically, it extracts topics and keywords from each word segmentation result and outputs the probability values ​​of the text data to be processed belonging to each customer complaint topic, as well as the probability values ​​of the keywords in the text data to be processed. For example, the topic analysis model can be an LDA topic model.

[0085] Specifically, the terminal inputs multiple word segmentation results into a pre-trained topic analysis model. The topic analysis model extracts topics and keywords from each word segmentation result and outputs the probability value of the text data to be processed belonging to each customer complaint topic, as well as the probability value of the keywords in the text data to be processed.

[0086] S208, the first vector value, the second vector value and the third vector value are concatenated to obtain the concatenated feature, and the concatenated feature is used as the feature of the text data to be processed.

[0087] If the first vector value of each word segmentation result is used to represent the feature of the text data to be processed, there is a problem of single feature. In order to solve the above problem, this embodiment adds a second vector value of the text data to be processed belonging to each customer complaint topic, and a third vector value of the keywords of the text data to be processed, on the basis of the first vector value of each word segmentation result. By adding the second vector value and the third vector value, the influence of topics and keywords on the classification model can be enhanced, thereby improving the accuracy of the classification model.

[0088] The terminal can use a deep learning feature fusion method to concatenate the first vector value, the second vector value, and the third vector value to obtain concatenated features.

[0089] S210, the features of the text data to be processed are input into the pre-trained classification model of the topic classification model and the perception classification model to obtain the customer complaint topic category and customer complaint perception category to which the text data to be processed belongs.

[0090] The topic classification model identifies the topic categories of the text data to be processed. In this embodiment, the topic classification model can be the LDA topic model. The perceptual classification model perceives the hidden questions under the topics of the text data to be processed and obtains the perceptual category corresponding to the questions. In this embodiment, the perceptual classification model can be the hierarchical attention model. The hierarchical attention model assigns different weights to each word vector and sums them to obtain sentence vectors. It then assigns different weights to each sentence vector and sums them to obtain text vectors. The perceptual question classification is achieved through fully connected layers and a softmax activation function.

[0091] In the aforementioned customer complaint text classification method, the first vector value of each word segmentation result of the text data to be processed, the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed are concatenated to obtain concatenated features. The concatenated features are then used as features of the text data to be processed and input into the pre-trained classification model of the domain classification model and the perception classification model to obtain the domain category and perception category to which the text data to be processed belongs. This not only enhances the influence of topics and keywords on the classification model, thereby improving the accuracy of the classification model, but also divides the text data to be processed into specific topic dimensions and the corresponding questions of topic dimensions, deeply analyzes the specific customer complaint content under the domain category, and improves the accuracy of the classification results.

[0092] In training a classification model using historical customer complaint data, it is necessary to label the historical customer complaint data with topic and keyword tags. Existing labeling methods generally include two approaches: the first is that business personnel manually classify and label the customer complaint data based on their business experience; the second is to train a customer complaint data classifier using a large amount of manually labeled data, and then classify and label the data using the classifier. Existing labeling methods not only require significant manpower for data labeling, but the specific labeling system is also fixed, which has certain limitations. Furthermore, the varying levels of expertise and subjective judgment among labelers can lead to inaccurate labeling, resulting in low reliability of the trained classification model. Therefore, to address these issues, this embodiment constructs a two-dimensional labeling system. This two-dimensional labeling system refers to a multi-level labeling system encompassing both topic classification and perception classification dimensions, including multi-level topic tags and multi-level perception tags. The label levels of the two-dimensional labeling system are determined based on the specific application scenario. In this embodiment, the two-dimensional labeling system has three levels, such as... Figure 3 As shown, the construction process of a two-dimensional labeling system includes the following steps:

[0093] S302, preprocess each historical customer complaint data to obtain the word segmentation results corresponding to each historical customer complaint data.

[0094] The implementation process of this step is the same as that of S202, and will not be repeated here.

[0095] S304, extract the theme and keywords from the word segmentation results to obtain the basic theme and basic keywords corresponding to each historical customer complaint data.

[0096] The implementation process of this step is the same as that of S206, and will not be repeated here.

[0097] S306, tag each basic topic and each basic keyword to obtain the three-level topic tags corresponding to each basic topic and the three-level perception tags corresponding to each basic keyword.

[0098] The terminal retrieves topic tags for specific application scenarios from a topic thesaurus, and obtains perception tags corresponding to specific topics from a perception thesaurus. The topic thesaurus includes all topic tags corresponding to specific application scenarios; the perception thesaurus includes all perception tags corresponding to each topic. Third-level topic tags represent topic tags for specific business scenarios, and third-level perception tags represent perception tags under specific topics.

[0099] Specifically, the terminal determines the topic tags of historical customer complaint data in the specific business scenario from the topic thesaurus, and uses the topic tags as the third-level topic tags of the historical customer complaint data; the terminal determines the corresponding problem of the historical customer complaint data under the specific third-level topic tags, and obtains the perception tags of the historical customer complaint data under the third-level topic tags from the perception thesaurus, and uses the perception tags as the third-level perception tags of the historical customer complaint data.

[0100] S308, perform clustering processing on the third-level topic tags and third-level perception tags respectively to obtain multiple topic cluster sets and perception cluster sets. Perform tag labeling processing on the topic cluster sets and perception cluster sets respectively to obtain the second-level topic tags corresponding to each topic cluster set and the second-level perception tags corresponding to each perception cluster set.

[0101] Specifically, the terminal can use clustering algorithms to cluster the third-level topic tags and third-level perception tags respectively. The topic cluster set includes third-level topic tags of multiple similar business scenarios; the perception cluster set includes third-level perception tags corresponding to multiple similar topics.

[0102] Specifically, the terminal uses a clustering algorithm to cluster the three-level topic tags separately, obtaining multiple topic cluster sets; the terminal uses a clustering algorithm to cluster the three-level perception tags separately, obtaining multiple perception cluster sets; the terminal summarizes the topic tags of multiple similar business scenarios included in each topic cluster set, obtaining the corresponding secondary topic tags for each topic cluster set, and performs secondary topic tag annotation processing on each topic cluster set, obtaining multiple topic cluster sets with secondary topic tags; the terminal summarizes the perception tags corresponding to multiple similar topics included in each perception cluster set, obtaining the corresponding secondary perception tags for each perception cluster set, and performs secondary perception tag annotation processing on each perception cluster set, obtaining multiple perception cluster sets with secondary perception tags.

[0103] S310, perform clustering processing on the secondary topic labels and secondary perception labels respectively to obtain multiple topic cluster subsets and perception cluster subsets. Perform label annotation processing on the topic cluster subsets and perception cluster subsets respectively to obtain the primary topic label corresponding to each topic cluster subset and the primary perception label corresponding to each perception cluster subset.

[0104] Specifically, the terminal can use clustering algorithms to cluster the secondary topic tags and secondary perception tags separately. The topic clustering subset includes secondary topic tags from multiple similar business scenarios; the perception clustering subset includes secondary perception tags corresponding to multiple similar topics.

[0105] Specifically, the terminal uses clustering algorithms to cluster the secondary topic tags separately, obtaining multiple topic cluster subsets; the terminal uses clustering algorithms to cluster the secondary perception tags separately, obtaining multiple perception cluster subsets; the terminal summarizes the topic tags of multiple similar business scenarios included in each topic cluster subset, obtaining the primary topic tags corresponding to each topic cluster subset, and performs primary topic tag annotation processing on each topic cluster subset, obtaining multiple topic cluster subsets with primary topic tags; the terminal summarizes the perception tags corresponding to multiple similar topics included in each perception cluster subset, obtaining the primary perception tags corresponding to each perception cluster subset, and performs primary perception tag annotation processing on each perception cluster subset, obtaining multiple perception cluster subsets with primary perception tags.

[0106] S312, a two-dimensional labeling system is determined based on three-level topic labels, three-level perception labels, two-level topic labels, two-level perception labels, one-level topic labels, and one-level perception labels.

[0107] Among them, such as Figure 4As shown, by sequentially connecting the third-level topic tags, the second-level topic tags, and the first-level topic tags, a topic tag system can be obtained. By sequentially connecting the third-level perception tags, the second-level perception tags, and the first-level perception tags, a perception tag system can be obtained. The topic tag system and the perception tag system are collectively referred to as a two-dimensional tag system.

[0108] In this embodiment, historical customer complaint data is labeled using a two-dimensional to three-dimensional labeling system. On the one hand, this allows for flexible construction of a classification system, and on the other hand, it enables batch labeling of historical customer complaint data. This avoids the limitations of manual labeling and the problem of low quality of labeled data caused by subjective factors, reduces the impact of low quality of training data caused by the subjectivity of manual labeling on the classification model's performance, and improves the reliability of the classification model.

[0109] In one embodiment, such as Figure 5 As shown, the training process of the classification model includes:

[0110] S502, obtain historical customer complaint data.

[0111] S504, based on a pre-built two-dimensional labeling system, performs topic and perception labeling on historical customer complaint data to obtain first labeled data with topic labels and second labeled data with perception labels.

[0112] Specifically, the terminal uses a pre-built two-dimensional labeling system to annotate historical customer complaint data with multi-level topic tags and multi-level perception tags, which can yield at least three types of annotation results. The first type of annotation result is: first annotated data with three-level topic tags and second annotated data with three-level perception tags. The second type of annotation result is: first annotated data with three-level topic tags and two-level topic tags and second annotated data with three-level perception tags and two-level perception tags. The third type of annotation result is: first annotated data with three-level topic tags, two-level topic tags and one-level topic tags and second annotated data with three-level perception tags, two-level perception tags and one-level perception tags.

[0113] S506, input the first labeled data into the topic classification model of the classification model to obtain the topic prediction data of the first labeled data.

[0114] Specifically, if the first labeled data includes third-level topic tags, then the topic prediction data output by the topic classification model also includes third-level topic tags; if the first labeled data includes third-level topic tags and second-level topic tags, then the topic prediction data output by the topic classification model also includes third-level topic tags and second-level topic tags; if the first labeled data includes third-level topic tags and second-level topic tags, then the topic prediction data output by the topic classification model also includes third-level topic tags, second-level topic tags, and first-level topic tags.

[0115] S508, input the second labeled data into the perception classification model of the classification model to obtain the perception prediction data of the second labeled data.

[0116] Specifically, if the second labeled data includes a third-level perception label, then the perception prediction data output by the perception classification model also includes a third-level perception label; if the second labeled data includes both third-level and second-level perception labels, then the perception prediction data output by the perception classification model also includes both third-level and second-level perception labels; if the second labeled data includes both third-level and second-level perception labels, then the perception prediction data output by the perception classification model also includes third-level, second-level, and first-level perception labels.

[0117] S510: Based on the error values ​​between the topic prediction data and the first labeled data, as well as the error values ​​between the perception prediction data and the second labeled data, adjust the parameters of the topic classification model and the perception classification model to complete one iteration.

[0118] The terminal can use a matrix analysis-based method to determine the error values ​​between the topic prediction data and the first labeled data, as well as the error values ​​between the perceived prediction data and the second labeled data.

[0119] Specifically, the terminal determines the error value between the topic prediction data and the first labeled data based on the variability of the matrix output by the topic classification model relative to the classification sample grouping. The terminal determines the error value between the perceptual prediction data and the second labeled data based on the variability of the matrix output by the perceptual classification model relative to the classification sample grouping. When the error value is greater than the preset value, the parameters of the topic classification model and / or the perceptual classification model are adjusted to complete one iteration.

[0120] S512, if the error value is less than the preset value, stop the iteration and obtain the trained classification model.

[0121] If the error value is less than the preset value, it means that the reliability of the topic classification model and / or the perception classification model meets the business requirements. At this time, the iteration stops and the trained classification model is obtained.

[0122] If the training samples of the classification model are labeled with only three levels of topic labels and three levels of perception labels, then the customer complaint topic category predicted by the classification model to belong to the text data to be processed will be a three-level topic category, and the customer complaint perception category will be a three-level perception category. Similarly, if the training samples of the classification model are labeled with three levels of topic labels and perception labels, then the customer complaint topic category predicted by the classification model to belong to the text data to be processed will be a three-level topic category, and the customer complaint perception category will be a three-level perception category.

[0123] In this embodiment, by batch labeling historical customer complaint data from two dimensions—domain and perception—not only can the training data required for the classification model be prepared quickly, but the problem of low quality of labeled data caused by different human experience and subjective factors can also be reduced, thereby improving the reliability of the classification model.

[0124] In one embodiment, historical customer complaint data is labeled with themes and perceptions based on a pre-built two-dimensional labeling system to obtain first labeled data with theme labels and second labeled data with perception labels, including:

[0125] Step 1: Vectorize the basic themes and keywords corresponding to the current historical customer complaint data to obtain the theme vector value and keyword vector value of the current historical customer complaint data.

[0126] This step is the same as the step in S206 where the topic analysis model is used to obtain topic vector values ​​and keyword vector values, and will not be repeated here.

[0127] Step 2: Based on the topic vector value and keyword vector value of the current historical customer complaint data, determine the third-level topic tag and third-level perception tag corresponding to the current historical customer complaint data, as well as the first labeled data with the third-level topic tag and the second labeled data with the third-level perception tag.

[0128] Specifically, the topic with the largest topic vector value is selected as the third-level topic label for the current historical customer complaint data, and the keyword with the largest keyword vector value is selected as the third-level perception label for the current historical customer complaint data.

[0129] In this embodiment, customer complaint data is classified from two dimensions: subject and perception. This allows for a more accurate identification of specific issues in the specific areas reflected in the complaints, making the analysis of customer complaint data clearer and more accurate.

[0130] In one embodiment, historical customer complaint data is labeled with themes and perceptions based on a pre-built two-dimensional labeling system to obtain first labeled data with theme labels and second labeled data with perception labels, and the method further includes:

[0131] Based on the three-level topic tag, three-level perception tag and two-dimensional tag system, determine the two-level topic tag and two-level perception tag corresponding to the current historical customer complaint data, as well as the first labeled data with the three-level topic tag and the second labeled data with the three-level perception tag and the second labeled data.

[0132] In this process, after knowing the three-level topic tags and three-level perception tags of the current historical customer complaint data, the two-dimensional tag system can be used to determine the two-level topic tags and two-level perception tags after clustering the three-level topic tags and three-level perception tags. These two-level topic tags and two-level perception tags are then used as the two-level topic tags and two-level perception tags corresponding to the current historical customer complaint data.

[0133] In this embodiment, by performing secondary labeling on the current historical customer complaint data, we can obtain labeled data with deeper levels and broader scope, which can more accurately uncover the higher-level issues reflected in the customer complaints.

[0134] In one embodiment, historical customer complaint data is labeled with themes and perceptions based on a pre-built two-dimensional labeling system to obtain first labeled data with theme labels and second labeled data with perception labels, and the method further includes:

[0135] Based on the secondary topic tags, secondary perception tags, and two-dimensional tag system, determine the primary topic tags and primary perception tags corresponding to the current historical customer complaint data, as well as the first labeled data with tertiary topic tags, secondary topic tags, and primary topic tags, and the second labeled data with tertiary perception tags, secondary perception tags, and primary perception tags.

[0136] In this process, after knowing the secondary topic tags and secondary perception tags of the current historical customer complaint data, the primary topic tags and primary perception tags after clustering the secondary topic tags and secondary perception tags can be determined according to the two-dimensional tag system. The primary topic tags and primary perception tags are then used as the primary topic tags and primary perception tags corresponding to the current historical customer complaint data.

[0137] In this embodiment, by performing three-level labeling on the current historical customer complaint data, the highest-level domain labeling data can be obtained, which can more accurately uncover the advanced issues reflected in the customer complaints.

[0138] In one embodiment, a customer complaint text classification method is applied to... Figure 1 Taking terminal 102 as an example, the specific steps are as follows:

[0139] S602, preprocess each historical customer complaint data to obtain the word segmentation results corresponding to each historical customer complaint data.

[0140] S604 extracts themes and keywords from the word segmentation results to obtain the basic themes and keywords corresponding to each historical customer complaint data.

[0141] S606 performs tagging processing on each basic topic and each basic keyword to obtain three-level topic tags corresponding to each basic topic and three-level perception tags corresponding to each basic keyword.

[0142] S608, perform clustering processing on the third-level topic tags and third-level perception tags respectively to obtain multiple topic cluster sets and perception cluster sets. Perform tag labeling processing on the topic cluster sets and perception cluster sets respectively to obtain the second-level topic tags corresponding to each topic cluster set and the second-level perception tags corresponding to each perception cluster set.

[0143] S610, perform clustering processing on the secondary topic labels and secondary perception labels respectively to obtain multiple topic cluster subsets and perception cluster subsets. Perform label annotation processing on the topic cluster subsets and perception cluster subsets respectively to obtain the primary topic label corresponding to each topic cluster subset and the primary perception label corresponding to each perception cluster subset.

[0144] S612, based on three-level topic tags, three-level perception tags, two-level topic tags, two-level perception tags, one-level topic tags, and one-level perception tags, a two-dimensional tagging system is determined.

[0145] S616, based on a pre-built two-dimensional labeling system, performs topic and perception labeling on historical customer complaint data to obtain first labeled data with topic labels and second labeled data with perception labels.

[0146] S618, input the first labeled data into the topic classification model of the classification model to obtain the topic prediction data of the first labeled data.

[0147] S620, input the second labeled data into the perception classification model of the classification model to obtain the perception prediction data of the second labeled data.

[0148] S622, based on the error values ​​of the topic prediction data and the first labeled data, as well as the error values ​​of the perception prediction data and the second labeled data, adjust the parameters of the topic classification model and the perception classification model to complete one iteration.

[0149] S624. If the error value is less than the preset value, stop the iteration and obtain the trained classification model.

[0150] S626: Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results.

[0151] S628 performs vectorization processing on multiple word segmentation results to obtain the first vector value of each word segmentation result.

[0152] S630: Input multiple word segmentation results into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0153] S632, the first vector value, the second vector value and the third vector value are concatenated to obtain the concatenated feature, and the concatenated feature is used as the feature of the text data to be processed.

[0154] S634, the features of the text data to be processed are input into the pre-trained classification model of the topic classification model and the perception classification model to obtain the customer complaint topic category and customer complaint perception category to which the text data to be processed belongs.

[0155] In this embodiment, historical customer complaint data is labeled using a two-dimensional to three-dimensional labeling system. On the one hand, this allows for flexible construction of a classification system, and on the other hand, it enables batch labeling of historical customer complaint data. This avoids the limitations of manual labeling and the problem of low quality of labeled data caused by subjective factors, reduces the impact of low quality of training data caused by the subjectivity of manual labeling on the classification model's performance, and improves the reliability of the classification model.

[0156] By concatenating the first vector values ​​of each word segmentation result of the text data to be processed, the second vector values ​​of the text data belonging to each customer complaint topic, and the third vector values ​​of the keywords of the text data to be processed, a concatenated feature is obtained. This concatenated feature is then used as the feature of the text data to be processed and input into the pre-trained classification model's domain classification model and perception classification model to obtain the domain category and perception category to which the text data to be processed belongs. This not only enhances the influence of topics and keywords on the classification model, thereby improving the accuracy of the classification model, but also divides the text data to be processed into specific topic dimensions and the corresponding perception dimensions, allowing for in-depth analysis of the specific customer complaint content under the domain category and improving the accuracy of the classification results.

[0157] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0158] Based on the same inventive concept, this application also provides a customer complaint text classification device for implementing the aforementioned customer complaint text classification method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more customer complaint text classification device embodiments provided below can be found in the limitations of the customer complaint text classification method described above, and will not be repeated here.

[0159] In one embodiment, such as Figure 6 As shown, a customer complaint text classification device is provided, including:

[0160] The word segmentation module 100 is used to acquire the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results;

[0161] The first vectorization module 200 is used to perform vectorization processing on multiple word segmentation results to obtain the first vector value of each word segmentation result;

[0162] The second vectorization module 300 is used to input multiple word segmentation results into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0163] The feature concatenation module 400 is used to concatenate the first vector value, the second vector value, and the third vector value to obtain concatenated features, which are then used as features of the text data to be processed.

[0164] The classification module 500 is used to input the features of the text data to be processed into the pre-trained classification model of the topic classification model and the perception classification model to obtain the customer complaint topic category and customer complaint perception category to which the text data to be processed belongs.

[0165] In one embodiment, the word segmentation module 100 is further configured to acquire text data to be processed, perform duplicate word removal and stop word removal on the text data to be processed to obtain initial text, and perform word segmentation on the initial text to obtain multiple word segmentation results.

[0166] In one embodiment, the customer complaint text classification device further includes: a training module 600;

[0167] The training module 600 is used to acquire historical customer complaint data; based on a pre-built two-dimensional labeling system, the historical customer complaint data is labeled with topics and perceptions to obtain first labeled data with topic labels and second labeled data with perception labels; the first labeled data is input into the topic classification model of the classification model to obtain topic prediction data of the first labeled data; the second labeled data is input into the perception classification model of the classification model to obtain perception prediction data of the second labeled data; based on the error value between the topic prediction data and the first labeled data, as well as the error value between the perception prediction data and the second labeled data, the parameters of the topic classification model and the perception classification model are adjusted to complete one iteration; if the error value is less than the preset value, the iteration stops and the trained classification model is obtained.

[0168] In one embodiment, the customer complaint text classification device further includes: a construction module 700;

[0169] The construction module 700 is used to preprocess historical customer complaint data to obtain word segmentation results corresponding to the historical customer complaint data; extract topics and keywords from the word segmentation results to obtain basic topics and basic keywords corresponding to each historical customer complaint data; perform tagging processing on each basic topic and each basic keyword to obtain three-level topic tags corresponding to each basic topic and three-level perception tags corresponding to each basic keyword; perform clustering processing on the three-level topic tags and three-level perception tags respectively to obtain multiple topic cluster sets and perception cluster sets, and perform tagging processing on the topic cluster sets and perception cluster sets respectively. The annotation process yields secondary topic labels for each topic cluster set and secondary perception labels for each perception cluster set. Clustering is then performed on the secondary topic labels and secondary perception labels to obtain multiple topic cluster subsets and perception cluster subsets. Labeling is then performed on these subsets to obtain primary topic labels for each topic cluster subset and primary perception labels for each perception cluster subset. Based on the tertiary topic labels, tertiary perception labels, secondary topic labels, secondary perception labels, primary topic labels, and primary perception labels, a two-dimensional labeling system is determined.

[0170] In one embodiment, the training module 600 is further configured to: vectorize the basic topics and basic keywords corresponding to the current historical customer complaint data to obtain the topic vector value and keyword vector value of the current historical customer complaint data; and based on the topic vector value and keyword vector value of the current historical customer complaint data, determine the third-level topic label and third-level perception label corresponding to the current historical customer complaint data, as well as the first labeled data with the third-level topic label and the second labeled data with the third-level perception label.

[0171] In one embodiment, the training module 600 is further configured to: determine the secondary topic tags and secondary perception tags corresponding to the current historical customer complaint data, as well as the first labeled data with the secondary topic tags and the secondary topic tags and the second labeled data with the secondary perception tags and the secondary perception tags, based on the three-level topic tags, the three-level perception tags and the two-level tag system.

[0172] In one embodiment, the training module 600 is further configured to: determine the first-level topic label and first-level perception label corresponding to the current historical customer complaint data, as well as the first labeled data with the third-level topic label, the second-level topic label and the first-level topic label, and the second labeled data with the third-level perception label, the second-level perception label and the first-level perception label, based on the second-level topic label, the second-level perception label and the two-dimensional label system.

[0173] Each module in the aforementioned customer complaint text classification method device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0174] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a customer complaint text classification method. The display unit of the computer device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0175] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0176] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0177] Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results;

[0178] The multiple word segmentation results are vectorized to obtain the first vector value of each word segmentation result;

[0179] Multiple word segmentation results are input into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed.

[0180] The first vector value, the second vector value, and the third vector value are concatenated to obtain the concatenated feature, which is then used as the feature of the text data to be processed.

[0181] The features of the text data to be processed are input into the pre-trained classification models of the topic classification model and the perception classification model to obtain the customer complaint topic category and the customer complaint perception category to which the text data to be processed belongs.

[0182] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0183] Obtain the text data to be processed, and perform duplicate word removal and stop word removal on the text data to obtain the initial text;

[0184] The initial text is segmented into words to obtain multiple segmentation results.

[0185] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0186] Obtain historical customer complaint data;

[0187] Based on a pre-built two-dimensional labeling system, the historical customer complaint data is labeled with themes and perceptions to obtain first labeled data with theme labels and second labeled data with perception labels.

[0188] The first labeled data is input into the topic classification model of the classification model to obtain the topic prediction data of the first labeled data;

[0189] The second labeled data is input into the perceptual classification model of the classification model to obtain the perceptual prediction data of the second labeled data;

[0190] Based on the error values ​​between the topic prediction data and the first labeled data, as well as the error values ​​between the perception prediction data and the second labeled data, the parameters of the topic classification model and the perception classification model are adjusted to complete one iteration.

[0191] If the error value is less than the preset value, the iteration stops, and the trained classification model is obtained.

[0192] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0193] Preprocessing historical customer complaint data yields word segmentation results corresponding to the historical customer complaint data;

[0194] The word segmentation results are used to extract themes and keywords to obtain the basic themes and keywords corresponding to each of the historical customer complaint data.

[0195] Each basic topic and each basic keyword is tagged to obtain three-level topic tags for each basic topic and three-level perception tags for each basic keyword;

[0196] Clustering is performed on the third-level topic tags and third-level perception tags respectively to obtain multiple topic cluster sets and perception cluster sets. Labeling is then performed on the topic cluster sets and perception cluster sets respectively to obtain the second-level topic tags corresponding to each topic cluster set and the second-level perception tags corresponding to each perception cluster set.

[0197] Clustering is performed on the secondary topic tags and secondary perception tags respectively to obtain multiple topic cluster subsets and perception cluster subsets. Labeling is then performed on the topic cluster subsets and perception cluster subsets respectively to obtain the primary topic tags corresponding to each topic cluster subset and the primary perception tags corresponding to each perception cluster subset.

[0198] A two-dimensional tagging system is determined based on three-level topic tags, three-level perception tags, two-level topic tags, two-level perception tags, one-level topic tags, and one-level perception tags.

[0199] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0200] Vectorize the basic themes and keywords corresponding to the current historical customer complaint data to obtain the theme vector value and keyword vector value of the current historical customer complaint data;

[0201] Based on the topic vector value and keyword vector value of the current historical customer complaint data, the third-level topic tag and third-level perception tag corresponding to the current historical customer complaint data are determined, as well as the first labeled data with the third-level topic tag and the second labeled data with the third-level perception tag.

[0202] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0203] Based on the three-level topic tag, three-level perception tag and two-dimensional tag system, determine the two-level topic tag and two-level perception tag corresponding to the current historical customer complaint data, as well as the first labeled data with the three-level topic tag and the second labeled data with the three-level perception tag and the second labeled data.

[0204] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0205] Based on the secondary topic tags, secondary perception tags, and two-dimensional tag system, determine the primary topic tags and primary perception tags corresponding to the current historical customer complaint data, as well as the first labeled data with tertiary topic tags, secondary topic tags, and primary topic tags, and the second labeled data with tertiary perception tags, secondary perception tags, and primary perception tags.

[0206] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0207] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0208] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0209] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0210] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0211] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for classifying customer complaint texts, characterized in that, The method includes: Obtain the text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results; The multiple word segmentation results are vectorized to obtain the first vector value of each word segmentation result; The multiple word segmentation results are input into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed. The first vector value, the second vector value, and the third vector value are concatenated to obtain concatenated features, which are then used as features of the text data to be processed. The features of the text data to be processed are respectively input into the topic classification model and the perception classification model of the pre-trained classification model to obtain the customer complaint topic category and customer complaint perception category to which the text data to be processed belongs. The training data for the topic classification model and the perception classification model are generated based on a two-dimensional labeling system. The construction process of the two-dimensional labeling system includes: preprocessing historical customer complaint data to obtain the word segmentation results corresponding to the historical customer complaint data. The word segmentation results are used to extract themes and keywords to obtain the basic themes and keywords corresponding to each of the historical customer complaint data. Each basic topic and each basic keyword is labeled to obtain a three-level topic tag for each basic topic and a three-level perception tag for each basic keyword; Clustering is performed on the third-level topic tags and the third-level perception tags to obtain multiple topic cluster sets and perception cluster sets. The topic tags of multiple similar business scenarios included in each topic cluster set are summarized to obtain the second-level topic tags corresponding to each topic cluster set. Second-level topic tag annotation is performed on each topic cluster set to obtain multiple topic cluster sets with second-level topic tags. Similarly, the perception tags corresponding to multiple similar topics included in each perception cluster set are summarized to obtain the second-level perception tags corresponding to each perception cluster set. Second-level perception tag annotation is performed on each perception cluster set to obtain multiple perception cluster sets with second-level perception tags. Clustering is performed on the secondary topic tags and the secondary perception tags respectively to obtain multiple topic cluster subsets and perception cluster subsets. Tag labeling is performed on the topic cluster subsets and the perception cluster subsets respectively to obtain the primary topic tag corresponding to each topic cluster subset and the primary perception tag corresponding to each perception cluster subset. A two-dimensional tagging system is determined based on the three-level topic tags, the three-level perception tags, the two-level topic tags, the two-level perception tags, the one-level topic tags, and the one-level perception tags.

2. The method according to claim 1, characterized in that, The method further includes: Obtain historical customer complaint data; Based on a pre-built two-dimensional labeling system, the historical customer complaint data is labeled with themes and perceptions to obtain first labeled data with theme labels and second labeled data with perception labels. The first labeled data is input into the topic classification model of the classification model to obtain the topic prediction data of the first labeled data; The second labeled data is input into the perceptual classification model of the classification model to obtain the perceptual prediction data of the second labeled data; Based on the error values ​​between the topic prediction data and the first labeled data, and the error values ​​between the perception prediction data and the second labeled data, the parameters of the topic classification model and the perception classification model are adjusted to complete one iteration. If the error value is less than the preset value, the iteration stops, and the trained classification model is obtained.

3. The method according to claim 1, characterized in that, The historical customer complaint data is labeled with themes and perception based on a pre-built two-dimensional labeling system to obtain first labeled data with theme tags and second labeled data with perception tags, including: Vectorize the basic themes and keywords corresponding to the current historical customer complaint data to obtain the theme vector value and keyword vector value of the current historical customer complaint data; Based on the topic vector value and keyword vector value of the current historical customer complaint data, the third-level topic tag and third-level perception tag corresponding to the current historical customer complaint data are determined, as well as the first labeled data with the third-level topic tag and the second labeled data with the third-level perception tag.

4. The method according to claim 3, characterized in that, The process of using a pre-built two-dimensional labeling system to perform topic and perception labeling on the historical customer complaint data to obtain first labeled data with topic labels and second labeled data with perception labels also includes: Based on the three-level topic tags, the three-level perception tags, and the two-dimensional tag system, determine the two-level topic tags and two-level perception tags corresponding to the current historical customer complaint data, as well as the first labeled data with the three-level topic tags and the second labeled data with the three-level perception tags and the second labeled data.

5. The method according to claim 4, characterized in that, The process of using a pre-built two-dimensional labeling system to perform topic and perception labeling on the historical customer complaint data to obtain first labeled data with topic labels and second labeled data with perception labels also includes: Based on the secondary topic tags, the secondary perception tags, and the two-dimensional tag system, determine the primary topic tags and primary perception tags corresponding to the current historical customer complaint data, as well as the first labeled data with tertiary topic tags, secondary topic tags, and primary topic tags, and the second labeled data with tertiary perception tags, secondary perception tags, and primary perception tags.

6. The method according to claim 1, characterized in that, The process involves acquiring the text data to be processed, preprocessing the text data to obtain multiple word segmentation results, including: Obtain the text data to be processed, and perform duplicate word removal and stop word removal on the text data to be processed to obtain the initial text; The initial text is segmented into words to obtain multiple segmentation results.

7. A customer complaint text classification device, characterized in that, The device includes: The word segmentation module is used to acquire text data to be processed, preprocess the text data to be processed, and obtain multiple word segmentation results; The first vectorization module is used to perform vectorization processing on the multiple word segmentation results to obtain the first vector value of each word segmentation result; The second vectorization module is used to input multiple word segmentation results into a pre-trained topic analysis model to obtain the second vector value of the text data to be processed belonging to each customer complaint topic, and the third vector value of the keywords of the text data to be processed. The feature concatenation module is used to concatenate the first vector value, the second vector value, and the third vector value to obtain concatenated features, and use the concatenated features as features of the text data to be processed. The classification module is used to input the features of the text data to be processed into the topic classification model and the perception classification model of the pre-trained classification model, respectively, to obtain the customer complaint topic category and customer complaint perception category to which the text data to be processed belongs; The training data for the topic classification model and the perception classification model are generated based on a two-dimensional labeling system. The construction process of the two-dimensional labeling system includes: preprocessing historical customer complaint data to obtain the word segmentation results corresponding to the historical customer complaint data. The word segmentation results are used to extract themes and keywords to obtain the basic themes and keywords corresponding to each of the historical customer complaint data. Each basic topic and each basic keyword is labeled to obtain a three-level topic tag for each basic topic and a three-level perception tag for each basic keyword; Clustering is performed on the third-level topic tags and the third-level perception tags to obtain multiple topic cluster sets and perception cluster sets. The topic tags of multiple similar business scenarios included in each topic cluster set are summarized to obtain the second-level topic tags corresponding to each topic cluster set. Second-level topic tag annotation is performed on each topic cluster set to obtain multiple topic cluster sets with second-level topic tags. Similarly, the perception tags corresponding to multiple similar topics included in each perception cluster set are summarized to obtain the second-level perception tags corresponding to each perception cluster set. Second-level perception tag annotation is performed on each perception cluster set to obtain multiple perception cluster sets with second-level perception tags. Clustering is performed on the secondary topic tags and the secondary perception tags respectively to obtain multiple topic cluster subsets and perception cluster subsets. Tag labeling is performed on the topic cluster subsets and the perception cluster subsets respectively to obtain the primary topic tag corresponding to each topic cluster subset and the primary perception tag corresponding to each perception cluster subset. A two-dimensional tagging system is determined based on the three-level topic tags, the three-level perception tags, the two-level topic tags, the two-level perception tags, the one-level topic tags, and the one-level perception tags.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images