Large model-based work order generation method, apparatus and device, and medium

Abstract

The invention discloses a work order generation method and device based on a large model, equipment and a medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the preprocessing of an initial work order request, so as to obtain target text data; performing semantic analysis on the target text data by using the target work order generation large model, and extracting work order feature information from the target text data based on a semantic analysis result; and classifying the work order feature information based on work order attributes in the work order feature information to obtain corresponding classification results, determining work order templates corresponding to the work order feature information according to the classification results, and filling the corresponding work order templates with the work order feature information to generate a target work order. Semantic analysis is performed by using the work order generation large model, and the work order feature information is extracted based on the semantic analysis result, so that the problems of insufficient semantic understanding depth and incomplete information extraction are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular relates to a work order generation method and device based on a large model, equipment and medium. BACKGROUND

[0002] With the continuous expansion of the scale of enterprises and the increase of business complexity, the enterprise customer service department needs to handle a large number of work order requests every day. The traditional work order processing method mainly relies on manual problem classification, information extraction and work order creation. This processing method has the problem of insufficient understanding of work order data, which leads to inaccurate work order problem classification.

[0003] Therefore, how to improve the understanding of work order data in the process of work order creation has become a technical problem to be solved at present. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a work order generation method and device based on a large model, which can extract work order feature information based on semantic analysis results by using a work order generation large model for semantic analysis, solve the problems of insufficient semantic understanding and incomplete information extraction. The specific scheme is as follows:

[0005] In a first aspect, the present application provides a work order generation method based on a large model, applied to a work order generation system, comprising:

[0006] Receiving initial work order requests sent by a target user through different channels, and preprocessing each initial work order request by using a target data processing method to obtain target text data corresponding to each initial work order request;

[0007] Performing semantic analysis on each target text data by using a target work order generation large model, and extracting work order feature information from each target text data based on the corresponding semantic analysis result; wherein the work order feature information includes work order attributes in the corresponding target text data;

[0008] Classifying each work order feature information based on each work order attribute to obtain a classification result corresponding to each work order feature information, determining a work order template corresponding to each work order feature information according to each classification result, and filling each work order feature information into the corresponding work order template to generate a target work order corresponding to each initial work order request.

[0009] Optionally, the preprocessing of each initial work order request by using a target data processing method comprises:

[0010] automatically identifying sensitive data in each of the initial work order requests, and performing data desensitization on each of the initial work order requests to obtain corresponding desensitized data;

[0011] performing text cleaning on each of the desensitized data to obtain a corresponding cleaned work order request, and performing file format conversion on each of the cleaned work order requests to obtain a standardized work order request in a target file format;

[0012] performing noise filtering processing on each of the standardized work order requests, and determining whether missing data exists in a corresponding filtered work order request, and if missing data exists in any filtered work order request, marking the missing data.

[0013] Optionally, the preprocessing of each of the initial work order requests by using the target data processing method comprises:

[0014] retrieving industry knowledge data corresponding to each of the initial work order requests from a target data source by using a retrieval enhancement generation technology;

[0015] fusing each of the industry knowledge data with a corresponding initial work order request to obtain corresponding fused data, and preprocessing each of the fused data by using the target data processing method.

[0016] Optionally, after classifying each of the work order attribute-based work order feature information, the method further comprises:

[0017] calculating a confidence score corresponding to each of the classification results, and determining whether each of the confidence scores is less than a pre-set confidence threshold, and if any of the confidence scores is less than the pre-set confidence threshold, performing manual review on the any of the confidence scores.

[0018] Optionally, the filling of each of the work order feature information into a corresponding work order template to generate a target work order corresponding to each of the initial work order requests comprises:

[0019] filling each of the work order feature information into a corresponding work order template to obtain an initial work order corresponding to each of the initial work order requests;

[0020] determining whether missing data exists in each of the initial work orders, and if missing data exists in any of the initial work orders, supplementing the missing data in the any of the initial work orders by using a pre-set data supplement method to generate the target work order corresponding to each of the initial work order requests.

[0021] Optionally, after filling each of the work order feature information into a corresponding work order template, the method further comprises:

[0022] display each of the target work orders by using a preset visual interactive interface, so that the target user interacts with the local by using the preset visual interactive interface;

[0023] generate target guide information according to the corresponding interaction condition, and display the target guide information by using the preset visual interactive interface, so that the target user optimizes the target work order according to the target guide information.

[0024] Optionally, after filling each of the work order characteristic information into the corresponding work order template, the method further includes:

[0025] obtaining target feedback information corresponding to the target work order uploaded by the target user, obtaining intermediate data in the target work order generation process, and adjusting model parameters of the target work order generation large model based on the target feedback information corresponding to the target work order and the intermediate data.

[0026] In a second aspect, the present application provides a work order generation device based on a large model, applied to a work order generation system, comprising:

[0027] a data preprocessing module, configured to receive initial work order requests sent by a target user through different channels, and pre-process each of the initial work order requests by using a target data processing method to obtain target text data corresponding to each of the initial work order requests;

[0028] a characteristic information extraction module, configured to perform semantic analysis on each of the target text data by using a target work order generation large model, and extract work order characteristic information from each of the target text data based on the corresponding semantic analysis result; wherein the work order characteristic information includes work order attributes in the corresponding target text data;

[0029] a work order generation module, configured to classify each of the work order characteristic information based on each of the work order attributes to obtain a classification result corresponding to each of the work order characteristic information, determine a work order template corresponding to each of the work order characteristic information according to each of the classification result, and fill each of the work order characteristic information into the corresponding work order template to generate a target work order corresponding to each of the initial work order requests.

[0030] In a third aspect, the present application provides an electronic device, comprising:

[0031] a memory, configured to save a computer program;

[0032] a processor, configured to execute the computer program to implement the foregoing work order generation method based on a large model.

[0033] In a fourth aspect, the present application provides a computer readable storage medium for storing a computer program, which, when executed by a processor, implements the aforementioned large model-based work order generation method.

[0034] The present application first receives initial work order requests sent by a target user through different channels, and pre-processes each initial work order request by using a target data processing method to obtain target text data corresponding to each initial work order request, then performs semantic analysis on each target text data by using a target work order generation large model, and extracts work order feature information from each target text data based on the corresponding semantic analysis result; wherein the work order feature information includes work order attributes in the corresponding target text data, and finally classifies each work order feature information based on each work order attribute to obtain a classification result corresponding to each work order feature information, determines a work order template corresponding to each work order feature information according to each classification result, and fills each work order feature information into the corresponding work order template to generate a target work order corresponding to each initial work order request. As can be seen, by using the target work order generation large model to perform semantic analysis and extracting work order feature information based on the corresponding semantic analysis result, the present application solves the problems of insufficient semantic understanding depth and incomplete information extraction in the prior art; by classifying work order feature information based on work order attributes and matching and filling corresponding work order templates, the present application realizes automatic classification and generation of work orders, solves the problems of inconsistent accuracy and low knowledge utilization efficiency caused by strong subjectivity of manual classification, and improves the efficiency of work order generation. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0036] Figure 1 A large model-based work order generation method flowchart disclosed by the present application;

[0037] Figure 2 A large model-based work order generation system structure schematic diagram disclosed by the present application;

[0038] Figure 3 A large model-based work order generation system data flow schematic diagram disclosed by the present application;

[0039] Figure 4 A work order classification flowchart disclosed by the present application;

[0040] Figure 5 A work order generation and verification flowchart disclosed by the present application;

[0041] Figure 6 A work order generation system interaction timing diagram disclosed by the present application;

[0042] Figure 7 A large model-based work order generation device structure diagram disclosed by the present application;

[0043] Figure 8 A structure diagram of an electronic device disclosed by the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] The current work order processing method has the problem of insufficient understanding of work order data, which leads to inaccurate classification of work order problems. Therefore, the present application provides a large model-based work order generation method, which solves the problems of insufficient semantic understanding and incomplete information extraction by using a work order generation large model to perform semantic analysis and extracting work order feature information based on the semantic analysis results.

[0046] Referring to Figure 1 The embodiments of the present application disclose a large model-based work order generation method applied to a work order generation system, which comprises:

[0047] Step S11, receiving initial work order requests sent by a target user through different channels, and pre-processing each initial work order request by using a target data processing method to obtain target text data corresponding to each initial work order request.

[0048] The structure of the work order generation system in the present embodiment is shown in Figure 2 It comprises a data processing module, a large language model inference module, a work order information extraction module, a multi-dimensional intelligent classification module, a rule engine and knowledge base module, a work order automatic generation and template filling module, a man-machine collaboration verification module, a continuous learning and optimization module, and a system integration and interface module. The data flow in the system is shown in Figure 3 It generally comprises data input, data preprocessing, feature extraction, knowledge enhancement, feedback learning, intelligent analysis, and result output processes.

[0049] The system integration and interface module comprises:

[0050] Standard API (Application Programming Interface): Provides RESTful API (a programming interface) and GraphQL (a query language for API) interfaces, supporting integration with mainstream ticket systems;

[0051] Message queue processor: Supports asynchronous message processing, improving system response performance;

[0052] Data synchronization manager: Ensures data consistency and real-time with other enterprise systems;

[0053] Permission manager: Provides fine-grained user permission control and data security protection;

[0054] Monitoring and alarm system: Real-time monitoring of system running status, timely discovery and handling of abnormal situations.

[0055] It should be noted that the ticket generation system in this embodiment uses TensorFlow, PyTorch and other mainstream deep learning frameworks, which can achieve:

[0056] Model parallel processing: Supports distributed inference of large-scale models;

[0057] Dynamic graph calculation: Provides flexible calculation graph construction and optimization;

[0058] GPU accelerated computing: Fully utilizes hardware resources to improve processing efficiency;

[0059] Model version management: Supports model version control and hot updates.

[0060] In addition, the embodiment is also configured with multi-modal processing capability and security and privacy protection mechanism:

[0061] Multi-modal processing capability:

[0062] Image understanding module: Supports processing of customer-provided screenshot, error interface and other image information;

[0063] Speech-to-text engine: Converts customer phone recordings into processable text;

[0064] Document parser: Processes PDF, Word and other formats of attachment documents;

[0065] Multi-modal fusioner: Integrates text, image, voice and other information sources for analysis.

[0066] Security and privacy protection mechanism:

[0067] To ensure system security and output quality, a complete security protection system is implemented:

[0068] Input validation gateway: filtering malicious inputs, SQL (Structured Query Language) injection, and other security threats;

[0069] Output content censor: preventing the model from generating inappropriate or harmful content;

[0070] Sensitive information identifier: automatically identifying and desensitizing customer privacy information;

[0071] Data encryption transmission: using AES-256 (Advanced Encryption Standard 256) and RSA (Rivest-Shamir-Adleman) encryption to protect data transmission security;

[0072] Access control: fine-grained permission management based on the RBAC (Role-Based Access Control) model;

[0073] Audit log system: complete record of system operation and data access track.

[0074] In addition, the rule engine and knowledge base module include:

[0075] Business rule manager: maintain enterprise-specific ticket processing rules, such as priority promotion for VIP customers, automatic routing of specific issues, etc.

[0076] Knowledge graph constructor: establish the association between products, problems, and solutions;

[0077] Historical case library: store and manage historical ticket processing cases to support quick matching of similar problems;

[0078] Expert knowledge injector: convert the experience and knowledge of domain experts into executable rules.

[0079] The ticket processing flow in this implementation is as follows Figure 4 and Figure 5 As shown, it includes receiving a ticket request (original request), pre-processing and semantic analysis of the received ticket request to obtain corresponding ticket feature information, classifying ticket feature information according to ticket attributes to determine the corresponding ticket template for each ticket request, and finally generating a ticket based on the ticket template, and verifying the generated ticket by the user. Correspondingly, the system timing interaction diagram of the system is as follows Figure 6As shown, the user interacts with the system through the interactive interface, the system interacts with the large language model according to the user's work order request, and the large language model also interacts with the knowledge base, and finally generates the target work order corresponding to the user's original request.

[0080] In this embodiment, the process of preprocessing each initial work order request by using the target data processing method can specifically include: automatically identifying sensitive data in each initial work order request, and desensitizing each initial work order request to obtain corresponding desensitized data; performing text cleaning on each desensitized data to obtain corresponding cleaned work order requests, and performing file format conversion on each cleaned work order request to obtain standardized work order requests in a target file format; performing noise filtering processing on each standardized work order request, and judging whether there is missing data in the corresponding filtered work order request, if there is missing data in any filtered work order request, the missing data is marked.

[0081] Specifically, the above process is responsible for the data preprocessing module, which is used to receive the problem description and related information submitted by the customer through multiple channels (including telephone, email, online chat, mobile application, Web portal, etc.), and comprehensively clean and standardize the input data. Specifically, it includes:

[0082] Text cleaning unit: remove HTML (Hyper Text Markup Language, Hyper Text Markup Language) tags, special characters, redundant spaces, correct common spelling errors, and handle encoding exception problems;

[0083] Multilingual processing unit: identify the input language type, standardize non-standard languages, and support unified processing of multilingual work orders;

[0084] Format standardization unit: convert work order data from different sources to a unified JSON format and establish a standardized data structure (i.e. standardized work order request);

[0085] Noise filtering unit: filter out redundant information unrelated to the work order, such as polite words, repeated descriptions, etc.

[0086] Data integrity checking unit: verify the existence of necessary fields and mark missing key information.

[0087] In addition, the process of preprocessing each initial work order request by using the target data processing method in this embodiment includes: using retrieval enhancement generation technology to retrieve industry knowledge data corresponding to each initial work order request from the target data source; fuse each industry knowledge data with the corresponding initial work order request to obtain corresponding fused data, and preprocess each fused data by using the target data processing method.

[0088] That is, the system integrates advanced RAG (Retrieval-Augmented Generation) technology, including the following core components:

[0089] Enterprise knowledge base: store structured knowledge such as product manuals, technical documents, common questions, solutions, etc.

[0090] Vector database: use high-performance vector storage technologies such as Faiss, Pinecone, Weaviate, etc.

[0091] Semantic search engine: implement semantic similarity calculation based on DPR (Dense Passage Retrieval) and Sentence-BERT technology;

[0092] Context fusioner: intelligently fuse the retrieved relevant knowledge with the original question;

[0093] Knowledge update manager: provides real-time update and maintenance functions for the knowledge base.

[0094] Through retrieval-enhanced generation technology, the value of knowledge resources is fully realized, and the technical problem of not being able to effectively integrate and utilize historical data, knowledge base, and expert experience resources is solved.

[0095] It should be noted that the initial work order request in this embodiment is multi-source heterogeneous data, which may include pictures and voice information. Through the data preprocessing module, the multi-source heterogeneous data is preprocessed, solving the technical problem that the prior art can only process text information and cannot effectively process multi-modal input such as pictures and voice, providing comprehensive multi-modal information understanding capability.

[0096] Step S12, using the target work order generation large model to perform semantic analysis on each target text data, and extracting work order feature information from each target text data based on the corresponding semantic analysis result; wherein the work order feature information includes the work order attributes in the corresponding target text data.

[0097] In this embodiment, the process of semantic analysis is responsible for the aforementioned large language model inference module, which uses a large language model that has been fine-tuned for domain adaptability (i.e. target work order generation large model) as the core understanding engine, specifically including:

[0098] Pre-trained model loader: Supports loading various mainstream large language models (GPT (Generative Pre-trained Transformer), BERT (Bidirectional Encoder Representations from Transformers), etc.);

[0099] Semantic encoder: Converts pre-processed ticket text into high-dimensional semantic vector representation;

[0100] Context understanding engine: Analyzes the business background, technical context and implied needs of the ticket;

[0101] Deep reasoning engine: Understands the deep meaning of customer description through multi-layer attention mechanism;

[0102] Domain knowledge fusion: Combines enterprise-specific business knowledge and historical cases for reasoning enhancement.

[0103] In addition, the process of extracting ticket feature information from each target text data based on the semantic analysis result in the present embodiment is responsible for the aforementioned ticket information extraction module. Specifically, the ticket information extraction module, based on the understanding result (i.e. semantic analysis result) of the large language model, uses structured information extraction technology to identify and extract the key information required by the ticket:

[0104] Entity recognition component: Recognize named entities in the description, including product name, version number, error code, user name, device model, etc.

[0105] Relationship extraction component: Analyze the semantic relationship between entities, such as the ternary relationship of user-operation-system;

[0106] Attribute classifier: Extract ticket attribute information;

[0107] Time information extractor: Identify the time of problem occurrence, first occurrence, duration, etc.

[0108] Sentiment analyzer: Evaluate customer emotional state and satisfaction, assist in priority judgment.

[0109] Among them, the above-mentioned ticket attribute information includes:

[0110] Problem description: Structured summary and core problem identification of customer problem;

[0111] Problem type: Software failure, hardware failure, network problem, account management, service request, consultation, etc.

[0112] Influence scope: Single user, partial user, system level, global, etc.

[0113] Emergency level: multiple levels of classification such as emergency (P0), high (P1), medium (P2), low (P3), etc.

[0114] Department: IT operation department, network department, application development department, customer service department, etc.

[0115] Product line: core business system, office system, network infrastructure, etc.

[0116] Customer information: contact name, phone, email, department, permission level, etc.

[0117] Technical environment: operating system, browser version, network environment, device configuration, etc.

[0118] By using a large model to deeply understand the semantic understanding of target text data, the technical problem that the existing technology cannot accurately understand the implicit requirements and complex business context in the customer's natural language description is solved, providing deep semantic understanding ability, and solving the technical problem that the existing solution can only perform simple classification and cannot fully extract multi-dimensional information required by the work order. The information such as work order type, priority, impact range, technical environment, etc. is automatically identified and extracted.

[0119] Step S13, classifying each work order feature information based on each work order attribute to obtain the classification result corresponding to each work order feature information respectively, determining the work order template corresponding to each work order feature information respectively according to each classification result, and filling each work order feature information into the corresponding work order template to generate the target work order corresponding to each initial work order request respectively.

[0120] In this embodiment, the process of classifying work order feature information is responsible for the aforementioned multi-dimensional intelligent classification module, which adopts a multi-level and multi-dimensional classification strategy to ensure the accuracy and comprehensiveness of classification:

[0121] Hierarchical classification algorithm:

[0122] First-level classification: determine the main category of the work order (technical problem, business problem, service request, etc.);

[0123] Second-level classification: subdivision under the main category (such as technical problems divided into software, hardware, network, etc.);

[0124] Third-level classification: determine the specific problem type (such as software problems divided into installation, configuration, fault, etc.);

[0125] Multi-dimensional feature fusion: consider multiple features such as problem type, impact range, emergency level, customer level, etc.

[0126] Context-aware matcher: intelligent matching based on customer historical work orders, similar cases and business rules;

[0127] Confidence calculator: Calculate confidence scores for each classification result, trigger human review when below a preset threshold;

[0128] Dynamic weight adjuster: Dynamically adjust the weights of each dimension based on historical data and feedback results.

[0129] Correspondingly, in the embodiment, after classifying the work order attribute-based work order feature information, it further includes: calculating the confidence scores corresponding to each classification result, and judging whether each confidence score is less than a preset confidence threshold. If any confidence score is less than the preset confidence threshold, then manually review any confidence score.

[0130] In addition, in the embodiment, the process of filling each work order feature information into the corresponding work order template to generate the target work order corresponding to each initial work order request can specifically include: filling each work order feature information into the corresponding work order template to obtain the initial work order corresponding to each initial work order request; judging whether there is missing data in each initial work order, if there is missing data in any initial work order, then supplementing the missing data in any initial work order using a preset data supplement method to generate the target work order corresponding to each initial work order request.

[0131] The above process is responsible for the work order automatic generation and template filling module, which includes:

[0132] Dynamic template selector: Select the most suitable work order template according to the work order classification result;

[0133] Intelligent filling engine: Automatically map the extracted key information to the corresponding fields of the template;

[0134] Information completion reasoner: intelligently infer missing non-key information or supplement from historical data;

[0135] Formatted outputter: Generate structured work order documents that meet enterprise standards;

[0136] Integrity verifier: Check if the generated work order contains all necessary information.

[0137] In addition, in the embodiment, after filling each work order feature information into the corresponding work order template, it further includes: displaying each target work order using a preset visual interactive interface, so that the target user interacts with the local using the preset visual interactive interface; generate target guide information according to the corresponding interaction, and display the target guide information using the preset visual interactive interface, so that the target user optimizes the target work order according to the target guide information.

[0138] The above process is responsible for the man-machine cooperation module, and the man-machine cooperation module includes:

[0139] Visual presenter: visually presents the automatically generated work order information to the user;

[0140] Interactive correction interface (i.e., preset visual interactive interface): provides convenient modification functions and supports field-level quick adjustment;

[0141] Confidence prompter: special marking and prompting of system uncertain information;

[0142] Multi-round dialogue manager: supports multi-round interaction with customers and guides the supplement of missing information;

[0143] Collaborative workflow engine: manages the workflow of manual review to ensure efficient collaboration.

[0144] In addition, after filling each work order feature information into the corresponding work order template, it further includes: obtaining target feedback information corresponding to the target work order uploaded by the target user, obtaining intermediate data in the target work order generation process, and adjusting the model parameters of the target work order generation model based on the target feedback information and the intermediate data corresponding to the target work order.

[0145] The above process is completed by the continuous learning and optimization module, and the continuous learning and optimization module includes:

[0146] Feedback data collector: records the whole process data and result feedback (i.e., target feedback information) of each work order processing;

[0147] Error pattern analyzer: identifies common error types and failure modes of the system;

[0148] Incremental learning engine: continuously updates model parameters based on new data and feedback information;

[0149] Performance monitor: real-time monitoring of performance indicators and accuracy of each module of the system;

[0150] Automatic optimization scheduler: periodically triggers model retraining and parameter optimization tasks.

[0151] The specific workflow of the embodiment is as follows:

[0152] Stage one: data reception and preprocessing:

[0153] 1. The system receives work order requests from multiple channels through a unified access layer;

[0154] 2. The data preprocessing module cleans, formats and standardizes the original data;

[0155] 3. Perform data integrity checks and mark missing or abnormal information.

[0156] Phase Two: Semantic Understanding and Analysis:

[0157] 4. The large language model reasoning module performs deep semantic analysis on the preprocessed text.

[0158] 5. Combined with enterprise knowledge base retrieval enhancement, relevant background information is obtained.

[0159] 6. Generate semantic vector representation of text, prepare for subsequent classification.

[0160] Phase Three: Information Extraction and Feature Recognition:

[0161] 7. The ticket information extraction module identifies and extracts key entities and attribute information.

[0162] 8. Analyze the relationship and dependency between entities and construct a structured representation of the problem.

[0163] 9. Perform sentiment analysis and urgency assessment.

[0164] Phase Four: Intelligent Classification and Priority Determination:

[0165] 10. The multi-dimensional intelligent classification module performs hierarchical classification.

[0166] 11. Combine business rules and historical cases for classification verification.

[0167] 12. Calculate the confidence of the classification result and mark uncertain items.

[0168] Phase Five: Ticket Generation and Filling:

[0169] 13. Select the appropriate ticket template according to the classification result.

[0170] 14. Automatically fill the extracted information into the corresponding fields.

[0171] 15. Intelligently infer or mark missing information as to be supplemented.

[0172] Phase Six: Human-Machine Collaboration Verification:

[0173] 16. Show the generated ticket information to the customer or customer service.

[0174] 17. Provide an interactive interface for confirmation and correction.

[0175] 18. Collect supplementary information for uncertain or missing information.

[0176] Phase Seven: Final Ticket Generation and Distribution:

[0177] 19. Generate the final ticket according to the confirmation result.

[0178] 20. Route the ticket to the corresponding processing department according to the classification result;

[0179] 21. Update the ticket system and send notifications.

[0180] Phase Eight: Feedback Learning and Optimization:

[0181] 22. Record the complete processing process and result data;

[0182] 23. Analyze the processing effect and customer satisfaction;

[0183] 24. Use feedback information for continuous optimization of the model.

[0184] In one specific embodiment, the large language model inference module in this embodiment adopts a pre-trained language model based on the Transformer architecture, and is deeply optimized through the following ways:

[0185] 1. Domain knowledge injection: supervised fine-tuning through enterprise historical ticket data, product documents, FAQ (Frequently Asked Questions), etc.

[0186] 2. Few-shot learning: use Prompt Engineering and In-Context Learning technology;

[0187] 3. Thought chain reasoning: guide the model to perform step-by-step logical reasoning;

[0188] 4. Knowledge graph integration: combine structured knowledge graphs to enhance reasoning capabilities;

[0189] 5. Multi-task learning: simultaneously optimize classification, information extraction, generation, and other tasks.

[0190] In another specific embodiment, this embodiment can implement an advanced confidence assessment mechanism, including:

[0191] 1. Intrinsic model confidence: calculate confidence based on model output probability distribution;

[0192] 2. Consistency check: assess credibility through consistency of multiple inference results;

[0193] 3. Cross-validation: use different models or methods to verify results;

[0194] 4. Historical accuracy: based on the historical processing accuracy of similar cases;

[0195] 5. Expert rule verification: logical consistency check through business rules.

[0196] Therefore, by using the target work order generation large model to perform semantic analysis and extracting work order feature information based on the corresponding semantic analysis result, the application solves the problems of insufficient semantic understanding depth and incomplete information extraction in the prior art. By classifying the work order feature information based on the work order attributes and matching and filling the corresponding work order template, the application realizes the automatic classification and generation of work orders, solves the problems of inconsistent accuracy and low knowledge utilization efficiency caused by strong subjectivity of manual classification, and improves the efficiency of work order generation.

[0197] Referring to Figure 7 As shown in the figure, the embodiment of the application discloses a work order generation device based on a large model, applied to a work order generation system, comprising:

[0198] The data preprocessing module 11 is configured to receive initial work order requests sent by a target user through different channels, and preprocess each initial work order request by using a target data processing method to obtain target text data corresponding to each initial work order request.

[0199] The feature information extraction module 12 is configured to perform semantic analysis on each target text data by using a target work order generation large model, and extract work order feature information from each target text data based on the corresponding semantic analysis result. The work order feature information includes work order attributes in the corresponding target text data.

[0200] The work order generation module 13 is configured to classify each work order feature information based on each work order attribute to obtain a classification result corresponding to each work order feature information, determine a work order template corresponding to each work order feature information according to each classification result, and fill each work order feature information into the corresponding work order template to generate a target work order corresponding to each initial work order request.

[0201] In some embodiments, the data preprocessing module 11 can specifically include:

[0202] The data desensitization unit is configured to automatically identify sensitive data in each initial work order request, and perform data desensitization on each initial work order request to obtain desensitized data.

[0203] The format conversion unit is configured to perform text cleaning on each desensitized data to obtain a cleaned work order request, and perform file format conversion on each cleaned work order request to obtain a standardized work order request in a target file format.

[0204] The data labeling unit is configured to perform noise filtering processing on each standardized work order request, and determine whether there is missing data in the filtered work order request. If there is missing data in any filtered work order request, the missing data is labeled.

[0205] In some embodiments, the data preprocessing module 11 can specifically include:

[0206] a data retrieval unit configured to retrieve, from a target data source, industry knowledge data corresponding to each of the initial work order requests by using a retrieval enhancement generation technique;

[0207] a data fusion unit configured to fuse each of the industry knowledge data with the corresponding initial work order request to obtain corresponding fused data, and to preprocess each of the fused data by using the target data processing method.

[0208] In some embodiments, the work order generation module 13 further includes:

[0209] a confidence score calculation unit configured to calculate a confidence score corresponding to each of the classification results, and to determine whether each of the confidence scores is less than a preset confidence threshold, and to perform manual review on any confidence score that is less than the preset confidence threshold.

[0210] In some embodiments, the work order generation module 13 can specifically include:

[0211] an information filling unit configured to fill each of the work order feature information into a corresponding work order template to obtain an initial work order corresponding to each of the initial work order requests;

[0212] a data supplement unit configured to determine whether there is missing data in each of the initial work orders, and to supplement the missing data in any initial work order by using a preset data supplement method if there is missing data in the initial work order, to generate the target work order corresponding to each of the initial work order requests.

[0213] In some embodiments, the work order generation module 13 further includes:

[0214] a work order display unit configured to display each of the target work orders by using a preset visual interactive interface, so that the target user interacts with the local by using the preset visual interactive interface;

[0215] a guide information generation unit configured to generate target guide information according to a corresponding interaction, and to display the target guide information by using the preset visual interactive interface, so that the target user optimizes the target work order according to the target guide information.

[0216] In some embodiments, the work order generation module 13 further includes:

[0217] The model parameter adjustment unit is configured to obtain target feedback information uploaded by the target user and corresponding to the target work order, obtain intermediate data in a target work order generation process, and adjust model parameters of the target work order generation large model based on the target feedback information corresponding to the target work order and the intermediate data.

[0218] Further, the embodiment of the present application further discloses an electronic device, Figure 8 The electronic device 20 is shown in the structure diagram of the electronic device 20 according to an exemplary embodiment, and the content in the figure cannot be considered as any limitation on the use range of the present application.

[0219] Figure 8 The electronic device 20 is shown in the structure diagram of the electronic device 20 according to an exemplary embodiment, and the content in the figure cannot be considered as any limitation on the use range of the present application.

[0220] In the embodiment, the power supply 23 is configured to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol followed by the communication interface 24 can be any communication protocol applicable to the technical solution of the present application, which is not limited here; the input and output interface 25 is configured to obtain external input data or output data to the outside, and the specific interface type can be selected according to the specific application needs, which is not limited here.

[0221] In addition, the memory 22 as a carrier for resource storage can be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc., and the resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage mode can be temporary storage or permanent storage.

[0222] The operating system 221 is configured to manage and control each hardware device on the electronic device 20 and the computer program 222, and can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program capable of completing the large model-based work order generation method executed by the electronic device 20 disclosed in any of the preceding embodiments, the computer program 222 can further include a computer program capable of completing other specific work.

[0223] Further, the application also discloses a computer readable storage medium for storing a computer program, wherein the computer program is executed by a processor to realize the foregoing large model-based work order generation method. For specific steps of the method, refer to the corresponding content disclosed in the foregoing embodiments, which will not be repeated here.

[0224] The various embodiments are described in the specification by progressive stages, and each embodiment focuses on the difference from other embodiments. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts refer to the method part.

[0225] The skilled person can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in the above description in general terms. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0226] The steps of the method or algorithm described in combination with the embodiments disclosed herein can be directly implemented by hardware, a software module executed by a processor, or a combination of both. The software module can be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0227] Finally, it should be noted that, in this document, relational terms such as first and second are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0228] The technical solutions provided by the present application are described in detail above, and the principles and implementation manners of the present application are described by using specific examples. The above description of the examples is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will be changed, and the above description of the content of the specification should not be understood as a limitation on the present application.

Claims

1. A large model-based work order generation method, characterized in that, The application is applied to a work order generation system, comprising: receiving initial work order requests sent by target users through different channels, and preprocessing each initial work order request by using a target data processing method to obtain target text data corresponding to each initial work order request; performing semantic analysis on each target text data by using a target work order generation large model, and extracting work order feature information from each target text data based on the corresponding semantic analysis result; wherein the work order feature information includes work order attributes in the corresponding target text data; classifying each work order feature information based on each work order attribute to obtain a classification result corresponding to each work order feature information, determining a work order template corresponding to each work order feature information according to each classification result, and filling each work order feature information into the corresponding work order template to generate a target work order corresponding to each initial work order request.

2. The large model-based work order generation method of claim 1, wherein, The preprocessing of each initial work order request by using a target data processing method comprises: automatically identifying sensitive data in each initial work order request, and desensitizing each initial work order request to obtain desensitized data; performing text cleaning on each desensitized data to obtain cleaned work order requests, and converting each cleaned work order request into a standardized work order request in a target file format; performing noise filtering processing on each standardized work order request, and determining whether there is missing data in the filtered work order request, if there is missing data in any filtered work order request, marking the missing data. 3.The large model-based work order generation method of claim 1, wherein, The preprocessing of each initial work order request by using a target data processing method comprises: retrieving industry knowledge data corresponding to each initial work order request from a target data source by using a retrieval enhancement generation technology; fuse each industry knowledge data with the corresponding initial work order request to obtain fusion data, and preprocess each fusion data by using the target data processing method. 4.The large model-based work order generation method of claim 1, wherein, After classifying each work order feature information based on each work order attribute, the method further comprises: calculating a confidence score corresponding to each classification result, and determining whether each confidence score is less than a pre-set confidence threshold, if any confidence score is less than the pre-set confidence threshold, performing manual review on the confidence score. 5.The large model-based work order generation method of claim 1, wherein, The filling of each work order feature information into the corresponding work order template to generate a target work order corresponding to each initial work order request comprises: filling each work order feature information into the corresponding work order template to obtain an initial work order corresponding to each initial work order request; determining whether there is missing data in each initial work order, if there is missing data in any initial work order, supplementing the missing data in the initial work order by using a pre-set data supplementing method to generate the target work order corresponding to each initial work order request.

6. The large model-based work order generation method of claim 1, wherein, After filling each work order feature information into the corresponding work order template, the method further comprises: Display each target work order by using a preset visual interactive interface, so that the target user interacts with the local by using the preset visual interactive interface; Generate target guide information according to the corresponding interaction condition, and display the target guide information by using the preset visual interactive interface, so that the target user optimizes the target work order according to the target guide information.

7. The large model-based work order generation method according to any one of claims 1 to 6, characterized in that, After filling each work order feature information into the corresponding work order template, it further includes: Obtain the target feedback information corresponding to the target work order uploaded by the target user, obtain the intermediate data in the target work order generation process, and adjust the model parameters of the target work order generation large model based on the target feedback information corresponding to the target work order and the intermediate data. 8.A large model-based work order generation device, characterized by, Applied to a work order generation system, comprising: A data preprocessing module is configured to receive initial work order requests sent by a target user through different channels, and preprocess each initial work order request by using a target data processing method to obtain target text data corresponding to each initial work order request. A feature information extraction module is configured to perform semantic analysis on each target text data by using a target work order generation large model, and extract work order feature information from each target text data based on the corresponding semantic analysis result; wherein the work order feature information includes work order attributes in the corresponding target text data. A work order generation module is configured to classify each work order feature information based on each work order attribute to obtain a classification result corresponding to each work order feature information, determine a work order template corresponding to each work order feature information according to each classification result, and fill each work order feature information into the corresponding work order template to generate a target work order corresponding to each initial work order request.

9. An electronic device, comprising: Comprising: A memory for saving a computer program; A processor for executing the computer program to realize the large model-based work order generation method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program for saving, which is executed by a processor to realize the large model-based work order generation method of any one of claims 1 to 7.

Images