Deep learning model-based work order classification method and system

Work order classification through deep learning models solves the time consumption and accuracy problems caused by artificial dependence in traditional methods, and achieves efficient and accurate automated classification, reducing costs and improving the scalability and self-optimization capabilities of the system.

CN118964618APending Publication Date: 2024-11-15SHANDONG LANGCHAO YUNTOU INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411156211.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-22
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Traditional work order classification methods rely on manual labor, resulting in large time consumption, poor accuracy, limited scalability, and inefficient data utilization, which affects operation and maintenance efficiency and customer satisfaction.

Method used

The work order classification method and system based on deep learning models are adopted to realize the automated classification of work orders through data preprocessing, feature type coding, embedding calculation, model coding and business classification matching.

Benefits of technology

It significantly reduces the work order processing time, improves classification accuracy and efficiency, reduces human resources costs, enhances the scalability and self-optimization capabilities of the system, and improves user experience and resource utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964618A_ABST
    Figure CN118964618A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of computers, in particular to a work order classification method and system based on a deep learning model, and the method comprises the following steps: data preprocessing and word segmentation are carried out, and text data in a work order are processed; word segmentation processing is carried out on the cleaned text, token sequences of work order titles, question description, creator information, areas and product line fields are obtained, [SEP] marks are used for separating the fields, finally, a single input sequence is constructed, and the sequence starts with a special mark [CLS]; the work order classification method and system based on the deep learning model have the beneficial effects that a large number of work orders can be quickly processed through an advanced technical framework, the work order processing time is remarkably shortened, and the work order classification speed and efficiency are greatly optimized. Compared with a traditional manual classification method, the system not only accelerates the work order processing speed, but also ensures timely processing of various work order requests through an intelligent mode.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a work order classification method and system based on a deep learning model. Background Art

[0002] Work order classification is an important part of information technology management, which is directly related to the efficiency and service quality of the operation and maintenance team. In the traditional model, work order classification usually relies on the individual submitting the work order to select a preset business classification list. This manual classification method has many limitations:

[0003] (1) Time consumption: Manual ticket classification is usually a time-consuming process, especially when the number of tickets is large or the problem complexity is high. The time-consuming classification process may lead to overall response delays, affecting service efficiency and customer satisfaction.

[0004] (2) Accuracy: The accurate classification of work orders depends largely on the experience and judgment of the operator. Different personnel may make different classification decisions for the same work order due to differences in understanding and judgment. The influence of personal experience and prejudice may result in the work order not being accurately classified into the most appropriate business category. This inaccurate classification may lead to improper resource allocation and increase the complexity and difficulty of subsequent processing.

[0005] (3) Poor scalability: As the business expands and becomes more complex, new business categories continue to emerge, which limits the scalability and flexibility of traditional manual classification methods and makes it difficult to quickly adapt to new business needs.

[0006] (4) Inefficient data utilization: Traditional methods fail to effectively utilize historical work order data to optimize the classification process. A large amount of valuable data is not used for analysis and improvement, missing the opportunity to improve efficiency and accuracy through data-driven decision-making.

[0007] Limited user experience: User experience may be negatively impacted due to delayed processing times and classification errors. In the current market environment where customer service requirements are increasingly high, this may lead to decreased customer satisfaction and even customer churn. Summary of the invention

[0008] The purpose of the present invention is to provide a work order classification method and system based on a deep learning model to solve the problems raised in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solution: a work order classification method based on a deep learning model, the method comprising the following steps:

[0010] Data preprocessing and word segmentation: Clean and standardize the text data in the work order, including the work order title, problem description, creator information, region, and product line, remove HTML tags, special symbols, and non-printing characters, and unify the character set and uppercase and lowercase characters;

[0011] The cleaned text is segmented to obtain the token sequence of the work order title, problem description, creator information, region, and product line fields, and the [SEP] tag is used to separate the fields. Finally, a single input sequence is constructed, and the sequence starts with a special tag [CLS].

[0012] Preferably, the method further includes a feature type encoding step:

[0013] Build a feature type dictionary: assign unique identifiers to all possible input feature types, such as ticket title, problem description, creator information, region, product line, and special tags;

[0014] Feature type embedding: Using a learnable embedding layer, feature type IDs are converted into fixed-length dense vectors, which are updated during model training to express the semantic and grammatical properties of each feature.

[0015] Preferably, the process further includes a work order characterization step:

[0016] Embedding calculation: Token embedding, feature type encoding and position encoding are performed on each token in the input sequence, and the three embedding vectors are added element by element in the same dimensional space to form the final embedding vector;

[0017] Model encoding: The final embedding vector is input into the representation model encoding layer for n calculations to capture the deep semantic information of the text.

[0018] Preferably, the method further includes a business classification matching step:

[0019] Classification probability calculation: Extract the vector at a specific position from the model output as the feature vector of the work order, and calculate the probability distribution of each business classification corresponding to the work order through a feedforward neural network;

[0020] Classification result determination: During inference in the production stage, the business classification ID whose probability exceeds the set threshold is returned. If there is no classification that meets the conditions, a null value is returned and the user is prompted that the automatic classification has failed.

[0021] Preferably, the method further includes a model training step:

[0022] Training process: Use the labeled work order data to train the model, and use the back propagation algorithm to optimize the parameters of the embedding layer and the model encoding layer to improve the accuracy and efficiency of work order classification;

[0023] Model optimization: Adjust the model structure, embedding vector dimension, and training parameters based on the training results to further improve the performance of the model.

[0024] A work order classification system based on a deep learning model, the system comprising:

[0025] The work order intelligent assistant, as the main interface for users to interact with the system, is integrated into the work order system's order creation page. It is used to receive the work order information entered by the user, convert the information into a structured model input format, send it to the back-end work order classification model, and present the model's output results to the user;

[0026] The work order classification model is deployed on a high-performance server and uses deep learning and natural language processing technology to accurately analyze the received work order information and realize automatic classification of work orders.

[0027] Preferably, the work order intelligent assistant includes the following functions:

[0028] The user input receiving function receives the work order information entered by the user through an intuitive and easy-to-use work order creation page, including the work order title, problem description, creator information, region and product line, and supports multiple input methods;

[0029] Information parsing and structuring functions use natural language processing technology to parse input text, extract key information, and convert it into structured data that can be processed by the model;

[0030] The classification request processing function sends the parsed and structured data to the back-end work order classification model for processing;

[0031] The result verification and feedback function verifies the output results of the classification model, queries the database to obtain detailed business process information, and finally presents the results to the user;

[0032] Multi-level classification information processing function supports multi-level classification processing of complex work orders to ensure data integrity and accuracy;

[0033] The data backfill and user feedback function automatically fills the classification results back to the order creation page, allowing users to view and modify the pre-classification results, and provides a user feedback mechanism to optimize the model.

[0034] Preferably, the work order classification model includes:

[0035] The work order representation model is a pre-trained language model based on the Bert architecture. It is used to extract and understand the semantic information of the work order text and embed features of various attributes of the work order in combination with the feature type embedding module.

[0036] The business classification matching model is built based on a feedforward neural network. It uses the feature vector sequence output by the work order representation model as input and calculates the probability distribution of each business classification through neural network layer processing, thereby realizing the automatic classification of work orders.

[0037] Preferably, the user input receiving function further includes:

[0038] Front-end input verification function to ensure the correctness of the data format entered by the user, such as verifying the email format and phone number, to reduce invalid requests processed by the back-end;

[0039] User behavior analysis function is used to optimize interface layout and improve the speed and accuracy of users filling in information.

[0040] Preferably, the result verification and feedback function also includes:

[0041] Result verification rule setting function to ensure the legality and accuracy of model output;

[0042] The business logic layer builds functions to process data query and integration, and provide complete business information for front-end display;

[0043] The user interaction design function allows users to easily modify classification results through reasonable user interaction design, and realizes real-time data binding between the front and back ends to ensure timely response to user operations.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The work order classification method and system based on the deep learning model proposed in the present invention can quickly process a large number of work orders through its advanced technical framework, significantly reduce the time of work order processing, and greatly optimize the speed and efficiency of work order classification. Compared with the traditional manual classification method, this system not only speeds up the speed of work order processing, but also ensures the timely processing of various work order requests in an intelligent way.

[0046] The introduction of deep learning models allows the system to self-learn and self-optimize based on a large amount of historical data. This learning ability ensures that work orders can be accurately and intelligently classified into appropriate system business categories, significantly reducing classification errors caused by human judgment errors. Compared with traditional methods, this automated system reduces the reliance on employees' ability to understand business classification, thereby reducing human resource costs and improving resource allocation and utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a system architecture diagram of the present invention;

[0048] Figure 2 This is a structural diagram of the work order classification model of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Embodiment 1, the present invention provides a technical solution: a work order classification method based on a deep learning model, the method comprising the following steps:

[0051] (1) Model input construction: The system needs to preprocess various text data in the work order, including fields such as work order title, problem description, creator information, region and product line. The main purpose of preprocessing is to clean and standardize the data to ensure that the data input into the model is clean and consistent. Text cleaning requires removing noise, such as HTML tags, special symbols, non-printing characters, etc. Standardized text converts all texts into a unified character set (such as UTF-8), unifies uppercase and lowercase, and replaces or deletes numbers and special characters. For a given work order data object, the text content of the work order title, problem description, creator information, region and product line fields is word segmented, which is the process of breaking down continuous text strings into meaningful units (vocabulary or tokens). Word segmentation is not just a simple separation by space, but also needs to consider language characteristics, such as abbreviations in English and space-free writing in Chinese. After word segmentation, the token sequence T is obtained 标题 , T 描述 , T 创建 , T 区域 and T 产品 . Concatenate all preprocessed and tokenized field text sequences to form a single input sequence. The order and manner of concatenation are critical to the model's understanding. Generally, the most important information is placed first, such as the problem description, followed by the work order title, and then other auxiliary information such as the creator, region, and product line. This order can help the model capture key information more effectively. Use the [SEP] tag at the end of each field, except for the last field. Doing so helps the model build a clear information boundary internally. Concatenate the token sequence as the model's input X = {[CLS], T 标题 ,[SEP],T 描述 ,[SEP],T 创建 ,[SEP],T 区域 ,[SEP],T 产品}, where [CLS] is a special marker for the head of the input sequence, and [SEP] is a special marker for separating feature sequences of different fields.

[0052] (2) Feature type encoding: Feature type encoding is a crucial step in building an efficient and accurate work order intelligent classification model, which allows the model to properly process and understand various input features. This part is mainly composed of a learnable embedding layer and a corresponding feature type dictionary. Through these components, the model can better process and parse different types of input data. First, the system needs to build a feature type dictionary, which includes all possible input feature types, such as work order title, work order problem description, work order creator information, region, product line, and special tags for sequence construction such as [CLS] and [SEP]. This dictionary assigns a unique identifier (ID) to each feature, which will be used in the subsequent encoding process. The feature type embedding module includes an embedding layer, which is learnable, meaning that its parameters are updated during the model training process. The main function of the embedding layer is to convert the feature type ID into a fixed-length dense vector that can express the semantic and grammatical properties of each feature in the model. The dimension of the embedding vector usually depends on the complexity of the model and the number of features processed. The larger the dimension, the richer the information that can be expressed, but at the same time, the more computation is required. The parameters of the embedding layer are usually randomly initialized before training, and then optimized during the training process through the back propagation algorithm. The feature type encoding is used as a training parameter to participate in the training. After obtaining the input sequence X, the feature type information C corresponding to the elements in the input sequence is recorded. n}, where c i Indicates the feature type id of the element at position i in X. Feature types include work order title, work order problem description, work order creator information, region, product line, and special tags. Input C into the embedding layer to obtain the feature type encoding sequence

[0053] (3) Ticket representation: The first step in ticket representation is to perform embedding calculations on the input sequence to convert each input text unit (such as a word or token) into a corresponding vector form. This process usually involves the following three types of embeddings: Token embedding, which converts each word or symbol in the text into a predefined vector. These vectors are usually learned during the pre-training process and can capture the semantic information of the vocabulary. Feature type encoding, which provides a vector with its type information for each element in the input sequence. As mentioned earlier, this includes ticket titles, problem descriptions, etc. Feature type encoding helps the model understand the role of each word in the text. Position encoding, since the model architecture used (such as Transformer) does not naturally handle the order information of sequence data, position encoding supplements this by adding the position information of each word. This enables the model to handle the order relationship of words, such as the semantics of phrase or sentence structure. All of these embedding vectors are in the same dimensional space and are combined together by element-by-element addition to form the final embedding vector. The calculation method is in and p i They are token embedding, feature type encoding and position encoding respectively. 0 After n times of calculation of the encoding layer of the representation model, H is obtained n , the calculation process is H n =transformer n (H n-1 ).

[0054] (4) Matching business classification: Take H n The corresponding vector of [CLS] is used as the feature vector of the work order input, and the probability distribution of the business classification corresponding to the work order is calculated using a feedforward neural network. During the inference in the production stage, the business classification id whose probability exceeds the set threshold is returned. If it does not exist, a null value is returned. At this time, the intelligent work order assistant does not backfill the data and prompts the user that the automatic classification failed.

[0055] During the training phase, the basic representation model part of the model is pre-trained based on large-scale language data to enable the model to have basic language understanding capabilities. On this basis, the model is enhanced based on the structural features of the work order input and a classification task head is added. Further fine-tuning training is performed on production data in the specific field of work order classification. The training of the work order classification model is completed in a high-performance computing environment, using a large amount of annotated historical work order data. During the training process, the model will continuously adjust its internal parameters, and use techniques such as cross-validation and early stopping to ensure that the model achieves the best performance. Model optimization includes not only adjusting model parameters, but also periodically retraining the model to adapt to new business scenarios and data. In addition, by monitoring the performance of the model in real time and collecting user feedback, the system can continuously optimize the model to make it more accurate and reliable.

[0056] Embodiment 2, refer to the attached Figure 1 to Figure 2 As shown, based on the first embodiment, a work order classification system based on a deep learning model is proposed, and the system includes:

[0057] By combining modern deep learning technology and natural language processing technology, we are committed to improving the automation level and accuracy of work order processing. The system consists of two core parts: the work order classification model and the work order intelligent assistant, which work together to achieve efficient and accurate work order classification.

[0058] 1. Work Order Intelligent Assistant

[0059] The work order intelligent assistant is the main interface for users to interact with the system. It is integrated into the work order system's order creation page. The main functions of the intelligent assistant include receiving the work order information input by the user, converting this information into a structured model input format, sending it to the back-end work order classification model, and presenting the model's output results to the user.

[0060] The operation process of the work order intelligent assistant is as follows:

[0061] The process involved is as follows:

[0062] (1) User input reception: During the user input reception stage, the system's front-end interface will provide an intuitive and easy-to-use ticket creation page, where users can enter various ticket information, including but not limited to ticket title, problem description, creator information, region and product line, etc. This interface will support multiple input methods such as text input, selection box, date picker, etc. to meet different types of ticket requirements. The front-end implements input validation to ensure the correctness of the data format, such as verifying the email format, phone number, etc., to reduce invalid requests processed by the back-end, and optimizes the interface layout through user behavior analysis, so that users can fill in information more quickly and accurately.

[0063] (2) Information parsing and structuring: The intelligent assistant parses the input text, extracts key information using natural language processing technology, and converts it into structured data that can be processed by the model.

[0064] (3) Classification request processing: The parsed and structured data will be sent to the deep learning work order classification model at the backend. This model is deployed on a high-performance server and can handle high-concurrency data requests to ensure that the classification results are returned within the user's waiting time. The model uses pre-trained models such as BERT and GPT and is fine-tuned according to actual business needs. Model deployment uses distributed computing technologies, such as Kubernetes container management, to optimize resource usage and improve the system's response speed and processing capabilities.

[0065] (4) Result verification and feedback: After the classification model processes the data, the system will verify the results, confirm the legitimacy of the returned business classification ID, and query the database to obtain detailed business process information, responsible persons, etc., so that users can better understand the classification results. The result verification ensures the legitimacy and accuracy of the model output by setting verification rules. The back-end builds a business logic layer to handle data query and integration, and provide complete business information for the front-end display.

[0066] (5) Multi-level classification information processing: For complex work orders, multi-level classification may be required. The system will recursively query the next level of classification based on the first level classification results until all necessary classification information is completed to ensure data integrity and accuracy. The back-end introduces a cache mechanism to reduce the number of database queries and improve system efficiency.

[0067] (6) Data backfilling and user feedback: The system will automatically backfill the classification results to the order creation page and allow users to view and modify the pre-classification results to ensure that the final submitted work order information meets the user's actual needs. The classification results are displayed in the form of labels, and there is an edit button next to each label, allowing users to click to modify when necessary. In addition, in order to enhance the user experience, a brief description or prompt information will be displayed under each classification label to explain the business meaning of the classification label to help users understand why it is automatically classified into this category. The front end uses a reasonable user interaction design so that users can easily modify the classification results. Real-time data binding between the front and back ends is implemented to ensure timely response to user operations. In order to further optimize the work order classification model and user experience, the system provides a feedback mechanism, where users can provide feedback on the accuracy of the classification results through simple clicks. There are "correct" and "wrong" buttons next to each classification result, and users can choose according to the actual situation. The collected user feedback will be used as important data to input into the subsequent model training and optimization process to help the system better learn and adapt to the specific needs of different users.

[0068] 2. Work Order Classification Model

[0069] The work order classification model is the technical core of this system. It uses deep learning and natural language processing technology to accurately analyze work order information. The model structure diagram is as follows Figure 2 ,The model is mainly composed of the work order representation model and the ,business classification matching model.

[0070] The main structure of the model is as follows:

[0071] (1) Work Order Representation Model: The core of the work order representation model is a pre-trained language model based on the Bert architecture. The Bert (Bidirectional Encoder Representations from Transformers) model is a model based on the Transformer architecture. It can understand the deep semantics of language by pre-training a large amount of language data. In this system, the Bert model is used to extract and understand the semantic information of the work order text. In order to enhance the model's ability to understand different types of work order attributes, a feature type embedding module is added to the Bert model. The main function of this module is to embed features of various attributes in the work order and convert them into a format that the model can understand. In this way, the model can not only understand the text content, but also perform more accurate analysis based on the specific attributes of the work order.

[0072] (2) Business classification matching model: The business classification matching model is built based on a feedforward neural network, which directly uses the feature vector sequence output by the work order representation model as input. The task of this model is to extract the most critical information from the work order feature vector and calculate the probability distribution of each business classification. The work order feature vector is generated by the work order representation model and contains a compressed representation of all relevant information about the work order. In the business classification matching model, these vectors are processed through a series of neural network layers, including activation functions, normalization layers, and dropout layers, to improve the generalization ability of the model and avoid overfitting. After the feature vector is processed, the model uses a softmax layer to calculate the probability of each business classification. This probability distribution helps the system determine which business classification the work order is most likely to belong to, so that the work order is automatically classified into the correct category.

[0073] In modern enterprise operations, an efficient and accurate work order processing system is the key to ensuring service quality and operational efficiency. The automated work order classification system based on deep learning can quickly process a large number of work orders through its advanced technical framework, significantly reducing the time of work order processing and greatly optimizing the speed and efficiency of work order classification. Compared with traditional manual classification methods, this system not only speeds up the speed of work order processing, but also ensures the timely processing of various work order requests in an intelligent way.

[0074] The introduction of deep learning models allows the system to self-learn and self-optimize based on a large amount of historical data. This learning ability ensures that work orders can be accurately and intelligently classified into appropriate system business categories, significantly reducing classification errors caused by human judgment errors. Compared with traditional methods, this automated system reduces the reliance on employees' ability to understand business classification, thereby reducing human resource costs and improving resource allocation and utilization efficiency.

[0075] As the number of business and work orders continues to grow, the automated work order classification system based on deep learning has shown excellent scalability. The system can easily adapt to changes in business needs by adding training data and adjusting learning algorithms, without investing a lot of manpower and training costs. This is especially important for a fast-growing and changing corporate environment, as it can quickly adapt to new business needs and market changes without affecting existing operational processes.

[0076] In addition, the deep learning-based system also has the ability to self-learn and optimize. As more and more work order data is accumulated, the system can continuously learn from it and optimize its classification strategy to ensure the continued accuracy and efficiency of the prediction and classification strategy. Through the feature type embedding technology in the work order representation model, the system can adapt to various types of work order content and provide more personalized and accurate work order classification suggestions. This personalized processing not only improves user satisfaction, but also improves the response speed and processing capacity of the entire system.

[0077] The intelligent work order assistant in the system provides classification suggestions in an auxiliary form, and users do not need to undergo additional training or adapt to a new system interface. This design ensures that even in abnormal situations where the model cannot provide accurate classification suggestions, it will not affect the normal user workflow, thereby ensuring business continuity and system reliability.

[0078] In summary, the automatic work order classification system based on deep learning not only improves the efficiency and accuracy of work order processing, but also brings significant results in resource optimization. The introduction of this system has brought enterprises reduced operating costs, improved service quality, and better preparation for future challenges. Through these improvements, enterprises can manage and respond to customer needs more effectively, ensuring that they can maintain their leading position in a highly competitive market environment. This technological innovation not only improves the level of automation in work order processing, but also contributes to the progress of the information technology service industry.

[0079] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A work order classification method based on deep learning model, Features: The method comprises the following steps: Data preprocessing and word segmentation: Clean and standardize the text data in the work order, including the work order title, problem description, creator information, region, and product line, remove HTML tags, special symbols, and non-printing characters, and unify the character set and uppercase and lowercase characters; The cleaned text is segmented to obtain the token sequence of the work order title, problem description, creator information, region, and product line fields, and the [SEP] tag is used to separate the fields. Finally, a single input sequence is constructed, and the sequence starts with a special tag [CLS].

2. According to the work order classification method based on the deep learning model according to claim 1, Features: It also includes the feature type encoding step: Build a feature type dictionary: assign unique identifiers to all possible input feature types, such as ticket title, problem description, creator information, region, product line, and special tags; Feature type embedding: Using a learnable embedding layer, feature type IDs are converted into fixed-length dense vectors, which are updated during model training to express the semantic and grammatical properties of each feature.

3. According to the work order classification method based on deep learning model according to claim 1, Features: It also includes the work order characterization steps: Embedding calculation: Token embedding, feature type encoding and position encoding are performed on each token in the input sequence, and the three embedding vectors are added element by element in the same dimensional space to form the final embedding vector; Model encoding: The final embedding vector is input into the representation model encoding layer for n calculations to capture the deep semantic information of the text.

4. According to the work order classification method based on deep learning model according to claim 1, Features: It also includes the business classification matching step: Classification probability calculation: Extract the vector at a specific position from the model output as the feature vector of the work order, and calculate the probability distribution of each business classification corresponding to the work order through a feedforward neural network; Classification result determination: During inference in the production stage, the business classification ID whose probability exceeds the set threshold is returned. If there is no classification that meets the conditions, a null value is returned and the user is prompted that the automatic classification has failed.

5. According to the work order classification method based on deep learning model according to claim 1, Features: Also includes the model training steps: Training process: Use the labeled work order data to train the model, and use the back propagation algorithm to optimize the parameters of the embedding layer and the model encoding layer to improve the accuracy and efficiency of work order classification; Model optimization: Adjust the model structure, embedding vector dimension, and training parameters based on the training results to further improve the performance of the model.

6. A work order classification system based on a deep learning model according to the work order classification method based on a deep learning model according to any one of claims 1 to 5, Features: The system comprises: The work order intelligent assistant, as the main interface for users to interact with the system, is integrated into the work order system's order creation page. It is used to receive the work order information entered by the user, convert the information into a structured model input format, send it to the back-end work order classification model, and present the model's output results to the user; The work order classification model is deployed on a high-performance server and uses deep learning and natural language processing technology to accurately analyze the received work order information and realize automatic classification of work orders.

7. A work order classification system based on a deep learning model according to claim 6, Features: The work order intelligent assistant includes the following functions: The user input receiving function receives the work order information entered by the user through an intuitive and easy-to-use work order creation page, including the work order title, problem description, creator information, region and product line, and supports multiple input methods; Information parsing and structuring functions use natural language processing technology to parse input text, extract key information, and convert it into structured data that can be processed by the model; The classification request processing function sends the parsed and structured data to the back-end work order classification model for processing; The result verification and feedback function verifies the output results of the classification model, queries the database to obtain detailed business process information, and finally presents the results to the user; Multi-level classification information processing function supports multi-level classification processing of complex work orders to ensure data integrity and accuracy; The data backfill and user feedback function automatically fills the classification results back to the order creation page, allowing users to view and modify the pre-classification results, and provides a user feedback mechanism to optimize the model.

8. A work order classification system based on a deep learning model according to claim 6, Features: The work order classification model includes: The work order representation model is a pre-trained language model based on the Bert architecture. It is used to extract and understand the semantic information of the work order text and embed features of various attributes of the work order in combination with the feature type embedding module. The business classification matching model is built based on a feedforward neural network. It uses the feature vector sequence output by the work order representation model as input and calculates the probability distribution of each business classification through neural network layer processing, thereby realizing the automatic classification of work orders.

9. A work order classification system based on a deep learning model according to claim 7, Features: The user input receiving function also includes: Front-end input verification function to ensure the correctness of the data format entered by the user, such as verifying the email format and phone number, to reduce invalid requests processed by the back-end; User behavior analysis function is used to optimize interface layout and improve the speed and accuracy of users filling in information.

10. A work order classification system based on a deep learning model according to claim 7, Features: The result verification and feedback functions also include: Result verification rule setting function to ensure the legality and accuracy of model output; The business logic layer builds functions to process data query and integration, and provide complete business information for front-end display; The user interaction design function allows users to easily modify classification results through reasonable user interaction design, and realizes real-time data binding between the front and back ends to ensure timely response to user operations.

Citation Information

Patent Citations

  • Method for applying deep learning to Chinese word segmentation

    CN111209749A

  • Work order retrieval method, server, medium and product

    CN115640378A