Complaint ticket classification methods, devices, equipment, storage media, and program products

Abstract

This application discloses a method, apparatus, device, storage medium, and program product for classifying complaint work orders, aiming to solve the problems of low processing efficiency, insufficient classification accuracy, and high labor costs caused by the reliance on manual classification of complaint work orders. The method includes: acquiring user complaint text data; inputting the complaint text data into a trained work order classification model, and outputting the classification result of the complaint text data; wherein the work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; the feature extraction layer is used to extract text features at least two scales from the complaint text data, and fuse the text features at least two scales to obtain fused features; the attention layer is used to assign weights to features corresponding to different words in the fused features to obtain weighted features that reflect the differences in word importance in the complaint text data; the classification layer is used to output the classification result based on the weighted features.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, equipment, storage medium, and program product for classifying complaint work orders. Background Technology

[0002] In the daily operations of telecommunications operators, handling complaint tickets is an important part of the customer service system.

[0003] In the existing complaint handling process, when a complaint handler receives a complaint, they typically manually categorize it based on the content of the complaint before forwarding it to the appropriate department for processing. However, due to the diverse and complex nature of complaints, manual categorization is not only time-consuming and labor-intensive, but may also lead to inaccurate results due to subjective factors of the complaint handlers, thus affecting the efficiency of complaint handling. Summary of the Invention

[0004] This application provides a method for classifying complaint work orders to solve the problems of low processing efficiency, insufficient classification accuracy, and high labor costs in the prior art due to the reliance on manual classification of complaint work orders.

[0005] This application also provides a complaint work order classification device, an electronic device, a computer-readable storage medium, and a computer program product.

[0006] The embodiments of this application adopt the following technical solutions: In a first aspect, embodiments of this application provide a method for classifying complaint work orders, including: Obtain user complaint text data; Input the complaint text data into the trained work order classification model and output the classification result of the complaint text data; The work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; The feature extraction layer is used to extract text features at least two scales from the complaint text data and fuse the text features at least two scales to obtain fused features. The attention layer is used to assign weights to the features corresponding to different lexical units in the fused features, so as to obtain weighted features that can reflect the differences in the importance of lexical units in the complaint text data. The classification layer is used to output classification results based on weighted features.

[0007] Optionally, the work order classification model also includes an input layer, which takes the complaint text data as input to the trained work order classification model and outputs the classification results of the complaint text data, including: The complaint text data is input into the input layer to obtain the text vector of the complaint text data; The text vector is input into the feature extraction layer to obtain text features at least two scales, and the text features at least two scales are fused to obtain fused features; The fused features are input into the attention layer, and the word-level attention mechanism of the attention layer assigns weights to the features corresponding to different words in the fused features to obtain weighted features. The weighted features are input into the classification layer to obtain the classification result.

[0008] Optionally, the feature extraction layer includes at least two parallel convolutional neural network structures; Each convolutional neural network structure is configured with convolutional kernels of different sizes to capture text features of different granularities; the feature extraction layer obtains fused features by splicing and fusing the feature maps output by each convolutional neural network structure.

[0009] Optionally, at least two parallel convolutional neural network structures can be used to perform convolution operations on text vectors with 3×3 and 5×5 convolutional kernels, respectively, to capture text features of different granularities.

[0010] Optionally, the feature maps output by each convolutional neural network structure can be concatenated and fused to obtain fused features, including: The feature maps output by each convolutional neural network structure are aligned in the same time dimension and then fused to obtain fused features.

[0011] Optionally, before inputting the complaint text data into the trained work order classification model and outputting the classification result of the complaint text data, the method further includes: Obtain the historical complaint work order dataset, which includes historical complaint texts and the corresponding real category labels for the historical complaint texts; The historical complaint work order dataset is divided into a training set, a validation set, and a test set; Construct an initial work order classification model, which includes at least a feature extraction layer, an attention layer, and a classification layer. The initial work order classification model is iteratively trained based on the training set, the training process is validated based on the validation set, and the iteratively trained work order classification model is tested based on the test set to obtain the trained work order classification model.

[0012] Secondly, embodiments of this application provide a complaint work order classification device, including an acquisition module and a classification module, wherein: The acquisition module is used to acquire the text data of user complaints; The classification module is used to input complaint text data into the trained work order classification model and output the classification results of the complaint text data. The work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; The feature extraction layer is used to extract text features at least two scales from the complaint text data and fuse the text features at least two scales to obtain fused features. The attention layer is used to assign weights to the features corresponding to different lexical units in the fused features, so as to obtain weighted features that can reflect the differences in the importance of lexical units in the complaint text data. The classification layer is used to output classification results based on weighted features.

[0013] Thirdly, embodiments of this application provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the complaint work order classification method as described above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the complaint work order classification method described above.

[0015] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the complaint work order classification method as described above.

[0016] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: The method provided in this application involves inputting complaint text data into a work order classification model that includes a multi-scale feature extraction layer, an attention layer, and a classification layer. In this way, the model can automatically fuse text features at different scales to comprehensively capture semantic information and dynamically focus on keyword elements using an attention mechanism. This enables efficient and accurate automatic classification of complaint work orders, significantly reducing human intervention and subjective errors, and improving complaint processing efficiency. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A schematic diagram illustrating the implementation process of a complaint work order classification method provided in this application embodiment; Figure 2 A schematic diagram of a network structure for work order classification provided in this application embodiment; Figure 3 This application provides a schematic diagram of the specific structure of a complaint work order classification device. Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] As will be known to those skilled in the art, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0020] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0021] It should be understood that the training and prediction processes of the AI ​​models involved in the various embodiments of this specification all adhere to multiple legal and compliant principles, including legal data sources, compliant data content, compliant data governance, compliant training objectives and schemes, compliant training processes, compliant training environments and tools, and compliant ethical verification of training results, and comply with the requirements of Article 5 of the Patent Law. Among them: Data source legitimacy: All datasets used for AI model training were obtained through legal means, covering three categories: publicly authorized data, data authorized by partners, and self-collected compliant data. Publicly authorized data comes from compliant data sources following open-source licenses such as Apache 2.0, with complete copyright attribution and authorization scope clearly marked, and no unauthorized open-source code or data reuse. Data authorized by partners has been subject to formal data usage agreements, clearly defining the scope, duration, and confidentiality obligations, and possessing a complete authorization chain. For self-collected data involving personal information, strict informed consent procedures have been followed, and anonymization processes (including but not limited to field masking, feature anonymization, and differential privacy technology applications) have been implemented to remove personally identifiable information, fully complying with the requirements of relevant laws and regulations such as the "Interim Measures for the Administration of Generative Artificial Intelligence Services" and the "Personal Information Protection Law."

[0022] Data content compliance: The AI ​​model's dataset undergoes multiple screenings and cleaning processes to remove all content that may violate social morality or harm public interests. It contains no obscene, pornographic, violent, discriminatory, or information that endangers national or public safety, nor does it involve the illegal acquisition or use of genetic resources. For data in sensitive fields (such as healthcare and finance), an additional privacy-preserving computation module (including federated learning and secure multi-party computation technologies) ensures that the data is "usable but not visible," avoiding compliance risks during the original data transmission process and ensuring that the data application scenarios and uses comply with public order and good morals and industry regulatory requirements.

[0023] Data governance norms: A complete data traceability system is established during the AI ​​model training process to automatically record the source, collection time, annotation process, cleaning rules, and permission allocation of training data, generating traceable compliance reports to ensure that the data is verifiable throughout its entire lifecycle. The dataset annotation process for AI models is completed by a professional human R&D team, clearly defining the proportion of human creative contributions and avoiding reliance on AI-generated data that has not undergone substantial human modification, thus meeting the examination requirements for "human main contributions" in AI patent applications.

[0024] Training objectives and plans are compliant: The AI ​​model training objective focuses on classifying and intelligently assigning and transferring work orders to responsible departments for telecommunications operator complaint texts, aiming to improve complaint handling efficiency and service quality. The training scheme and final output results do not violate any mandatory provisions of laws and administrative regulations, do not harm the public interest or the legitimate rights and interests of others, and do not pose any potential risks of being used for illegal activities, privacy infringement, or public safety disruption. It strictly adheres to the ethical principle of "intelligent for good".

[0025] Training process compliance: A closed-loop training framework is adopted to ensure compliance and controllability of the training process. The specific process is as follows: First, training samples are obtained through compliant data sources. After the aforementioned data cleaning and desensitization, they are input into the neural network model to generate preliminary training results. Second, an expert system is introduced to verify the preliminary results. Based on preset rules and human expert experience, the feasibility of the results is evaluated, and outputs that may pose ethical risks or compliance hazards are corrected (such as removing decision-making logic that violates public order and good morals, and adjusting model parameters that do not comply with safety regulations). Finally, the loss function weights are dynamically optimized based on expert system feedback to strengthen the model's learning of compliant results, avoid overfitting errors or non-compliant labels, and form a closed-loop control of "data input - model training - expert verification - parameter optimization - result feedback" to ensure that the entire training process complies with A5 ethical review requirements.

[0026] Training environment and tool compliance: AI model training is implemented using nationally licensed chips and a compliant training platform. All open-source frameworks and components used in the training process have obtained their corresponding licenses, and copyright statements and patent citation information are fully retained, with no instances of infringement or reuse. The training environment is built using virtual devices (containers / virtual machines) with fixed random seeds and initial parameter configurations to ensure the reproducibility of the training process. Furthermore, through access control and operation log recording, risks such as data leakage and parameter tampering during training are prevented, ensuring the security and compliance of the training process.

[0027] Training results ethical verification compliance: After the model is trained, it undergoes additional third-party ethical compliance assessment and algorithm filing review to verify that the model output does not violate social morality or harm public interests. For potentially sensitive scenarios (such as public services and intelligent decision-making), a special result verification mechanism is established to ensure that the model always complies with Article 5 of the Patent Law and relevant laws and regulations in practical applications.

[0028] In summary, the data and training process used in the AI ​​model of this specification strictly comply with the relevant provisions of Article 5 of the Patent Law and the Patent Examination Guidelines (2023 Edition), and there are no violations of laws, social ethics, public interests, or illegal use of genetic resources. It fully meets the compliance requirements for patent authorization.

[0029] Example 1 To address the problems of low processing efficiency, insufficient classification accuracy, and high labor costs caused by the reliance on manual classification of complaint work orders in existing technologies, this application provides a complaint work order classification method.

[0030] The execution subject of this method can be various types of computing devices, or it can be an application or app installed on the computing device. The computing device can be a user terminal such as a mobile phone, tablet computer, or smart wearable device, or it can be a server.

[0031] For ease of description, this application uses a server as the execution subject of the method in its embodiments to illustrate the method. Those skilled in the art will understand that this embodiment uses a server as an example to describe the method, which is merely an illustrative example and does not limit the scope of protection of the corresponding claims.

[0032] Specifically, the implementation flow of the method provided in this application embodiment is as follows: Figure 1 As shown, it includes the following steps: Step 102: Obtain the user's complaint text data.

[0033] In this embodiment of the application, complaint text data refers to natural language text content that can represent user complaint matters. User complaint text data can be obtained through methods such as customer service hotline transcription, APP / webpage forms, WeChat official account messages, and content transferred from the Ministry of Industry and Information Technology / 12345.

[0034] In some embodiments, when obtaining user complaint text data, metadata such as work order number, channel identifier, generation time, and city of origin can also be associated to facilitate subsequent tracking and statistical analysis.

[0035] Optionally, in some embodiments, in order to improve the classification stability of the work order classification model, the complaint text data can be standardized after obtaining it. For example, the full-width / half-width characters can be unified, meaningless repeated symbols can be deleted, and sensitive personal fields such as mobile phone number / ID number / address can be replaced with placeholders to reduce the risk of privacy leakage and reduce noise interference.

[0036] Optionally, in some embodiments, in order to enhance the ability of the feature extraction layer of the subsequent work order classification model to focus on semantic information, the complaint text data can be further preprocessed. This preprocessing can include word segmentation, stop word removal, category field matching, or synonym grouping. For example, disconnection / network outage / inability to access the internet can be grouped into the same semantic cluster.

[0037] Step 104: Input the complaint text data into the trained work order classification model and output the classification result of the complaint text data.

[0038] In this embodiment of the application, the completed work order classification model refers to a deep learning text classification model that has completed parameter learning using historical complaint work order samples and can be directly used for online inference.

[0039] Optionally, the classification results may include target category labels (such as communication quality / cost / service processing, etc.), category confidence (probability value), and an optional Top-K candidate category list.

[0040] The work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; The feature extraction layer is used to extract text features at least two scales from the complaint text data and fuse these features to obtain fused features. In this embodiment, scale can be understood as the context range covered by the feature extraction window, such as the length of the n-gram or the number of lemmas covered by the convolution kernel. Different scales are used to capture semantic patterns of different granularities. Smaller scales are better at extracting local keyword combinations, such as "weak signal" or "unable to access the internet," while larger scales are better at extracting phrase-level patterns across lemmas, such as "failed to process business."

[0041] In one implementation, the feature extraction layer can be implemented using at least two parallel convolutional neural network structures. Multiple feature maps are obtained by performing convolution operations on the text vector of the complaint text using different convolutional kernel sizes in parallel, and the feature maps are then fused, such as by splicing and fusion to form a fused feature.

[0042] Optionally, at least two parallel convolutional neural network structures can use 3×3 and 5×5 convolutional kernels respectively to perform convolution operations on the text vectors to capture text features of different granularities. Among them, the 3x3 convolutional kernel can capture relatively local and detailed features in the complaint text data, while the 5x5 convolutional kernel focuses on extracting more global and comprehensive features from the complaint text data.

[0043] Optionally, to improve robustness to complaint text data, the word sequence of the complaint text data can be padded to align the length before the feature extraction layer performs convolution operations on the text vector; after convolution, non-linear activations, such as ReLU and pooling (e.g., max pooling or k-max pooling), can be configured to compress redundancy and enhance discriminability.

[0044] The attention layer is used to assign weights to the features corresponding to different tokens in the fused features, so as to obtain weighted features that can reflect the differences in the importance of tokens in the complaint text data. Among them, a token refers to the smallest semantic unit (character, word or sub-word) obtained after word segmentation, and each token corresponds to a vector representation with a fixed dimension.

[0045] In this embodiment, the attention layer employs a token-level attention mechanism. This mechanism calculates weight coefficients at the token level, allowing the work order classification model to assign different contributions to different tokens (or their corresponding convolutional output positions / channels) when forming the final representation. Specifically, an attention score is calculated for each token position in the fused features, then normalized (e.g., using Softmax normalization) to obtain a weight vector. Finally, the fused features are weighted and aggregated to output a weighted feature. This mechanism enables keywords containing core semantics to receive higher weights and suppresses interference from irrelevant information.

[0046] Optionally, the attention score can be obtained by taking the inner product of a lightweight feedforward network or a learnable parameter vector with the fused features; or positional encoding can be introduced into the attention calculation to distinguish word order information.

[0047] A classification layer is used to output classification results based on weighted features. In this embodiment, the classification layer may include a fully connected layer and a probability normalization module. Specifically, the weighted features can be input into the fully connected layer to obtain the logits of each category, and then the logits can be converted into probability values ​​for each category using the Softmax function. A probability vector is then output, and the category label is determined accordingly; for example, the category with the highest probability is taken as the classification result.

[0048] Optionally, to facilitate integration with business systems, the classification results can be further mapped to the responsible department identifier or work order routing rules. For example, if the classification result is a tariff issue, the output can also include the department identifier of the billing / package center or the work order routing rules; or, if the classification result is a communication quality issue, the output result can include the department identifier of the network optimization center or the work order routing rules. This mapping relationship is maintained in the form of a configuration table for easy updating when adding or deleting categories in the future.

[0049] The following describes the specific implementation process of step 104 in detail with examples.

[0050] In one implementation, such as Figure 2 As shown, the work order classification model includes an input layer in addition to the feature extraction layer, attention layer, and classification layer. The input layer is used to convert the complaint text data in natural language form into a numerical vector representation that can be processed by the neural network, thus forming the input for subsequent layers.

[0051] Specifically, when inputting complaint text data into the input layer, the complaint text is first segmented into words to obtain a sequence of multiple word units. Then, the word unit sequence is mapped to text vectors, which can be represented as a vector sequence formed by multiple word vectors arranged in order, also known as a word vector sequence. In one optional approach, the word vectors can be initialized with pre-trained word vectors and dynamically fine-tuned during the training phase to balance general semantics with the adaptability of operator-specific corpora. Secondly, for out-of-vocabulary words, UNK placeholder vectors or sub-word splitting can be used to generate vectors to reduce the impact of out-of-vocabulary words on classification accuracy.

[0052] After obtaining the text vector, it can be input into the feature extraction layer. The feature extraction layer is used to extract text features at at least two scales. The feature extraction layer can employ parallel multi-scale convolution operations to extract local semantic features at different scales, and then fuse the features from at least two scales to obtain fused features. The fusion method can be concatenation, weighted summation, or other methods that can integrate multi-scale information.

[0053] After the fused features are input into the attention layer, the attention layer calculates the weight coefficients of the features corresponding to different words in the fused features through a word-level attention mechanism, and then performs weighted processing on the fused features according to the weights, thereby obtaining weighted features that can reflect the differences in word importance. In this way, the proportion of features corresponding to keywords or key phrases related to complaint category determination can be increased in the weighted features, thereby enhancing the category discrimination ability and suppressing the interference of noisy words.

[0054] Finally, the weighted features are input into the classification layer, which performs classification calculations on the weighted features and outputs the classification results. The classification layer may include a fully connected layer and a Softmax processing module to output the probability value corresponding to each complaint category, and determine the category with the highest probability as the classification result; in an optional implementation, Top-K candidate categories and their probabilities may also be output for subsequent manual review or routing strategy selection.

[0055] For example, suppose we take a complaint ticket classification task from a certain operator as an example. There are 10 categories, the maximum sequence length is set to L=64, and the word vector dimension is d=200.

[0056] Input layer: The complaint text "Poor mobile signal, frequent disconnections, troublesome to handle" is segmented into words, resulting in a word sequence such as: [mobile, signal, poor, frequent, disconnections, troublesome, handle]. Each word is mapped to a 200-dimensional word vector, and the sequence is length-aligned: words shorter than 64 are padded at the end with PADs, resulting in a 64×200 text vector matrix X.

[0057] Feature extraction layer: Two parallel convolutional branches are set, with convolutional window lengths of 3 and 5 respectively, and each branch has 128 convolutional channels.

[0058] The convolutional window with a length of 3 outputs a 128-dimensional feature vector f3; The convolution window has a length of 5, and the window branches output a 128-dimensional feature vector f5. By splicing and fusing the two, we obtain a 256-dimensional fusion feature F=[f3;f5].

[0059] Attention layer: Attention weight vectors are calculated for each word position, and the fused features are weighted and converged to obtain a 256-dimensional weighted feature Z. For example, the weights for "signal" and "disconnected" are 0.18 and 0.22 respectively, which are higher than those for other positions, thus highlighting the key semantics.

[0060] Classification layer: Z is input into a fully connected layer and output as a 10-dimensional probability vector p via Softmax. For example, the output might be: Communication quality problem: 0.72; Pricing problem: 0.06; Service processing problem: 0.05; Probabilities for other categories are all less than 0.05. In this case, "Communication quality problem" is determined as the classification result for the complaint text, and a confidence score of 0.72 can be simultaneously output for subsequent routing / review strategies.

[0061] In one implementation, to obtain a work order classification model for classifying complaint work orders, an initial work order classification model needs to be trained before inputting the complaint text data into the trained work order classification model. The process for training the initial work order classification model is as follows: First, obtain the historical complaint ticket dataset. This dataset could be, for example, complaint ticket data from a provincial telecom operator in 2024, containing historical complaint texts and their corresponding category labels. The total sample size could be, for example, 73,000 entries, with 10 categories. Category labels could include issues such as communication quality, pricing, service processing, service attitude, and network failures.

[0062] Among them, the true category label refers to the target category identifier that has been confirmed by the business system, quality inspectors or existing rule system in the historical work order processing, and is used as the training target for supervised learning.

[0063] Secondly, before training the model using the dataset, the historical complaint texts can be preprocessed. Preprocessing is a crucial step in the training process, aiming to reduce noise interference and improve training stability and classification accuracy. Preprocessing includes at least three steps: word segmentation, stop word removal, and category matching. 1) Word segmentation: Using jieba or equivalent word segmentation methods, the complaint text is divided into several word units, where a word unit refers to the smallest semantic unit used for model processing (which can be a word / character / sub-word) to form a word unit sequence, laying the foundation for subsequent vectorization and feature extraction; 2) Stop word removal: Based on the preset stop word list and filtering rules, remove modal particles, conjunctions or meaningless symbols that do not contribute to the classification, such as “ah”, “ya”, “ne”, etc., so that the model can focus more on the effective key information in the complaint text, and improve training efficiency and classification accuracy. 3) Category matching: The category field in historical complaint work orders is made consistent. For example, synonyms / near-synonyms of category names, such as abnormal charges / unauthorized charges / billing issues, are merged into a unified standard label, or the multi-level categories in the historical system are mapped to a preset set of 10 labels to improve label consistency and reduce the interference of noisy labels on training.

[0064] Then, the historical complaint work order dataset is divided into a training set, a validation set, and a test set. For example, a 7:2:1 ratio can be used, where the training set is used for iterative learning of model parameters, the validation set is used for performance verification and hyperparameter tuning during training, and the test set is used for independent evaluation of the model's generalization ability after training. Optionally, stratified sampling can be used during the partitioning to maintain a relatively consistent sample proportion for each category in different subsets.

[0065] Next, an initial work order classification model is constructed. The initial work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer: the feature extraction layer is used to extract text features at least two scales and fuse them to obtain fused features; the attention layer is used to assign weights to features corresponding to different words in the fused features through a word-level attention mechanism to obtain weighted features; the classification layer is used to output the predicted category distribution based on the weighted features and form the classification result.

[0066] Finally, the initial work order classification model is iteratively trained based on the training set, and the training process is validated based on the validation set and the iteratively trained model is tested based on the test set to obtain the trained work order classification model. Iterative training may include inputting training samples in batches, outputting prediction results, calculating the training error between the prediction results and the true labels, and updating the model parameters; the training process may set stopping conditions, such as the validation set loss no longer decreasing, the validation set accuracy improvement being less than a threshold, or the number of training epochs reaching the upper limit; after training, the accuracy, macro-average F1, and other metrics are output using the test set to confirm the model's deployability and generalization performance.

[0067] The method provided in this application involves inputting complaint text data into a work order classification model that includes a multi-scale feature extraction layer, an attention layer, and a classification layer. This allows the model to automatically fuse text features at different scales to comprehensively capture semantic information and dynamically focus on keyword elements using an attention mechanism. As a result, efficient and accurate automatic classification of complaint work orders can be achieved, significantly reducing human intervention and subjective errors and improving complaint processing efficiency.

[0068] Example 2 To address the problems of low processing efficiency, insufficient classification accuracy, and high labor costs caused by the reliance on manual classification of complaint work orders in existing technologies, this application provides a complaint work order classification device, the specific structure of which is shown in the schematic diagram below. Figure 3 As shown, it includes an acquisition module 31 and a classification module 32. The functions of each module are as follows: Module 31 is used to acquire the user's complaint text data; Classification module 32 is used to input complaint text data into the trained work order classification model and output the classification result of the complaint text data; The work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; The feature extraction layer is used to extract text features at least two scales from the complaint text data and fuse the text features at least two scales to obtain fused features. The attention layer is used to assign weights to the features corresponding to different lexical units in the fused features, so as to obtain weighted features that can reflect the differences in the importance of lexical units in the complaint text data. The classification layer is used to output classification results based on weighted features.

[0069] Optional, classification module 32, used for: The complaint text data is input into the input layer to obtain the text vector of the complaint text data; The text vector is input into the feature extraction layer to obtain text features at least two scales, and the text features at least two scales are fused to obtain fused features; The fused features are input into the attention layer, and the word-level attention mechanism of the attention layer assigns weights to the features corresponding to different words in the fused features to obtain weighted features. The weighted features are input into the classification layer to obtain the classification result.

[0070] Optionally, the feature extraction layer includes at least two parallel convolutional neural network structures; Each convolutional neural network structure is configured with convolutional kernels of different sizes to capture text features of different granularities; the feature extraction layer obtains fused features by splicing and fusing the feature maps output by each convolutional neural network structure.

[0071] Optionally, at least two parallel convolutional neural network structures can be used to perform convolution operations on text vectors with 3×3 and 5×5 convolutional kernels, respectively, to capture text features of different granularities.

[0072] Optionally, the feature extraction layer aligns the feature maps output by each convolutional neural network structure in the same time dimension and then fuses them to obtain fused features.

[0073] Optionally, the device also includes a training module for: Before inputting the complaint text data into the trained work order classification model and outputting the classification result of the complaint text data, obtain the historical complaint work order dataset, which includes historical complaint texts and the corresponding true category labels of the historical complaint texts. The historical complaint work order dataset is divided into a training set, a validation set, and a test set; Construct an initial work order classification model, which includes at least a feature extraction layer, an attention layer, and a classification layer. The initial work order classification model is iteratively trained based on the training set, the training process is validated based on the validation set, and the iteratively trained work order classification model is tested based on the test set to obtain the trained work order classification model.

[0074] Using the device provided in this application embodiment, complaint text data is input into a work order classification model containing a multi-scale feature extraction layer, an attention layer, and a classification layer for processing. In this way, the model can automatically fuse text features of different scales to comprehensively capture semantic information and dynamically focus on keyword elements using an attention mechanism. This enables efficient and accurate automatic classification of complaint work orders, significantly reducing human intervention and subjective errors, and improving complaint processing efficiency.

[0075] Example 3 Figure 4 To illustrate the hardware structure of an electronic device according to various embodiments of this application, the electronic device 400 includes, but is not limited to, components such as: a radio frequency unit 401, a network module 402, an audio output unit 403, an input unit 404, a sensor 405, a display unit 406, a user input unit 407, an interface unit 408, a memory 409, a processor 410, and a power supply 411. Those skilled in the art will understand that... Figure 4 The electronic device structures shown are not intended to limit the electronic device. An electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. In the embodiments of this application, the electronic device includes, but is not limited to, mobile phones, tablets, laptops, PDAs, in-vehicle terminals, wearable devices, and pedometers.

[0076] The processor 410 is used to acquire user complaint text data; input the complaint text data into a trained work order classification model, and output the classification result of the complaint text data; wherein the work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; the feature extraction layer is used to extract text features of at least two scales from the complaint text data, and fuse the text features of at least two scales to obtain fused features; the attention layer is used to assign weights to the features corresponding to different words in the fused features to obtain weighted features that can reflect the differences in word importance of the complaint text data; the classification layer is used to output the classification result based on the weighted features.

[0077] The memory 409 is used to store a computer program that can run on the processor 410, which, when executed by the processor 410, implements the functions described above by the processor 410.

[0078] It should be understood that, in this embodiment, the radio frequency unit 401 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink data from the base station and processes it with the processor 410; additionally, it transmits uplink data to the base station. Typically, the radio frequency unit 401 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier, a duplexer, etc. Furthermore, the radio frequency unit 401 can also communicate with networks and other devices through a wireless communication system.

[0079] The electronic device provides users with wireless broadband internet access through the network module 402, such as helping users send and receive emails, browse web pages, and access streaming media.

[0080] The audio output unit 403 can convert audio data received by the radio frequency unit 401 or the network module 402 or stored in the memory 409 into audio signals and output them as sound. Furthermore, the audio output unit 403 can also provide audio output related to specific functions performed by the electronic device 400 (e.g., call signal reception sound, message reception sound, etc.). The audio output unit 403 includes a speaker, a buzzer, and a receiver, etc.

[0081] Input unit 404 is used to receive audio or video signals. Input unit 404 may include a graphics processing unit (GPU) 4041 and a microphone 4042. The GPU 4041 processes image data of still images or videos acquired by an image capture device (such as a camera) in video capture mode or image capture mode. The processed image frames can be displayed on display unit 406. The image frames processed by GPU 4041 can be stored in memory 409 (or other storage medium) or transmitted via radio frequency unit 401 or network module 402. Microphone 4042 can receive sound and process such sound into audio data. The processed audio data can be converted into a format that can be transmitted to a mobile communication base station via radio frequency unit 401 in telephone call mode.

[0082] The electronic device 400 also includes at least one sensor 405, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor includes an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 4061 according to the ambient light level, and the proximity sensor can turn off the display panel 4061 and / or backlight when the electronic device 400 is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used to identify the posture of the electronic device (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc. The sensor 405 may also include a fingerprint sensor, pressure sensor, iris sensor, molecular sensor, gyroscope, barometer, hygrometer, thermometer, infrared sensor, etc., which will not be described in detail here.

[0083] The display unit 406 is used to display information input by the user or information provided to the user. The display unit 406 may include a display panel 4061, which may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), or the like.

[0084] User input unit 407 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of electronic devices. Specifically, user input unit 407 includes a touch panel 4071 and other input devices 4072. Touch panel 4071, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near touch panel 4071). Touch panel 4071 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 410, which receives and executes commands from the processor 410. In addition, touch panel 4071 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. Besides touch panel 4071, user input unit 407 may also include other input devices 4072. Specifically, other input devices 4072 may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, power buttons, etc.), trackballs, mice, joysticks, etc., which will not be described in detail here.

[0085] Furthermore, the touch panel 4071 can cover the display panel 4061. When the touch panel 4071 detects a touch operation on or near it, it transmits the information to the processor 410 to determine the type of touch event. Subsequently, the processor 410 provides corresponding visual output on the display panel 4061 based on the type of touch event. Although in Figure 4 In this embodiment, the touch panel 4071 and the display panel 4061 are two independent components to realize the input and output functions of the electronic device. However, in some embodiments, the touch panel 4071 and the display panel 4061 can be integrated to realize the input and output functions of the electronic device. The specific implementation is not limited here.

[0086] Interface unit 408 serves as an interface for connecting external devices to electronic device 400. For example, external devices may include a wired or wireless headphone port, an external power supply (or battery charger) port, a wired or wireless data port, a memory card port, a port for connecting a device with an identification module, an audio input / output (I / O) port, a video I / O port, a headphone port, and so on. Interface unit 408 can be used to receive input (e.g., data, power, etc.) from external devices and transmit the received input to one or more components within electronic device 400, or it can be used to transmit data between electronic device 400 and external devices.

[0087] The memory 409 can be used to store software programs and various data. The memory 409 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback, image playback, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory 409 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0088] The processor 410 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 409, and by calling data stored in the memory 409, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. The processor 410 may include one or more processing units; preferably, the processor 410 may integrate an application processor and a modem processor. The application processor mainly handles the operating system, user interface, and applications, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 410.

[0089] The electronic device 400 may also include a power supply 411 (such as a battery) that supplies power to various components. Preferably, the power supply 411 can be logically connected to the processor 410 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system.

[0090] In addition, the electronic device 400 includes some functional modules not shown, which will not be described in detail here.

[0091] Preferably, this application embodiment also provides an electronic device, including a processor 410, a memory 409, and a computer program stored in the memory 409 and executable on the processor 410. When the computer program is executed by the processor 410, it implements the various processes of the above-described complaint work order classification method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0092] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described complaint work order classification method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0093] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0095] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0096] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0097] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0098] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0099] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0100] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0101] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for classifying complaint work orders, characterized in that, include: Obtain user complaint text data; The complaint text data is input into the trained work order classification model, and the classification result of the complaint text data is output. The work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; The feature extraction layer is used to extract text features at least two scales from the complaint text data, and to fuse the text features at least two scales to obtain fused features; The attention layer is used to assign weights to the features corresponding to different lexical units in the fused features, so as to obtain weighted features that can reflect the differences in the importance of lexical units in the complaint text data. The classification layer is used to output the classification result based on the weighted features.

2. The method as described in claim 1, characterized in that, The work order classification model further includes an input layer, wherein the process of inputting the complaint text data into the trained work order classification model and outputting the classification result of the complaint text data includes: The complaint text data is input into the input layer to obtain the text vector of the complaint text data; The text vector is input into the feature extraction layer to obtain text features at least two scales, and the text features at least two scales are fused to obtain the fused features; The fused features are input into the attention layer, and the word-level attention mechanism of the attention layer assigns weights to the features corresponding to different words in the fused features to obtain the weighted features; The weighted features are input into the classification layer to obtain the classification result.

3. The method as described in claim 1 or 2, characterized in that, The feature extraction layer includes at least two parallel convolutional neural network structures; Each of the convolutional neural network structures is configured with convolutional kernels of different sizes to capture text features of different granularities; the feature extraction layer obtains the fused features by splicing and fusing the feature maps output by each of the convolutional neural network structures.

4. The method as described in claim 3, characterized in that, The at least two parallel convolutional neural network structures use 3×3 and 5×5 convolutional kernels respectively to perform convolution operations on the text vector to capture text features of different granularities.

5. The method as described in claim 3 or 4, characterized in that, The feature maps output by each of the convolutional neural network structures are concatenated and fused to obtain the fused features, including: The feature maps output by each of the convolutional neural network structures are aligned in the same time dimension and then fused to obtain the fused feature.

6. The method according to any one of claims 1 to 5, characterized in that, Before inputting the complaint text data into the trained work order classification model and outputting the classification result of the complaint text data, the method further includes: Obtain a historical complaint work order dataset, which includes historical complaint texts and the corresponding real category labels of the historical complaint texts; The historical complaint work order dataset is divided into a training set, a validation set, and a test set; Construct an initial work order classification model, which includes at least a feature extraction layer, an attention layer, and a classification layer; The initial work order classification model is iteratively trained based on the training set, and the training process is verified based on the validation set and the iteratively trained work order classification model is tested based on the test set to obtain the completed work order classification model.

7. A complaint work order classification device, characterized in that, This includes an acquisition module and a classification module, wherein: The acquisition module is used to acquire the text data of user complaints; The classification module is used to input the complaint text data into the trained work order classification model and output the classification result of the complaint text data; The work order classification model includes at least a feature extraction layer, an attention layer, and a classification layer; The feature extraction layer is used to extract text features at least two scales from the complaint text data, and to fuse the text features at least two scales to obtain fused features; The attention layer is used to assign weights to the features corresponding to different lexical units in the fused features, so as to obtain weighted features that can reflect the differences in the importance of lexical units in the complaint text data. The classification layer is used to output the classification result based on the weighted features.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the complaint work order classification method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the complaint work order classification method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the complaint ticket classification method according to any one of claims 1 to 6.

Images