Intelligent customer service assistant self-growth method and device

Abstract

The embodiment of the application provides a kind of intelligent customer service assistant self-growth method and device, method includes: by obtaining historical multi-source heterogeneous data and carrying out cleaning and structured processing, constructs the service log library including three elements of user question, correct answer and processing result;After deploying basic customer service model, the newly added multi-source data is carried out denoising, duplicate removal and structured processing, and based on the high-quality work order in service log library as seed sample, AI learns labeling rule to realize the automatic labeling of new data;According to the training set obtained by automatic labeling, the basic customer service model is lightweight self-training using incremental fine-tuning, to obtain intelligent customer service model, then it is self-tested, and low-score cases are screened and the service log library is updated according to the knowledge, the updated knowledge is applied to the processing or training of subsequent new data, to form a self-growth closed loop, the application can improve the response efficiency and automation quality of IT service desk.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a method and apparatus for the self-generation of an intelligent customer service assistant. Background Technology

[0002] In traditional IT (Information Technology) help desk scenarios, there is typically a high concurrency service pressure of hundreds to thousands of phone inquiries and ticket processing per day. Common issues include high-frequency, repetitive inquiries such as password resets, account unlocking, network failures, software installation, and permission requests. Current mainstream solutions still primarily rely on "human agents + static knowledge bases," which suffers from the following technical bottlenecks: Lagging knowledge updates: Static FAQs and troubleshooting manuals cannot absorb new faults and solutions in real time, resulting in outdated and inaccurate answers from AI (artificial intelligence) customer service. High annotation costs: Training AI requires a large amount of high-quality dialogue data annotated by humans. The IT service desk generates a large number of unstructured call records and work orders every day, which are difficult to efficiently convert into training data. Model iteration is difficult: Traditional methods require periodic retraining of large models, which is costly, time-consuming, and prone to "catastrophic forgetting"—learning new things but forgetting old ones; Lack of closed-loop: AI service results (such as referral to human agents, negative user reviews, and agent corrections) cannot be automatically fed back for model optimization, resulting in stagnation of capabilities.

[0003] To overcome the above problems, there is an urgent need for a self-growing method for intelligent customer service assistants that can improve the response efficiency and automation quality of IT service desks. Summary of the Invention

[0004] To address the problems in the prior art, this application provides a method and apparatus for the self-generation of an intelligent customer service assistant, which can improve the response efficiency and automation quality of the IT service desk.

[0005] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a method for the self-generation of an intelligent customer service assistant, comprising: Acquire historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data; perform data cleaning and structuring processing on the multi-source heterogeneous data; and construct a service log library. A basic customer service model is deployed. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals. The new multi-source data is cleaned in three ways, including noise removal, deduplication, and structuring. The corresponding structured data is determined, and the structured data is self-labeled to determine the corresponding labeled dataset. The self-labeling includes obtaining high-quality work orders from the service log library as seed samples and reusing the labeling rules of the seed samples. The basic customer service model uses the labeled dataset as the training set for lightweight self-training to determine the corresponding intelligent customer service model. The intelligent customer service model is then self-evaluated to determine the corresponding model evaluation score. Self-evaluation cases with model evaluation scores below a preset threshold are selected. The service log library is then self-updated based on the self-evaluation cases. The updated service log library is then applied to the processing or training of subsequent new data, enabling the intelligent customer service model to achieve a self-growth closed loop and thus realize data processing for the information technology service desk.

[0006] Furthermore, the step of performing data cleaning and structuring processing on the multi-source heterogeneous data to construct a service log library includes: The recorded telephone data is automatically processed by speech recognition to convert it to text, and the corresponding text data is determined. Data cleaning is performed on the work order system data, the knowledge document data, and the text data. Core fields are extracted from the cleaned data to determine the user problem, standard solution, and processing result corresponding to each work order, and a service log library is constructed. The core field extraction includes extracting the work order title, fault description, fault category, processing steps, solution, closed-loop status, user satisfaction, and reason for transferring to human intervention.

[0007] Furthermore, the triple data cleaning process for the newly added multi-source data includes noise reduction, deduplication, and structuring to determine the corresponding structured data, including: The newly added multi-source data is subjected to noise reduction, including filtering out harassing data, test data, and invalid duplicate data without substantial content from the newly added multi-source data; The newly added multi-source data is deduplicated, including for duplicate data with the same fault type and the same solution, retaining the best processing case and eliminating the other duplicate cases according to a preset priority rule, which takes into account user satisfaction and processing efficiency. The data that has undergone noise reduction and deduplication is classified and information is extracted. The corresponding structured data is determined based on the extracted core field information.

[0008] Furthermore, the step of performing data self-annotation on the structured data to determine the corresponding labeled dataset includes: Obtain high-quality work orders from the service log library. These high-quality work orders include those that have been closed-loop, have user satisfaction exceeding a preset threshold, and have standardized processing procedures. Using these high-quality work orders as seed samples, construct initial labeling rules based on the labeling rules of the seed samples and the corresponding information technology service management process specifications. Each sample in the structured data is self-labeled according to the initial labeling rules. The self-labeling includes labeling the fault type, the degree of answer standardization, whether it can be solved automatically, whether it is a high-frequency problem, and whether it is a new fault that has not appeared in the service log library. The self-labeled samples are divided into positive samples, difficult samples, and negative samples according to their sample types, and stored in the corresponding sample sets to determine the corresponding labeled datasets.

[0009] Furthermore, the basic customer service model uses the labeled dataset as the training set for lightweight self-training to determine the corresponding intelligent customer service model, including: Using the labeled dataset as the training set, the basic customer service model is lightweightly self-trained using the LoRA incremental fine-tuning method; The training frequency of the lightweight self-training is dynamically adjusted according to the amount of data.

[0010] Furthermore, the self-evaluation of the intelligent customer service model to determine the corresponding model evaluation score includes: For the labeled dataset, a preset number of evaluation cases are extracted according to the preset fault type distribution ratio as a self-evaluation test set, which includes positive sample cases and difficult sample cases; The intelligent customer service model obtains its first answer on the self-evaluation test set, retrieves the corresponding agent's standard answer from the self-evaluation test set as the second answer, and performs a multi-dimensional quantitative comparison between the first answer and the second answer to determine the corresponding model evaluation score.

[0011] Furthermore, the step of updating the service log database based on the self-evaluation cases includes: The self-assessment cases are categorized into difficult cases, error-prone failure cases, and cases of misunderstanding of the process. For the aforementioned challenging cases, corresponding FAQ entries are automatically generated based on the agent's standard answer and added to the service log library; For the aforementioned error-prone fault cases, update the solutions for the corresponding fault types in the service log database, and mark the error-prone points and precautions; Cases where the process is not understood are marked as priority learning samples for reinforcement in the next round of training; At the same time, cases with incorrect answers are marked as unusable and the correct answers are added to the service log library.

[0012] Secondly, this application provides a self-generating intelligent customer service assistant device, comprising: The multi-source data acquisition module is used to acquire historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data. The module performs data cleaning and structuring processing on the multi-source heterogeneous data to build a service log library. The data self-labeling module is used to deploy a basic customer service model. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals, performs triple data cleaning on the new multi-source data, including noise removal, deduplication, and structuring processing, determines the corresponding structured data, and performs data self-labeling on the structured data to determine the corresponding labeled dataset. The data self-labeling includes obtaining high-quality work orders from the service log library as seed samples and reusing the labeling rules of the seed samples. The customer service model self-growth module is used to perform lightweight self-training of the basic customer service model using the labeled dataset as the training set, determine the corresponding intelligent customer service model, perform self-evaluation of the intelligent customer service model, determine the corresponding model evaluation score, filter out self-evaluation cases where the model evaluation score is lower than a preset threshold, perform knowledge self-update of the service log library based on the self-evaluation cases, and apply the updated service log library to the processing or training of subsequent new data, so that the intelligent customer service model can perform self-growth closed loop, thereby realizing information technology service desk data processing.

[0013] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the intelligent customer service assistant self-growth method.

[0014] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned intelligent customer service assistant self-growth method.

[0015] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the aforementioned intelligent customer service assistant self-growth method.

[0016] As can be seen from the above technical solution, this application provides a self-growing method and apparatus for intelligent customer service assistants. It acquires historical multi-source heterogeneous data and performs cleaning and structuring processing to construct a service log library containing three elements: user questions, correct answers, and processing results. After deploying a basic customer service model, it performs noise reduction, deduplication, and structuring processing on newly added multi-source data. Using high-quality work orders from the service log library as seed samples, it enables AI to learn annotation rules to automatically annotate new data. Based on the training set obtained from automatic annotation, it uses incremental fine-tuning to perform lightweight self-training on the basic customer service model to obtain an intelligent customer service model. Subsequently, it performs self-evaluation, filters low-scoring cases, and updates the service log library accordingly. The updated knowledge is then applied to the processing or training of subsequent new data, forming a self-growing closed loop, thereby improving the response efficiency and automation quality of the IT service desk. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the self-growth method for intelligent customer service assistants in this application embodiment; Figure 2 This is a structural diagram of the intelligent customer service assistant self-growth device in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0019] Figure label: Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0022] Considering the high concurrency service pressure faced by IT service desks, but the existing intelligent customer service assistants suffer from problems such as lagging knowledge updates, high annotation costs, and difficulty in iteration, this application provides a self-growth method and apparatus for intelligent customer service assistants. It acquires historical multi-source heterogeneous data and performs cleaning and structuring processing to construct a service log library containing three elements: user questions, correct answers, and processing results. After deploying a basic customer service model, it performs noise reduction, deduplication, and structuring processing on newly added multi-source data. Using high-quality work orders from the service log library as seed samples, it enables AI to learn annotation rules to automatically annotate new data. Based on the training set obtained from automatic annotation, it uses incremental fine-tuning to perform lightweight self-training of the basic customer service model to obtain an intelligent customer service model. Subsequently, it performs self-evaluation, filters low-scoring cases, and updates the service log library accordingly. The updated knowledge is applied to the processing or training of subsequent new data, forming a self-growth closed loop, thereby improving the response efficiency and automation quality of the IT service desk.

[0023] To improve the response efficiency and automation quality of the IT help desk, this application provides an embodiment of a self-growing method for intelligent customer service assistants, see [link to embodiment]. Figure 1 The intelligent customer service assistant self-growth method specifically includes the following: Step S101: Obtain historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data; perform data cleaning and structuring processing on the multi-source heterogeneous data; and construct a service log library. Optionally, in this embodiment, the core objective of this step is to integrate all existing data to build an "initial knowledge seed library" for the growth of the intelligent customer service assistant AI, ensuring that the data is searchable and usable.

[0024] Specifically, it involves accessing three types of core data: Telephone recording: Automatically transcribes calls using ASR technology, preserving all content including agent questions, user requests, fault descriptions, and resolution processes. Work order system: Automatically captures all work order fields, including but not limited to: work order title, fault description, fault category, handling steps, solution, whether it is closed-loop, user satisfaction, and reason for transferring to human agent; Existing knowledge documents, such as the IT Help Desk FAQ, troubleshooting manual, and ITSM process specifications, serve as the initial knowledge foundation.

[0025] After the data collection is completed, data cleaning and structuring processing are performed on the above multi-source heterogeneous data.

[0026] Data cleaning methods are well-established, and this embodiment does not impose any limitations. The main goal of cleaning is to remove abnormal data or supplement blank data.

[0027] The structured processing transforms the cleaned data into a unified three-element format of "user problem - correct answer - processing result", clearly labeling key fields such as fault type, root cause, standard solution, and resolution time.

[0028] Output: All data is integrated into a unified "IT Service Log Library". Each piece of data in the log library is stored in a standardized format to ensure that the content is searchable, reusable, and learnable, providing a foundation for subsequent learning.

[0029] This step's service log repository serves as the "initial knowledge seed repository" for subsequent AI self-training, self-labeling, and knowledge updates, fundamentally solving the problems of messy raw data formats, inconsistent quality, and inability to be directly used for model training. It enables AI to begin learning on a clean, standardized, and high-quality data foundation, avoiding the risk of "learning errors" caused by low-quality data. Simultaneously, the unified log repository structure provides a consistent standard for the automated cleaning and storage of newly added daily data, ensuring that the data flow throughout the closed-loop system is traceable and standardized.

[0030] Step S102: Deploy a basic customer service model. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals, performs triple data cleaning on the new multi-source data, including noise removal, deduplication, and structuring processing, determines the corresponding structured data, and performs data self-annotation on the structured data to determine the corresponding labeled dataset. The data self-annotation includes obtaining high-quality work orders from the service log library as seed samples and reusing the annotation rules of the seed samples. Optionally, in this embodiment, this step involves the automated data cleaning and self-labeling of the model. The core self-labeling step can replace manual labeling, reducing the implementation cost.

[0031] Specifically, the automated data cleaning step mainly targets newly accessed raw data. The core objective is to select high-quality and valuable training materials from the 2000+ newly accessed raw data entries each day, remove noisy data, and reduce training redundancy.

[0032] Deploy a basic customer service model based on an existing general-purpose language model or lightweight foundation model, possessing basic language understanding and generation capabilities. After deployment, the model will automatically pull newly added multi-source heterogeneous data from the IT service desk data source at preset time intervals, including newly generated call recordings (already converted to text via ASR), newly created work order records, and dialogue text between agents and users.

[0033] After obtaining the new data, perform three main cleaning actions: Noise reduction: Automatically filters spam calls, test tickets, and invalid duplicate (without substantive content) tickets / call logs; Deduplication: For duplicate work orders with the same fault and the same solution, the best handling case is automatically retained (prioritizing closed-loop work orders with high user satisfaction and high processing efficiency). Structured processing: Automatically classifies and labels the cleaned data, clarifying core information, including: fault type (account, network, software, hardware, permissions, email, etc.), root cause of the fault, standard solution, resolution time, and whether it was resolved in one go.

[0034] Specifically, the core objective of AI self-labeling is to use existing high-quality work orders as "seed samples" to enable AI to automatically learn labeling rules, complete the labeling of new data, and generate a dataset that can be directly used for training.

[0035] The annotation is based on high-quality work orders stored in the service log library. The service log library is a historical knowledge base built in step S101, which contains a large number of verified, standardized, and well-received high-quality cases. These high-quality work orders are automatically extracted from the service log library as seed samples, and the annotation rules of these seed samples are reused.

[0036] Specifically, AI automatically labels structured data with content including but not limited to: fault type, degree of answer standardization, whether it can be solved automatically, whether it is a high-frequency problem, and whether it is a new fault (a fault that did not appear in the initial knowledge base).

[0037] Furthermore, the labeled samples are categorized and stored in the corresponding sample sets, including: Positive sample set: Work orders / call records that have closed loops, positive user feedback, and standardized processing procedures (marked as "high-quality reusable cases"); Challenge / Blind Spot Sample Set: Complex work orders that require multiple transfers to human agents, cannot be answered directly by AI, and require manual intervention from agents (marked as "cases requiring focused learning"). Negative sample set: Cases of incorrect answers, user complaints, and duplicate orders (marked as "cases of errors to be avoided").

[0038] Output: AI-generated labeled training set (all sample sets), updated and expanded daily with new work orders, completely replacing manual labeling work.

[0039] Step S103: The basic customer service model performs lightweight self-training using the labeled dataset as the training set to determine the corresponding intelligent customer service model. The intelligent customer service model is then self-evaluated to determine the corresponding model evaluation score. Self-evaluation cases with model evaluation scores lower than a preset threshold are selected. The service log library is then self-updated based on the self-evaluation cases. The updated service log library is then applied to the processing or training of subsequent new data, enabling the intelligent customer service model to achieve a self-growth closed loop, thereby realizing data processing for the information technology service desk.

[0040] Optionally, in this embodiment, this step involves a complete small closed loop of "self-training → self-evaluation → knowledge update".

[0041] Specifically, the lightweight self-training has the core objective of: to perform lightweight updates to the AI ​​customer service assistant based on the daily newly added labeled training corpus, thereby achieving capability iteration while avoiding the forgetting of old knowledge and not affecting the existing service effect.

[0042] Training method: LoRA incremental fine-tuning is adopted (no need to retrain the base large model, low cost and high efficiency), focusing on updating the IT service desk exclusive capabilities without changing the original basic capabilities of AI.

[0043] Training frequency: When there is less new data added each day, intensive training can be conducted once a week; during peak periods of work orders / call volume, lightweight training can be conducted once a day (each training session should be controlled within 2-4 hours to ensure that the AI ​​can handle services normally).

[0044] Core training objective (focusing on IT service scenarios, avoiding trying to cover everything): Improved fault identification accuracy: It can quickly and accurately identify the fault type corresponding to the user's request; Solution standardization: The output processing steps and solutions strictly conform to ITSM processes and troubleshooting manuals; Reduced referral rate to human agents: Reduce the number of times simple, high-frequency questions are referred to human agents, thus reducing the workload of agents; Improved process adaptability: Adapts to the organization's dedicated processes for dispatching, upgrading, and access verification at the IT service desk.

[0045] Specifically, the core objective of AI self-assessment is to enable AI to "test itself," automatically evaluate training effectiveness, identify knowledge gaps and skill deficiencies, provide clear direction for the next round of training, and form a small closed loop of "training-assessment-optimization."

[0046] Execute 3 actions automatically every day: Automatic question selection: 100-200 questions (covering different fault types) are randomly selected from the high-quality work orders and blind spot / difficult work orders added on the same day as evaluation test questions; Automatic answering and comparison: The trained AI customer service assistant answers questions while retrieving standard answers (or high-quality case answers) processed by agents for two-way comparison; Automatic scoring: Based on preset scoring dimensions, the AI's answer performance is quantitatively scored. The core scoring dimensions include: Fault identification accuracy (whether the fault type is accurately matched). Solution matching rate (whether the answer conforms to the standard process and whether it is feasible); Compliance of the script (whether the answers comply with the IT service desk script specifications); Reasonableness of transferring to manual labor (whether there are situations where manual labor is transferred when it is not necessary, or where manual labor is transferred when it is necessary but not transferred).

[0047] Output results: Automatically filter out cases with scores below a preset threshold (e.g., 80 points), categorize them into three categories: "Knowledge Blind Spot", "Error-Prone Faults", and "Incomprehensible Process", and identify the content to be prioritized for the next round of learning.

[0048] Specifically, the core objective of automatic knowledge updates is to automatically add new knowledge and vulnerabilities discovered during self-assessment to the service log library, ensuring that similar problems can be handled correctly in the future, thus achieving self-accumulation of knowledge.

[0049] Automatic update action: New Fault Handling: For the "knowledge blind spots" (new faults) identified in the evaluation, AI automatically generates corresponding FAQ items based on the agent's standard answer and adds them to the knowledge base; Solution optimization: For "error-prone faults", automatically update the solutions in the original knowledge base, and mark the error-prone points and precautions; Prioritization: High-frequency blind spots and error-prone problems are marked as "priority learning" and will be emphasized in the next training session; Error correction: Mark cases where the AI ​​answers incorrectly as "not for reuse" and provide the correct answer to prevent the AI ​​from making the same mistake again.

[0050] The logic is as follows: new knowledge is discovered during the service process → automatically added to the knowledge base → reinforced in the next training session → directly reused during the service, thus achieving continuous accumulation of knowledge.

[0051] Optionally, in this embodiment, the complete implementation sequence is illustrated with an example: Week 1: Complete ASR voice-to-text integration, work order system interface connection, import existing knowledge documents, complete the structured processing of all data, and build the "IT Service Log Library". Week 2: Launch the basic version of the AI ​​customer service assistant to handle simple inquiries (such as password reset, account login, basic network fault inquiries, and FAQ queries), and simultaneously start testing the data cleaning and self-annotation functions; Weeks 3-4: Establish a complete closed loop of "self-labeling → self-training → self-evaluation → knowledge update", optimize cleaning rules, labeling rules and evaluation thresholds, and ensure that each link runs automatically without obvious loopholes; Starting from the second month: Fully automated closed-loop operation will be started. The AI ​​will complete lightweight self-training once a week, complete self-evaluation and knowledge update daily, and statistical analysis of iteration results (conversion to manual processing rate, accuracy, etc.) will be conducted monthly. Long-term: Continuously optimize closed-loop rules, and with the accumulation of data, gradually enable AI to handle more complex IT fault consultations, thereby achieving a continuous decrease in the rate of transferring to human agents and a continuous increase in user satisfaction.

[0052] This example demonstrates how this embodiment can automatically complete the self-training, self-evaluation, and self-knowledge growth of the AI ​​customer service assistant based on existing call and work order data, achieving a lightweight, low-management, and iterative service throughout the entire process.

[0053] As described above, the intelligent customer service assistant self-growth method provided in this application can construct a service log library containing three elements: user questions, correct answers, and processing results by acquiring historical multi-source heterogeneous data and performing cleaning and structuring processing. After deploying the basic customer service model, the newly added multi-source data is processed for noise reduction, deduplication, and structuring. Based on high-quality work orders in the service log library as seed samples, the AI ​​learns annotation rules to achieve automatic annotation of new data. According to the training set obtained by automatic annotation, the basic customer service model is lightweightly self-trained using incremental fine-tuning to obtain an intelligent customer service model. Subsequently, the model is self-evaluated, low-scoring cases are selected, and the service log library is updated accordingly. The updated knowledge is applied to the processing or training of subsequent new data, forming a self-growth closed loop, thereby improving the response efficiency and automation quality of the IT service desk.

[0054] In one embodiment of the intelligent customer service assistant self-growth method of this application, it may further include the following: Step S201: Perform automatic speech recognition to text processing on the telephone recording data to determine the corresponding text data; Step S202: Perform data cleaning on the work order system data, the knowledge document data, and the text data; extract core fields from the cleaned data; determine the user problem, standard solution, and processing result corresponding to each work order; and build a service log library. The core field extraction includes extracting the work order title, fault description, fault category, processing steps, solution, closed-loop status, user satisfaction, and reason for transferring to human intervention.

[0055] Optionally, in this embodiment, all telephone recording data generated by the IT service desk is acquired, and automatic speech recognition (ASR) technology is used to convert it into editable and searchable text data.

[0056] Specifically, through joint decoding of acoustic and language models, the dialogue content in the audio signal (including user problem descriptions, agent questions and answers, troubleshooting processes, etc.) is transcribed into text in real time or offline. After the conversion is completed, a structured text file is generated for each call record, preserving the time sequence of the dialogue, the speaker's role (user or agent), and the complete semantic content.

[0057] Optionally, in this embodiment, the text data obtained from the previous step, the original records in the work order system, and the existing knowledge document data are cleaned. The specific implementation of the cleaning steps is not limited, but the main purpose is to remove redundant, erroneous, and blank data.

[0058] Optionally, in this embodiment, after cleaning, a core field extraction operation is performed on each data entry to extract key information dimensions from the original text or work order record, including but not limited to: work order title (summarizing the core problem), fault description (a detailed description of the user's original request), fault classification (such as account issues, network failures, software anomalies, etc.), processing steps (the complete operation process from agent receiving the order to resolution), solution (the final standard answer or operation guide), closed-loop status (whether it was resolved in one go, whether multiple follow-ups are required), user satisfaction (the user's rating of the processing result), and reason for transferring to human intervention (if human intervention is involved, the specific reason is recorded). These fields together constitute a complete service case record.

[0059] Finally, the extracted fields are organized according to the triplet structure of "user problem - standard solution - processing result" and stored in the service log library.

[0060] Through step S202, this embodiment successfully constructs a high-quality structured IT service knowledge base. Each piece of data entering the database contains three core elements: question, answer, and result feedback. This provides a clean and standardized data foundation for subsequent AI seed sample selection, annotation rule learning, and model training. At the same time, the complete extraction of core fields enables the system to classify, retrieve, and statistically analyze data according to multiple dimensions such as fault type, satisfaction, and closed-loop status, facilitating the subsequent priority selection of high-quality cases as learning samples.

[0061] In one embodiment of the intelligent customer service assistant self-growth method of this application, it may further include the following: Step S301: Perform noise reduction operation on the newly added multi-source data, including filtering out harassing data, test data, and invalid duplicate data without substantial content from the newly added multi-source data; Step S302: Perform deduplication on the newly added multi-source data, including retaining the best processing case and eliminating the other duplicate cases for duplicate data with the same fault type and the same solution according to a preset priority rule. The priority rule takes into account user satisfaction and processing efficiency. Step S303: Extract classification information from the data after noise reduction and deduplication, and determine the corresponding structured data based on the extracted core field information.

[0062] Optionally, in this embodiment, noise filtering is performed on the newly added multi-source data.

[0063] The system automatically identifies and removes three types of invalid data: first, nuisance data, such as malicious duplicate calls or disruptive call records unrelated to IT services; second, test data, i.e., non-genuine service requests generated by maintenance personnel or the system during debugging; and third, invalid duplicate data without substantive content, such as blank work orders or records containing only symbols without fault descriptions. This filtering mechanism effectively eliminates interference, ensuring that all subsequent processed data consists of genuine and valuable service requests.

[0064] Optionally, in this embodiment, intelligent deduplication is performed on duplicate data with the same fault type and solution.

[0065] The system identifies multiple work orders or call records with the same fault category and handling solution. It then selects the unique optimal handling case according to a preset priority rule, eliminating duplicates. The priority rule considers two dimensions: user satisfaction (prioritizing cases with high user ratings) and processing efficiency (prioritizing cases with short resolution times and fast loop closure). In terms of effectiveness, the system retains the most valuable benchmark cases, ensuring that every sample encountered by the AI ​​during the learning process represents the best practice for that fault type, thereby improving the quality of the model's output solutions.

[0066] Optionally, in this embodiment, information extraction and structure transformation are performed on the data that has undergone noise reduction and deduplication.

[0067] The system automatically identifies and extracts preset core field information from each service record, including but not limited to key elements such as fault type, root cause, standard solution, resolution time, and whether the problem was resolved in one attempt. After extraction, the system reorganizes this unstructured information scattered in the original text into structured data in a unified format.

[0068] Through step S303, this embodiment realizes the structured transformation of the newly accessed raw data every day, which can provide a clear and computable data foundation for subsequent self-labeling and model training.

[0069] In one embodiment of the intelligent customer service assistant self-growth method of this application, it may further include the following: Step S401: Obtain high-quality work orders from the service log library. The high-quality work orders include work orders that have been closed-loop, have user satisfaction higher than a preset threshold, and have standard processing procedures. Using the high-quality work orders as seed samples, construct initial labeling rules based on the labeling rules of the seed samples and the corresponding information technology service management process specifications. Step S402: Perform self-annotation on each sample in the structured data according to the initial annotation rules, wherein the self-annotation includes annotating the fault type, the degree of answer standardization, whether it can be solved automatically, whether it is a high-frequency problem, and whether it is a new fault that has not appeared in the service log library; Step S403: Divide the self-labeled samples into positive samples, difficult samples and negative samples according to sample type, and store them into the corresponding sample sets to determine the corresponding labeled datasets.

[0070] Optionally, in this embodiment, initial annotation rules are constructed.

[0071] High-quality work orders are selected from the existing service log library. The criteria for high-quality work orders are based on three objective conditions: the work order status is closed, the user satisfaction score is higher than a preset threshold (e.g., 4.5 points), and the processing fully complies with the IT service management process specifications.

[0072] Using high-quality work orders as seed samples, the system extracts existing annotation features, such as fault classification methods, standardized descriptions of solutions, and the sequence of processing steps. Simultaneously, the system reads standard rules from the IT service management process specification document, such as fault level classification standards and escalation trigger conditions. The annotation features of the seed sample are then integrated and aligned with the rules in the process specification to form a set of initial annotation rules that can be used for subsequent automatic annotation.

[0073] Optionally, in this embodiment, the structured data is self-annotated.

[0074] For each newly accessed piece of structured data, each data point already contains core fields such as user question, agent response, processing result, closed-loop status, and satisfaction level. The above initial annotation rules are applied to each structured sample, automatically populating the annotation information for each sample.

[0075] Among them, the fault type of the labeled information is determined as account-related, network-related, software-related, hardware-related, permission-related, or email-related, etc.; the standardization of the answer is evaluated into three levels: high, medium, and low, based on whether it matches the standard solution; whether it can be automatically resolved is marked as yes or no, used to filter questions suitable for direct AI response; whether it is a high-frequency question is marked after counting the frequency of similar questions; whether it is a new fault not appearing in the service log library is determined by comparing with historical fault types.

[0076] Optionally, in this embodiment, the self-labeled samples are divided into three types according to preset rules.

[0077] Based on the annotation information, the samples are divided into positive samples, negative samples, and problem samples.

[0078] Positive samples are closed-loop, user-approved, and standardized processing cases, which are marked as high-quality reusable cases and used to train AI to learn the standard processing process.

[0079] The challenging samples are complex work orders that require multiple referrals to human agents, cannot be answered directly by AI, or require manual intervention from agents. These are marked as key learning cases to enhance AI's ability to handle complex problems.

[0080] Negative samples are cases of incorrect answers that lead to user complaints or duplicate orders. These are marked as error cases that need to be avoided and are used to train the AI ​​to avoid repeating the same type of error.

[0081] The three types of samples are stored in their respective sample sets, which together constitute a complete labeled dataset.

[0082] Through step S403, this embodiment successfully transforms high-quality cases and standardized processes accumulated manually into annotation logic that can be executed by the system, thereby achieving automatic conversion of training data and significantly improving data preparation efficiency.

[0083] In one embodiment of the intelligent customer service assistant self-growth method of this application, it may further include the following: Step S501: Using the labeled dataset as the training set, perform lightweight self-training of the basic customer service model using the LoRA incremental fine-tuning method; Step S502: Wherein, the training frequency of the lightweight self-training is dynamically adjusted according to the amount of data.

[0084] Optionally, in this embodiment, after cleaning and self-labeling the newly added multi-source data, the system obtains a high-quality labeled dataset. The dataset includes user questions, corresponding fault type labels, and standardized solutions. At this point, the system uses this labeled dataset as training material to perform lightweight self-training on the already deployed basic customer service model.

[0085] Training employs LoRA incremental fine-tuning technology. This refers to inserting a small number of trainable low-rank matrix modules into the model structure without changing the original base model parameters.

[0086] Specifically, all original weight parameters of the base model are frozen to maintain its original general language understanding and generation capabilities. Simultaneously, low-rank decomposition matrices are added as trainable parameters in specific layers of the model (such as the weight matrix of the attention mechanism). During training, only these newly added low-rank parameters are updated, while all original parameters of the base model remain unchanged.

[0087] In this way, the basic customer service model can quickly learn the fault identification and solution generation capabilities specific to the IT service desk, while fully retaining its original general dialogue and semantic understanding capabilities, thus avoiding the problem of "catastrophic forgetting".

[0088] Optionally, in this embodiment, a dynamic scheduling mechanism is used for the training frequency.

[0089] The system monitors the daily volume of newly labeled datasets and compares it to a preset threshold. When the amount of new data is low and below the preset threshold, it indicates a shortage of high-quality training samples available in the short term. In this case, the system sets the training frequency to once a week to avoid poor training results or overfitting due to insufficient sample size. When the amount of new data reaches or exceeds the preset threshold, it indicates that sufficient training samples have been accumulated that day. The system adjusts the training frequency to once a day to fully utilize the new data for rapid iteration.

[0090] Through step S502, this embodiment successfully achieved lightweight self-training of the model. The self-trained intelligent customer service model has added exclusive capabilities such as accurate identification of IT service desk fault types, standardized output of solutions, and process adaptation on the basis of its original capabilities.

[0091] In one embodiment of the intelligent customer service assistant self-growth method of this application, it may further include the following: Step S601: For the labeled dataset, extract a preset number of evaluation cases according to the preset fault type distribution ratio as a self-evaluation test set, the self-evaluation test set including positive sample cases and difficult sample cases; Step S602: Obtain the first answer of the intelligent customer service model on the self-evaluation test set, retrieve the standard answer of the agent corresponding to the self-evaluation test set as the second answer, and perform multi-dimensional quantitative comparison between the first answer and the second answer to determine the corresponding model evaluation score.

[0092] Optionally, in this embodiment, this step involves frequently self-testing the model to facilitate iterative optimization.

[0093] Specifically, the self-assessment requires sampling newly added work orders daily. The sampling is stratified according to a preset fault type distribution ratio to ensure that the extracted assessment cases can cover various fault scenarios and avoid the assessment results being biased towards a certain high-frequency fault type. The fault type distribution ratio is dynamically determined based on the specific implementation scenario.

[0094] After determining the number of samples for each type of fault, the system randomly selects a corresponding number of cases from the labeled dataset to form a self-evaluation test set. This test set contains two types of samples: one is positive sample cases, which refer to data that has been closed-loop, has good user feedback, and has a standardized processing procedure; the other is challenging sample cases, which refer to complex data that has been repeatedly transferred to human agents, cannot be directly answered by AI, or requires agent intervention. By including both types of samples simultaneously, the system can test the model's mastery of common problems and also evaluate the model's ability to handle complex blind spots.

[0095] Understandably, the core value of negative samples lies in teaching the model to avoid errors during "training," rather than in assessing the model's standard capabilities during "evaluation." Therefore, negative samples are not used for model evaluation during the self-evaluation phase.

[0096] Optionally, in this embodiment, user questions from the self-evaluation test set are input one by one into the trained intelligent customer service model to obtain the first answer generated autonomously by the model. Simultaneously, the system retrieves the corresponding agent's standard answer from the dataset as the second answer. This standard answer typically originates from solutions actually adopted and verified as effective by agents in high-quality work orders. The system then quantitatively compares the first and second answers from multiple preset dimensions.

[0097] The quantitative comparison dimensions include fault identification accuracy, i.e., whether the model correctly judges the type of fault the user belongs to; solution matching rate, i.e. whether the processing steps given by the model are consistent with or equivalent to the standard answer; language compliance, i.e. whether the model's answer conforms to the language norms of the service desk; and the rationality of transferring to human agent, i.e. whether the model judges that the timing of transferring to human agent is appropriate.

[0098] Next, each dimension is scored separately, and the overall score is calculated to obtain the model evaluation score. This score directly reflects the current capability level of the model, and cases with scores below a preset threshold (e.g., 80 points) will be selected as the focus of subsequent knowledge updates.

[0099] Through step S602, this embodiment successfully performed a self-evaluation of the intelligent customer service model, enabling the discovery of the model's weaknesses and laying the foundation for subsequent iterative optimization.

[0100] In one embodiment of the intelligent customer service assistant self-growth method of this application, it may further include the following: Step S701: Classify the self-evaluation cases into difficult cases, error-prone failure cases, and cases of misunderstanding of the process; Step S702: For the aforementioned difficult case, automatically generate corresponding FAQ entries based on the agent's standard answer and add them to the service log library; Step S703: For the error-prone fault cases, update the solutions for the corresponding fault types in the service log library, and mark the error-prone points and precautions; Step S704: For cases where the process is not understood, mark them as priority learning samples for reinforcement in the next round of training; Step S705: Simultaneously, mark incorrect answers as unusable and add correct answers to the service log database.

[0101] Optionally, in this embodiment, the low-scoring cases selected in the self-evaluation process are automatically categorized into three types.

[0102] Challenge (blind spot) cases refer to new faults or complex problems that AI was previously unable to handle and that required human intervention; Error-prone failure cases refer to problems that AI can answer but whose answers are not accurate or complete, or that deviate from standard solutions. Cases of misunderstanding the process refer to situations where the answer provided by the AI ​​is technically correct, but does not conform to the business processes stipulated by the IT service desk, such as dispatching, upgrading, or authorization verification.

[0103] Through this classification, the system can adopt differentiated knowledge update strategies for defects of different natures.

[0104] Optionally, in this embodiment, for records marked as difficult cases, the system automatically extracts the standard solution filled in by the agent when handling the work order and converts it into a structured FAQ entry. This FAQ entry includes fields such as problem description, fault type, standard solution steps, and expected resolution time, and is then automatically added to the service log library. This quickly precipitates the experience of human agents in handling new problems into reusable knowledge, enabling AI to directly call the correct answer when encountering similar problems in the future, thereby reducing the number of times human intervention is required.

[0105] Optionally, in this embodiment, for error-prone fault cases, the existing solutions corresponding to the fault type in the service log library are located and incrementally updated. The updates include supplementing missing processing steps, correcting inaccurate descriptions, and additionally annotating common error points and operational precautions for the fault. By continuously optimizing the accuracy and completeness of existing knowledge, the AI's answers gradually approach the best practices provided by the agent, reducing the error rate for similar questions.

[0106] Optionally, in this embodiment, cases where the model does not understand the process are marked as "priority learning samples" for focused reinforcement during the next round of lightweight self-training. This aims to make the model pay more attention to process standardization issues during training, thereby increasing the AI's adaptability to service desk-specific processes by increasing training weights.

[0107] Optionally, in this embodiment, the system handles cases where the AI's answer is obviously wrong and may lead to user complaints or duplicate orders separately. These cases are marked as "non-reusable" to prevent the AI ​​from using the incorrect answer again in subsequent responses, while the correct answer corrected by a human agent is added to the service log database. This step serves the dual purpose of error isolation and correction, preventing erroneous knowledge from continuously spreading and amplifying within the closed loop.

[0108] Through step S705, this embodiment successfully adopts differentiated knowledge update strategies to address the defects of different natures in the intelligent customer service model during self-evaluation, thereby achieving a self-growing closed-loop iteration of the intelligent customer service model.

[0109] To improve the response efficiency and automation quality of the IT service desk, this application provides an embodiment of an intelligent customer service assistant self-growth device for implementing all or part of the aforementioned intelligent customer service assistant self-growth method. See [link to embodiment]. Figure 2 The intelligent customer service assistant self-generating device specifically includes the following components: The multi-source data acquisition module 10 is used to acquire historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data, and to perform data cleaning and structuring processing on the multi-source heterogeneous data to build a service log library. The data self-labeling module 20 is used to deploy a basic customer service model. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals, performs triple data cleaning on the new multi-source data, including noise removal, deduplication, and structuring processing, determines the corresponding structured data, and performs data self-labeling on the structured data to determine the corresponding labeled dataset. The data self-labeling includes obtaining high-quality work orders from the service log library as seed samples and reusing the labeling rules of the seed samples. The customer service model self-growth module 30 is used to perform lightweight self-training of the basic customer service model using the labeled dataset as the training set, determine the corresponding intelligent customer service model, perform self-evaluation of the intelligent customer service model, determine the corresponding model evaluation score, filter out self-evaluation cases where the model evaluation score is lower than a preset threshold, perform knowledge self-update of the service log library based on the self-evaluation cases, and apply the updated service log library to the processing or training of subsequent new data, so that the intelligent customer service model can perform self-growth closed loop, thereby realizing information technology service desk data processing.

[0110] As described above, the intelligent customer service assistant self-growth device provided in this application embodiment can acquire historical multi-source heterogeneous data and perform cleaning and structuring processing to construct a service log library containing three elements: user questions, correct answers, and processing results. After deploying the basic customer service model, it performs noise reduction, deduplication, and structuring processing on the newly added multi-source data, and uses high-quality work orders in the service log library as seed samples to enable AI to learn annotation rules to achieve automatic annotation of new data. Based on the training set obtained by automatic annotation, it uses incremental fine-tuning to perform lightweight self-training of the basic customer service model to obtain an intelligent customer service model. Subsequently, it performs self-evaluation, filters low-scoring cases, and updates the service log library with knowledge accordingly. The updated knowledge is applied to the processing or training of subsequent new data to form a self-growth closed loop, thereby improving the response efficiency and automation quality of the IT service desk.

[0111] From a hardware perspective, in order to improve the response efficiency and automation quality of the IT service desk, this application provides an embodiment of an electronic device for implementing all or part of the intelligent customer service assistant self-growth method, wherein the electronic device specifically includes the following: The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the intelligent customer service assistant self-growth method and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the intelligent customer service assistant self-growth method in the present embodiment, and the contents of the embodiments of the intelligent customer service assistant self-growth method are incorporated herein, and repeated details will not be described again.

[0112] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0113] In practical applications, the self-growing method for intelligent customer service assistants can be partially executed on the electronic device side as described above, or all operations can be completed on the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed on the client device, the client device may further include a processor.

[0114] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0115] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0116] In one embodiment, the intelligent customer service assistant self-growth method function can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following controls: Step S101: Obtain historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data; perform data cleaning and structuring processing on the multi-source heterogeneous data; and construct a service log library. Step S102: Deploy a basic customer service model. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals, performs triple data cleaning on the new multi-source data, including noise removal, deduplication, and structuring processing, determines the corresponding structured data, and performs data self-annotation on the structured data to determine the corresponding labeled dataset. The data self-annotation includes obtaining high-quality work orders from the service log library as seed samples and reusing the annotation rules of the seed samples. Step S103: The basic customer service model performs lightweight self-training using the labeled dataset as the training set to determine the corresponding intelligent customer service model. The intelligent customer service model is then self-evaluated to determine the corresponding model evaluation score. Self-evaluation cases with model evaluation scores lower than a preset threshold are selected. The service log library is then self-updated based on the self-evaluation cases. The updated service log library is then applied to the processing or training of subsequent new data, enabling the intelligent customer service model to achieve a self-growth closed loop, thereby realizing data processing for the information technology service desk.

[0117] As described above, the electronic device provided in this application embodiment acquires historical multi-source heterogeneous data and performs cleaning and structuring processing to construct a service log library containing three elements: user questions, correct answers, and processing results. After deploying a basic customer service model, it performs noise reduction, deduplication, and structuring processing on newly added multi-source data. Based on high-quality work orders in the service log library as seed samples, it enables AI to learn annotation rules to achieve automatic annotation of new data. According to the training set obtained from automatic annotation, it uses incremental fine-tuning to perform lightweight self-training of the basic customer service model to obtain an intelligent customer service model. Subsequently, it performs self-evaluation, filters low-scoring cases, and updates the service log library with knowledge accordingly. The updated knowledge is applied to the processing or training of subsequent new data, forming a self-growing closed loop, thereby improving the response efficiency and automation quality of the IT service desk.

[0118] In another embodiment, the intelligent customer service assistant self-growth method can be configured separately from the central processing unit 9100. For example, the intelligent customer service assistant self-growth method can be configured as a chip connected to the central processing unit 9100, and the function of the intelligent customer service assistant self-growth method can be realized through the control of the central processing unit.

[0119] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technologies.

[0120] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0121] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0122] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0123] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0124] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0125] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0126] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 9130 is also coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0127] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the self-growing intelligent customer service assistant method with a server or client as the execution subject in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the self-growing intelligent customer service assistant method with a server or client as the execution subject in the above embodiments. For example, when the processor executes the computer program, it implements the following steps: Step S101: Obtain historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data; perform data cleaning and structuring processing on the multi-source heterogeneous data; and construct a service log library. Step S102: Deploy a basic customer service model. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals, performs triple data cleaning on the new multi-source data, including noise removal, deduplication, and structuring processing, determines the corresponding structured data, and performs data self-annotation on the structured data to determine the corresponding labeled dataset. The data self-annotation includes obtaining high-quality work orders from the service log library as seed samples and reusing the annotation rules of the seed samples. Step S103: The basic customer service model performs lightweight self-training using the labeled dataset as the training set to determine the corresponding intelligent customer service model. The intelligent customer service model is then self-evaluated to determine the corresponding model evaluation score. Self-evaluation cases with model evaluation scores lower than a preset threshold are selected. The service log library is then self-updated based on the self-evaluation cases. The updated service log library is then applied to the processing or training of subsequent new data, enabling the intelligent customer service model to achieve a self-growth closed loop, thereby realizing data processing for the information technology service desk.

[0128] As described above, the computer-readable storage medium provided in this application embodiment acquires historical multi-source heterogeneous data and performs cleaning and structuring processing to construct a service log library containing three elements: user questions, correct answers, and processing results. After deploying a basic customer service model, the newly added multi-source data is processed for noise reduction, deduplication, and structuring. Based on high-quality work orders in the service log library as seed samples, the AI ​​learns annotation rules to achieve automatic annotation of new data. According to the training set obtained from automatic annotation, the basic customer service model is lightweightly self-trained using an incremental fine-tuning method to obtain an intelligent customer service model. Subsequently, it performs self-evaluation, filters low-scoring cases, and updates the service log library with knowledge accordingly. The updated knowledge is applied to the processing or training of subsequent new data, forming a self-growing closed loop, thereby improving the response efficiency and automation quality of the IT service desk.

[0129] Embodiments of this application also provide a computer program product capable of implementing all steps of the intelligent customer service assistant self-growth method with the execution subject being a server or client in the above embodiments. When the computer program / instructions are executed by a processor, they implement the steps of the intelligent customer service assistant self-growth method. For example, the computer program / instructions implement the following steps: Step S101: Obtain historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data; perform data cleaning and structuring processing on the multi-source heterogeneous data; and construct a service log library. Step S102: Deploy a basic customer service model. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals, performs triple data cleaning on the new multi-source data, including noise removal, deduplication, and structuring processing, determines the corresponding structured data, and performs data self-annotation on the structured data to determine the corresponding labeled dataset. The data self-annotation includes obtaining high-quality work orders from the service log library as seed samples and reusing the annotation rules of the seed samples. Step S103: The basic customer service model performs lightweight self-training using the labeled dataset as the training set to determine the corresponding intelligent customer service model. The intelligent customer service model is then self-evaluated to determine the corresponding model evaluation score. Self-evaluation cases with model evaluation scores lower than a preset threshold are selected. The service log library is then self-updated based on the self-evaluation cases. The updated service log library is then applied to the processing or training of subsequent new data, enabling the intelligent customer service model to achieve a self-growth closed loop, thereby realizing data processing for the information technology service desk.

[0130] As described above, the computer program product provided in this application embodiment acquires historical multi-source heterogeneous data and performs cleaning and structuring processing to construct a service log library containing three elements: user questions, correct answers, and processing results. After deploying a basic customer service model, it performs noise reduction, deduplication, and structuring processing on newly added multi-source data. Based on high-quality work orders in the service log library as seed samples, it enables AI to learn annotation rules to achieve automatic annotation of new data. According to the training set obtained from automatic annotation, it uses incremental fine-tuning to perform lightweight self-training of the basic customer service model to obtain an intelligent customer service model. Subsequently, it performs self-evaluation, filters low-scoring cases, and updates the service log library with knowledge accordingly. The updated knowledge is applied to the processing or training of subsequent new data, forming a self-growing closed loop, thereby improving the response efficiency and automation quality of the IT service desk.

[0131] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0133] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0134] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0135] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A self-growing method for an intelligent customer service assistant, characterized in that, The method includes: Acquire historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data; perform data cleaning and structuring processing on the multi-source heterogeneous data; and construct a service log library. A basic customer service model is deployed. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals. The new multi-source data is cleaned in three ways, including noise removal, deduplication, and structuring. The corresponding structured data is determined, and the structured data is self-labeled to determine the corresponding labeled dataset. The self-labeling includes obtaining high-quality work orders from the service log library as seed samples and reusing the labeling rules of the seed samples. The basic customer service model uses the labeled dataset as the training set for lightweight self-training to determine the corresponding intelligent customer service model. The intelligent customer service model is then self-evaluated to determine the corresponding model evaluation score. Self-evaluation cases with model evaluation scores below a preset threshold are selected. The service log library is then self-updated based on the self-evaluation cases. The updated service log library is then applied to the processing or training of subsequent new data, enabling the intelligent customer service model to achieve a self-growth closed loop and thus realize data processing for the information technology service desk.

2. The self-growing method for intelligent customer service assistants according to claim 1, characterized in that, The step of cleaning and structuring the multi-source heterogeneous data to build a service log library includes: The recorded telephone data is automatically processed by speech recognition to convert it to text, and the corresponding text data is determined. Data cleaning is performed on the work order system data, the knowledge document data, and the text data. Core fields are extracted from the cleaned data to determine the user problem, standard solution, and processing result corresponding to each work order, and a service log library is constructed. The core field extraction includes extracting the work order title, fault description, fault category, processing steps, solution, closed-loop status, user satisfaction, and reason for transferring to human intervention.

3. The self-growing method for intelligent customer service assistants according to claim 1, characterized in that, The process of performing triple data cleaning on the newly added multi-source data includes noise removal, deduplication, and structured processing to determine the corresponding structured data, including: The newly added multi-source data is subjected to noise reduction, including filtering out harassing data, test data, and invalid duplicate data without substantial content from the newly added multi-source data; The newly added multi-source data is deduplicated, including for duplicate data with the same fault type and the same solution, retaining the best processing case and eliminating the other duplicate cases according to a preset priority rule, which takes into account user satisfaction and processing efficiency. The data that has undergone noise reduction and deduplication is classified and information is extracted. The corresponding structured data is determined based on the extracted core field information.

4. The self-growing method for intelligent customer service assistants according to claim 1, characterized in that, The step of performing data self-annotation on the structured data to determine the corresponding labeled dataset includes: Obtain high-quality work orders from the service log library. These high-quality work orders include those that have been closed-loop, have user satisfaction exceeding a preset threshold, and have standardized processing procedures. Using these high-quality work orders as seed samples, construct initial labeling rules based on the labeling rules of the seed samples and the corresponding information technology service management process specifications. Each sample in the structured data is self-labeled according to the initial labeling rules. The self-labeling includes labeling the fault type, the degree of answer standardization, whether it can be solved automatically, whether it is a high-frequency problem, and whether it is a new fault that has not appeared in the service log library. The self-labeled samples are divided into positive samples, difficult samples, and negative samples according to their sample types, and stored in the corresponding sample sets to determine the corresponding labeled datasets.

5. The self-growing method for intelligent customer service assistants according to claim 1, characterized in that, The basic customer service model uses the labeled dataset as the training set for lightweight self-training to determine the corresponding intelligent customer service model, including: Using the labeled dataset as the training set, the basic customer service model is lightweightly self-trained using the LoRA incremental fine-tuning method; The training frequency of the lightweight self-training is dynamically adjusted according to the amount of data.

6. The self-growing method for intelligent customer service assistants according to claim 1, characterized in that, The self-evaluation of the intelligent customer service model to determine the corresponding model evaluation score includes: For the labeled dataset, a preset number of evaluation cases are extracted according to the preset fault type distribution ratio as a self-evaluation test set, which includes positive sample cases and difficult sample cases; The intelligent customer service model obtains its first answer on the self-evaluation test set, retrieves the corresponding agent's standard answer from the self-evaluation test set as the second answer, and performs a multi-dimensional quantitative comparison between the first answer and the second answer to determine the corresponding model evaluation score.

7. The self-growing method for intelligent customer service assistants according to claim 1, characterized in that, The step of updating the service log database based on the self-evaluation cases includes: The self-assessment cases are categorized into difficult cases, error-prone failure cases, and cases of misunderstanding of the process. For the aforementioned challenging cases, corresponding FAQ entries are automatically generated based on the agent's standard answer and added to the service log library; For the aforementioned error-prone fault cases, update the solutions for the corresponding fault types in the service log database, and mark the error-prone points and precautions; Cases where the process is not understood are marked as priority learning samples for reinforcement in the next round of training; At the same time, cases with incorrect answers are marked as unusable and the correct answers are added to the service log library.

8. A self-growing device for an intelligent customer service assistant, characterized in that, The device includes: The multi-source data acquisition module is used to acquire historical multi-source heterogeneous data from the information technology service desk, including telephone recording data, work order system data, and knowledge document data. The module performs data cleaning and structuring processing on the multi-source heterogeneous data to build a service log library. The data self-labeling module is used to deploy a basic customer service model. The basic customer service model receives new multi-source data from the information technology service desk at preset time intervals, performs triple data cleaning on the new multi-source data, including noise removal, deduplication, and structuring processing, determines the corresponding structured data, and performs data self-labeling on the structured data to determine the corresponding labeled dataset. The data self-labeling includes obtaining high-quality work orders from the service log library as seed samples and reusing the labeling rules of the seed samples. The customer service model self-growth module is used to perform lightweight self-training of the basic customer service model using the labeled dataset as the training set, determine the corresponding intelligent customer service model, perform self-evaluation of the intelligent customer service model, determine the corresponding model evaluation score, filter out self-evaluation cases where the model evaluation score is lower than a preset threshold, perform knowledge self-update of the service log library based on the self-evaluation cases, and apply the updated service log library to the processing or training of subsequent new data, so that the intelligent customer service model can perform self-growth closed loop, thereby realizing information technology service desk data processing.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent customer service assistant self-growth method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the self-growing method for an intelligent customer service assistant as described in any one of claims 1 to 7.

Images