An operation and maintenance management method and system based on intelligent customer service model
By generating and managing event identification and event workflow for exception question and answer information, combined with automated operation and maintenance scripts, the problem of poor information transmission in the operation and maintenance management of intelligent customer service models is solved, and the efficiency and accuracy of exception handling are improved.
Patent Information
- Application Number
- CN202411271214.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-11
AI Technical Summary
In the operation and maintenance management of intelligent customer service models, information transmission between front-end customer service and back-end operation and maintenance teams often has poor communication or misunderstanding, resulting in the formation of information islands, and duplicate abnormal problems cannot be responded and resolved in a timely manner, which has increased The work burden of operation and maintenance personnel.
By generating event identifiers for exception question and answer information and assigning event workflows, performing event workflows to obtain exception processing results, and updating event workflows based on exception processing results, using information data sources such as question and answer caches, business databases and natural language models to automatically operate Dimension scripts execute exception repair strategies to realize information transmission standardization and real-time tracking.
Reduces miscommunication or misunderstanding, improves the efficiency of the operation and maintenance team in handling duplicate exception problems, realizes real-time tracking of event processing progress, reduces the work burden of operation and maintenance personnel, and improves the timeliness and accuracy of exception handling. .
Smart Images

Figure CN118798617B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to an operation and maintenance management method and system based on an intelligent customer service model. Background Art
[0002] With the rapid development of information technology, the application of intelligent customer service models in corporate business operations is becoming more and more extensive. Based on natural language processing technology and machine learning algorithms, intelligent customer service models can not only instantly understand and analyze user inquiries, directly generate replies for external users, but also provide business knowledge support to manual customer service, thereby improving overall service efficiency and quality.
[0003] Among them, Retrieval-Augmented Generation (RAG) has been widely used in intelligent customer service models. Through the method of retrieval-augmented generation, the model has richer background information when generating responses, so as to better meet the needs of users. However, as the query content processed by the intelligent customer service model becomes increasingly complex, various anomalies are also occurring more frequently.
[0004] Therefore, in order to solve the anomalies of the intelligent customer service model in a timely and effective manner, close collaboration between the front-end customer service and the back-end operation and maintenance department supported by the intelligent customer service model is crucial. When the feedback of the intelligent customer service model is abnormal, the front-end customer service needs to pass the abnormal status to the back-end operation and maintenance team as soon as possible for processing. However, in the existing operation and maintenance management method, miscommunication or misunderstanding often occurs during the information transmission process, and the lack of a unified exception management system often leads to the formation of information islands. The operation and maintenance team may face a large number of unclassified or unarchived exception feedback, which makes repeated exception problems unable to be responded to and resolved in a timely manner, and increases the workload of the operation and maintenance personnel. Summary of the invention
[0005] The present invention provides an operation and maintenance management method and system based on an intelligent customer service model to solve at least one problem mentioned in the above background technology.
[0006] The specific technical solutions provided by this application are as follows:
[0007] An operation and maintenance management method based on an intelligent customer service model comprises the following steps:
[0008] Generate event identification for abnormal Q&A information and assign event workflow;
[0009] Execute the event workflow to obtain the exception handling results;
[0010] Update the event workflow based on the exception handling results;
[0011] The method of generating an event identifier for abnormal question and answer information and assigning an event workflow comprises the following steps:
[0012] Collect abnormal question and answer information of the intelligent customer service model, and obtain the information data source of the abnormal question and answer information; the information data source includes question and answer cache, business database, knowledge base and natural language model;
[0013] Determine the event identifier and event workflow of abnormal question and answer information according to the information data source; the event workflow includes several event identifiers and corresponding word vector representations;
[0014] The execution event workflow obtains the exception handling result, including the steps of:
[0015] Determine a first monitoring instruction according to the information data source and word vector of the event workflow;
[0016] Acquiring back-end monitoring data of the information data source through a first monitoring instruction;
[0017] Determine the exception repair strategy based on the backend monitoring data, and execute the exception repair strategy through automated operation and maintenance scripts to obtain the exception handling results.
[0018] As a preferred solution, before collecting abnormal question and answer information of the intelligent customer service model, the following steps are included:
[0019] receiving input current questions and context information, and generating standardized questions according to the current questions and context information;
[0020] Search the question and answer cache according to the standardized question. If a match is successful, the question and answer cache is recorded as the information data source.
[0021] Otherwise, the business database is searched according to the preset regular matching rules and vocabulary matching rules. If the match is successful, the business database is recorded as the information data source;
[0022] Otherwise, keyword matching is performed according to the standardized question to obtain the knowledge base matching result. If the match is successful, the knowledge base is recorded as the information data source;
[0023] Otherwise, the natural language model is recorded as the information data source;
[0024] A Prompt template is generated based on the current question, context information and information data source, and the natural language model outputs answer information according to the Prompt template and associates the answer information with the information data source.
[0025] As a preferred solution, the process of determining the event identifier and event workflow of abnormal question and answer information according to the information data source includes the following steps:
[0026] Determine the event identifier of abnormal question and answer information based on the information data source;
[0027] Acquire a first event workflow set according to the information data source; the first event workflow set is a set of event workflows classified according to the information data source;
[0028] Obtaining a word vector representation of the abnormal question and answer information, and calculating a matching degree between the word vector representation and each event workflow in the first event workflow set;
[0029] An event workflow of abnormal question and answer information is determined according to the event identifier, word vector representation and matching degree; if there is an event workflow in the first event workflow set whose matching degree is not lower than a preset matching threshold, the event identifier and word vector representation are added to the event workflow with the highest matching degree; if the matching degrees of all event workflows in the first event workflow set are lower than the preset matching threshold, a new event workflow is created according to the event identifier and word vector representation.
[0030] As a preferred solution, the step of obtaining a word vector representation of abnormal question and answer information includes the following steps:
[0031] Use the WordPiece word segmentation method to split the input abnormal question and answer information into a token sequence, insert a [CLS] tag before the token sequence, and insert a [SEP] tag at the end of the token sequence to obtain the preprocessed abnormal question and answer information;
[0032] Input the preprocessed abnormal question and answer information into several Transformer layers;
[0033] The vector representation corresponding to the [CLS] tag is extracted from the Transformer of the last layer as the word vector representation of the entire abnormal question and answer information.
[0034] As a preferred solution, the calculating of the matching degree between the word vector representation and each event workflow in the first event workflow set comprises the steps of:
[0035] Obtaining a word vector representation of each event workflow in the first event workflow set;
[0036] Calculate the cosine similarity between the word vector representation of the abnormal question and answer information and each word vector representation in the first event workflow set;
[0037] The average value of the cosine similarity of each event workflow is calculated respectively, and the obtained result is recorded as the matching degree of the event workflow.
[0038] As a preferred solution, the method of determining the first monitoring instruction according to the information data source and word vector of the event workflow includes the following steps:
[0039] Determine the target data source based on the information data source of the event workflow, and mark the abnormal word vectors in the word vector representation of the event workflow;
[0040] Cluster the abnormal word vectors of the event workflow through a clustering algorithm to obtain key target parameter groups;
[0041] A first monitoring instruction is determined according to a target data source and a key target parameter group.
[0042] As a preferred solution, the first monitoring instruction is a natural language instruction; the method of obtaining the back-end monitoring data of the information data source through the first monitoring instruction comprises the following steps:
[0043] Parsing the first monitoring instruction to obtain a target data source and a key target parameter group in the first monitoring instruction;
[0044] Generate SQL monitoring instructions based on the target data source and key target parameter group;
[0045] Connect to the target data source through a database connection tool or API and send SQL monitoring instructions to the target data source;
[0046] Extract backend monitoring data from the query results returned by the database.
[0047] As a preferred solution, the method of determining an abnormality repair strategy based on backend monitoring data includes the following steps:
[0048] Analyze backend monitoring data to determine the type of anomaly and the scope of impact;
[0049] Determine anomaly repair strategy and generate repair plan;
[0050] The execution exception repair strategy includes fine-tuning the natural language model, updating the cache strategy, updating the business database and updating the knowledge base.
[0051] As a preferred solution, the updating of the event workflow according to the exception handling result comprises the following steps:
[0052] Based on the exception handling result obtained by executing the exception repair strategy through the automated operation and maintenance script, extract the status information in the exception handling result; the status information includes successful processing, failed processing, processing in progress and archived status;
[0053] Traverse the status information of all event identifiers in the event workflow, and update the event status information to archived for successfully processed event identifiers; store the archived event identifiers from the pending event storage area of the event workflow to the archived event storage area; for event identifiers that have been in a state of long-term processing failure, obtain the word vector representation and exception processing log corresponding to the event identifier, and reallocate the event workflow based on the event identifier, word vector representation and exception processing log.
[0054] The present invention also provides an operation and maintenance management system based on an intelligent customer service model, which is applied to the operation and maintenance management method based on an intelligent customer service model as described above, comprising:
[0055] An event identification management module is used to determine the event identification of abnormal question and answer information according to the information data source;
[0056] An event workflow management module is used to determine the event workflow of abnormal question and answer information according to the information data source, and to update the event workflow according to the abnormal processing result;
[0057] A data monitoring module, used to determine a first monitoring instruction according to the information data source and word vector of the event workflow, and obtain backend monitoring data of the information data source through the first monitoring instruction;
[0058] The exception repair module is used to determine the exception repair strategy based on the backend monitoring data, and to execute the exception repair strategy through the automated operation and maintenance script to obtain the exception processing results.
[0059] Compared with the prior art, the present invention has the following beneficial effects:
[0060] This application collects abnormal question and answer information of the intelligent customer service model, determines the event identifier and event workflow of the abnormal question and answer information according to the information data source of the abnormal question and answer information, and realizes the standardization of information transmission between the front-end customer service and the back-end operation and maintenance team, thereby reducing miscommunication or misunderstandings, and enabling the operation and maintenance team to handle repeated abnormal problems more efficiently; obtains the exception handling results by executing the event workflow, and updates the event workflow according to the exception handling results, realizes real-time tracking of event handling progress, and executes exception repair strategies through automated operation and maintenance scripts, thereby reducing the workload of operation and maintenance personnel and improving the timeliness and accuracy of exception handling. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0063] Figure 1 A flowchart of an operation and maintenance management method based on an intelligent customer service model provided by an embodiment of the present invention;
[0064] Figure 2 A schematic diagram of a process for obtaining an exception handling result by executing an event workflow provided in an embodiment of the present invention;
[0065] Figure 3 A schematic diagram of a process for an intelligent customer service model provided in an embodiment of the present invention to output answer information according to a current question and context information and to associate the answer information with an information data source;
[0066] Figure 4 A schematic diagram of a flow chart of an event identification and an event workflow for determining abnormal question and answer information according to an information data source provided by an embodiment of the present invention;
[0067] Figure 5 A schematic diagram of a flow chart for calculating the degree of matching between the word vector representation and each event workflow in the first event workflow set provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0069] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.
[0070] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.
[0071] With the rapid development of information technology, the application of intelligent customer service models in corporate business operations is becoming more and more extensive. Based on natural language processing technology and machine learning algorithms, intelligent customer service models can not only instantly understand and analyze user inquiries, directly generate replies for external users, but also provide business knowledge support to manual customer service, thereby improving overall service efficiency and quality.
[0072] Among them, Retrieval-Augmented Generation (RAG) has been widely used in intelligent customer service models. Through the method of retrieval-augmented generation, the model has richer background information when generating responses, so as to better meet the needs of users. However, as the query content processed by the intelligent customer service model becomes increasingly complex, various anomalies are also occurring more frequently.
[0073] Therefore, in order to solve the anomalies of the intelligent customer service model in a timely and effective manner, close collaboration between the front-end customer service and the back-end operation and maintenance department supported by the intelligent customer service model is crucial. When the feedback of the intelligent customer service model is abnormal, the front-end customer service needs to pass the abnormal status to the back-end operation and maintenance team as soon as possible for processing. However, in the existing operation and maintenance management method, miscommunication or misunderstanding often occurs during the information transmission process, and the lack of a unified exception management system often leads to the formation of information islands. The operation and maintenance team may face a large number of unclassified or unarchived exception feedback, which makes repeated exception problems unable to be responded to and resolved in a timely manner, and increases the workload of the operation and maintenance personnel.
[0074] To solve the above problems, please refer to Figure 1 and Figure 2 The present invention provides an operation and maintenance management method based on an intelligent customer service model, comprising the steps of:
[0075] Generate event identification for abnormal Q&A information and assign event workflow;
[0076] Execute the event workflow to obtain the exception handling results;
[0077] Update the event workflow based on the exception handling results;
[0078] The method of generating an event identifier for abnormal question and answer information and assigning an event workflow comprises the following steps:
[0079] Collect abnormal question and answer information of the intelligent customer service model, and obtain the information data source of the abnormal question and answer information; the information data source includes question and answer cache, business database, knowledge base and natural language model;
[0080] Determine the event identifier and event workflow of abnormal question and answer information according to the information data source; the event workflow includes several event identifiers and corresponding word vector representations;
[0081] The execution event workflow obtains the exception handling result, including the steps of:
[0082] Determine a first monitoring instruction according to the information data source and word vector of the event workflow;
[0083] Acquiring back-end monitoring data of the information data source through a first monitoring instruction;
[0084] Determine the exception repair strategy based on the backend monitoring data, and execute the exception repair strategy through automated operation and maintenance scripts to obtain the exception handling results.
[0085] This application collects abnormal question and answer information of the intelligent customer service model, determines the event identifier and event workflow of the abnormal question and answer information according to the information data source of the abnormal question and answer information, and realizes the standardization of information transmission between the front-end customer service and the back-end operation and maintenance team, thereby reducing miscommunication or misunderstandings, and enabling the operation and maintenance team to handle repeated abnormal problems more efficiently; obtains the exception handling results by executing the event workflow, and updates the event workflow according to the exception handling results, realizes real-time tracking of event handling progress, and executes exception repair strategies through automated operation and maintenance scripts, thereby reducing the workload of operation and maintenance personnel and improving the timeliness and accuracy of exception handling.
[0086] Specifically, each step of the operation and maintenance management method based on the intelligent customer service model of the present invention is described in detail through the following content:
[0087] An operation and maintenance management method based on an intelligent customer service model, comprising:
[0088] Generate event identification for abnormal Q&A information and assign event workflow;
[0089] Execute the event workflow to obtain the exception handling results;
[0090] Update the event workflow based on the exception handling results.
[0091] The method of generating an event identifier for abnormal question and answer information and assigning an event workflow comprises the following steps:
[0092] SC1. Collect abnormal question and answer information of the intelligent customer service model and obtain the information data source of the abnormal question and answer information; the information data source includes question and answer cache, business database, knowledge base and natural language model;
[0093] In one embodiment, please refer to Figure 3 Before collecting abnormal question and answer information of the intelligent customer service model, the method further includes the following steps:
[0094] The intelligent customer service model outputs answer information based on the current question and context information and associates the answer information with the information data source, specifically including the following steps:
[0095] receiving input current questions and context information, and generating standardized questions according to the current questions and context information;
[0096] Search the question and answer cache according to the standardized question. If a match is successful, the question and answer cache is recorded as the information data source.
[0097] Otherwise, the business database is searched according to the preset regular matching rules and vocabulary matching rules. If the match is successful, the business database is recorded as the information data source;
[0098] Otherwise, keyword matching is performed according to the standardized question to obtain the knowledge base matching result. If the match is successful, the knowledge base is recorded as the information data source;
[0099] Otherwise, the natural language model is recorded as the information data source;
[0100] A Prompt template is generated based on the current question, context information and information data source, and the natural language model outputs answer information according to the Prompt template and associates the answer information with the information data source.
[0101] The Q&A cache is used to store and quickly retrieve answers to common questions in the past to reduce repeated calculations and response time. Specifically, the Q&A cache records the questions asked by users in the past and their corresponding answers, and uses key-value storage (such as Redis and Memcached) to speed up query speed.
[0102] The business database is used to store business data related to daily operations, supporting business operations, transaction processing and analysis. The business database includes actual business data such as customer information, transaction records, inventory data, order status, etc. The business database design focuses on data consistency and transaction processing capabilities, supporting complex queries and data analysis.
[0103] The knowledge base is used to provide structured and verified information to support the question-answering system and decision-making process. The knowledge base integrates various knowledge articles, FAQs, guides, tutorials, documents, flow charts, etc., aiming to provide common and reusable knowledge in various fields involved in the intelligent customer service model, which is convenient for quick retrieval and reference.
[0104] This embodiment provides a question-answering processing process of an intelligent customer service model. After receiving the question and context information input by the user, the intelligent customer service model first performs standardization processing on the information, including word segmentation, stop word removal, and part-of-speech tagging, to generate a standardized question. Subsequently, based on the standardized question, the model first searches the question-answering cache, and if a matching result is found, the question-answering cache is recorded as the information data source.
[0105] If no matching result is found in the cache, the model further searches the business database according to the preset regular matching rules and vocabulary matching rules. If a matching result is found in the business database, the business database is recorded as the information data source.
[0106] If no matching result is found in the business database, the model will further perform keyword matching in the knowledge base to obtain relevant information as the information data source. If a matching result is found in the knowledge base, the knowledge base is recorded as the information data source.
[0107] When none of the above search methods finds qualifying results, the model needs to process the current question and context information directly through the natural language model. At this time, the natural language model is recorded as the information data source. In the normal use of the intelligent customer service model, there may be a failure to match the cache, business database, and knowledge base. For example, questions of a casual or chat nature are usually not within the scope of the business database and knowledge base; questions are abnormal or incomplete, and the system cannot accurately segment and match words due to serious spelling errors or omissions of important information. New knowledge points are missing, the business database and knowledge base have not been updated, and users' questions involve new business or new function issues, such as "Introducing the latest version of the new xx function", but the system content update has not kept up with the time, such as the company's new products do not have complete question and answer pairs.
[0108] Based on the current question, context information and information data source (the actual content can be empty when the information data source is a natural language model), the model generates a Prompt template. The natural language model outputs answer information according to the Prompt template, associates the answer information with the corresponding information data source and integrates it into regular logs and system records.
[0109] It can be understood that the steps before collecting abnormal question and answer information of the intelligent customer service model provided in this embodiment provide a dialogue workflow of the intelligent customer service model, covering functional modules such as user input processing, data retrieval, natural language generation, information recording and feedback. The abnormal question and answer information refers to the answer information corresponding to the abnormal situation or problem that occurs during the question and answer process. By associating the answer information with the information data source in the above dialogue workflow of the intelligent customer service model, it is possible to facilitate the subsequent collection and improvement of abnormal answer information.
[0110] SC2. Determine the event identifier and event workflow of abnormal question and answer information according to the information data source; the event workflow includes several event identifiers and corresponding word vector representations;
[0111] For further information, see Figure 4 The process of determining the event identifier and event workflow of abnormal question and answer information according to the information data source comprises the following steps:
[0112] SC21. Determine the event identifier of the abnormal question and answer information based on the information data source; the event identifier is used to track and manage abnormal events, for example, a unique event identifier is generated using a timestamp + information data source code + serial number. The information data source code may include NLM (natural language model), QAC (question and answer cache), DB (business database) and KB (knowledge base). For example, based on the timestamp: 20231004, information data source code: NLM (natural language model), serial number: 001, the event identifier is generated: 20231004-NLM-001.
[0113] SC22. Acquire a first event workflow set according to the information data source;
[0114] The first event workflow set is a set of event workflows classified according to information data sources.
[0115] The event workflow in this embodiment is a structured method for tracking and managing the processing of abnormal question and answer information. The event workflow includes the event workflow including several event identifiers and word vector representations corresponding to each event identifier. In one embodiment, the event workflow also includes workflow configuration information; wherein the workflow configuration information is used to store the personnel information, tool information and process template of the event workflow. The first event workflow set in this embodiment is an event workflow classified according to different information data sources. For example, corresponding workflows can be defined for NLM, QAC, DB and KB respectively.
[0116] SC23. Obtain a word vector representation of the abnormal question and answer information, and calculate a matching degree between the word vector representation and each event workflow in the first event workflow set;
[0117] The word vector representation of the abnormal question and answer information is obtained by using word vector models such as TF-IDF, Word2Vec, GloVe, and BERT. If the Word2Vec or GloVe model is selected, the word vector mean of each word in the question and answer information can be obtained as the word vector representation. If a context model such as BERT is selected, the word vector representation of its output can be directly obtained.
[0118] In one embodiment, the step of obtaining the word vector representation of the abnormal question and answer information is to obtain the word vector representation of the abnormal question and answer information through a BERT model, including the following steps:
[0119] Preprocessing: Use the WordPiece word segmentation method to split the input abnormal question and answer information into several more fine-grained token sequences (i.e., word fragments). Then insert the [CLS] tag (representing the overall characteristics of the sequence) before the token sequence and the [SEP] tag (indicating the end of the sequence) at the end of the token sequence to obtain the preprocessed abnormal question and answer information.
[0120] The preprocessed abnormal question and answer information is in the input format that the BERT model can process. The preprocessed abnormal question and answer information includes: Token sequence, which is a sequence composed of the ID of each token generated by the word segmenter; Token type ID, which is used to distinguish different sentences (in the question and answer task, the token type is at least divided into questions and answers); position encoding, which is used to indicate the position of each token sequence in the sequence.
[0121] Obtain the vector representation of each token sequence: Input the preprocessed abnormal question and answer information into several Transformer layers;
[0122] The BERT model uses a bidirectional Transformer architecture, which includes multiple layers of self-attention mechanisms, each of which can capture semantic information at different levels. The self-attention mechanism allows each token sequence to interact with other token sequences to generate context-sensitive representations. This allows BERT to understand the deep semantics and complex relationships in sentences; the bidirectional Transformer architecture means that the BERT model uses a bidirectional encoder, which processes text sequences from left to right and from right to left at the same time, and can more comprehensively understand contextual information.
[0123] Extract word vector representation of abnormal question and answer information: Extract the vector representation corresponding to the [CLS] tag from the last layer of Transformer as the word vector representation of the entire abnormal question and answer information. Based on the above content, the [CLS] tag is designed by BERT to represent the summary feature representation of the entire input sequence. Using it as a word vector representation can effectively represent the context-sensitive information of abnormal question and answer information, and can accurately capture the semantic relationship and more complex language features in the sentence.
[0124] See also Figure 5 In one embodiment, the calculating of the matching degree between the word vector representation and each event workflow in the first event workflow set comprises the steps of:
[0125] SC231. Obtain a word vector representation of each event workflow in the first event workflow set;
[0126] Each event workflow in the first event workflow set is traversed, and the word vector representation of each event identifier of the event workflow is further traversed; wherein the word vector representation of each event identifier can be obtained and stored through the BERT model according to the method of the aforementioned embodiment.
[0127] SC232. Calculate the cosine similarity between the word vector representation of the abnormal question and answer information and each word vector representation in the first event workflow set;
[0128] SC233. Calculate the average value of the cosine similarity of each event workflow respectively, and record the obtained result as the matching degree of the event workflow.
[0129] The matching degree of the event workflow is expressed as:
[0130]
[0131]
[0132] in, is the word vector representation of abnormal question and answer information, is the word vector representation of the i-th event identifier in the event workflow, and n is the total number of event identifiers in the event workflow.
[0133] SC24. Determine the event workflow of abnormal question and answer information based on event identification, word vector representation and matching degree.
[0134] If there is an event workflow in the first event workflow set whose matching degree is not less than a preset matching threshold, adding the event identifier and word vector representation to the event workflow with the highest matching degree;
[0135] If the matching degrees of all event workflows in the first event workflow set are lower than a preset matching threshold, a new event workflow is created according to the event identifier and the word vector representation.
[0136] Based on the above content, if the abnormal question and answer information has a high degree of match with the current existing event workflow, similar abnormal question and answer information is classified into the same workflow, which is conducive to collective processing and management, and improves the consistency and reliability of problem solving; if the abnormal question and answer information has a low degree of match with the current existing event workflow, creating a new workflow ensures the independent processing of new problems, dynamically adapts to new abnormal types, and has the ability to flexibly expand.
[0137] This embodiment classifies and matches event workflows, and manages different types of problems with different workflows, thereby ensuring the efficiency, consistency, and traceability of the processing process and improving the management and processing quality of abnormal events.
[0138] In one embodiment, the process of determining the event identifier and event workflow of abnormal question and answer information according to the information data source further includes the following steps:
[0139] SC25. After creating a new event workflow, build an instant messaging chat room based on the workflow configuration information of the event workflow;
[0140] The instant messaging chat room is used for real-time communication and collaboration to ensure that relevant personnel can quickly respond to and handle abnormal situations. The workflow configuration information includes personnel information, tool information and process templates. Among them, all relevant personnel involved in event processing are added to the chat room through personnel information, and the tool information required, such as automation scripts, is integrated into the chat room through tool information to ensure that relevant personnel can easily access processing tools. The process template ensures the standardization and efficiency of event processing.
[0141] SC26: When the number of event identifiers in the event workflow increases, assign alerts to relevant personnel based on the number of event identifiers and the alert threshold.
[0142] The alarm threshold is configured according to the priority of the event workflow. The priority of the event workflow is set according to the word vector of the event workflow. For example, for business data anomalies that directly affect business operations (such as order data loss, payment anomalies), set it to high priority; for situations where services are available but functions are limited (such as weak recognition of user's vague business query input intention), set it to medium priority; for situations that do not directly affect core business (such as occasional inaccurate replies to chat), set it to low priority.
[0143] Based on the priority of the above event workflow, the weight parameter corresponding to the priority can be obtained, and the alarm threshold of the event workflow can be calculated based on the weight parameter and the preset alarm baseline threshold. When the alarm threshold is reached, an alarm is assigned to the relevant personnel to ensure that the relevant personnel can clearly understand the current emergency situation to ensure stable business operation.
[0144] The execution event workflow obtains the exception handling result, including the steps of:
[0145] SE1, determine the first monitoring instruction according to the information data source and word vector of the event workflow;
[0146] Specifically, the method of determining the first monitoring instruction according to the information data source and word vector of the event workflow includes the following steps:
[0147] SE11. Determine the target data source based on the information data source of the event workflow, and mark the abnormal word vectors in the word vector representation of the event workflow;
[0148] Based on the information data source of the event workflow, determine the target data source that needs to be monitored. The target data source corresponds to the information data source one by one, and the target data source includes the question and answer cache database, business database, knowledge base, and natural language model log cache database.
[0149] After determining the target data source, further annotate specific word vectors related to anomalies in the event workflow.
[0150] SE12. Cluster the abnormal word vectors of the event workflow through a clustering algorithm to obtain the key target parameter group;
[0151] Use clustering algorithms (such as K-means or DBSCAN) to perform cluster analysis on the above abnormal word vectors; through cluster analysis, similar abnormal word vectors are grouped together, and key target parameters are matched one by one for each group of abnormal word vectors. These key target parameters represent the unique characteristics of abnormal events, and all key target parameters are integrated into a key target parameter group.
[0152] SE13. Determine a first monitoring instruction according to the target data source and the key target parameter group.
[0153] In this embodiment, the monitoring instructions are specific instructions for extracting backend monitoring data from the information data source to ensure that the acquired data can reflect the characteristics of abnormal situations.
[0154] SE2, obtaining the backend monitoring data of the information data source through the first monitoring instruction;
[0155] Using the first monitoring instruction generated in step SE1, data is extracted from the target data source to obtain backend monitoring data. The backend monitoring data can provide detailed information about the abnormal event to support further determination and resolution of the abnormal problem.
[0156] As a preferred embodiment, the first monitoring instruction is a natural language instruction; the method of obtaining the back-end monitoring data of the information data source through the first monitoring instruction comprises the steps of:
[0157] Parsing the first monitoring instruction to obtain a target data source and a key target parameter group in the first monitoring instruction;
[0158] Generate SQL monitoring instructions based on the target data source and key target parameter group.
[0159] For example, the first monitoring instruction is "Get all exception-related data in [target data source]; the exception-related data includes [key target parameter group]" input by the operation and maintenance staff based on the target data source and key target parameter group. Then this step generates an SQL monitoring instruction based on the keywords "Get" and "[key target parameter group]" in the first monitoring instruction.
[0160] Connect to the target data source through a database connection tool or API (such as JDBC, ODBC, etc.) and send SQL monitoring instructions to the target data source;
[0161] The backend monitoring data is extracted from the query results returned by the database; the backend monitoring data is formatted into an appropriate structure for subsequent analysis or display.
[0162] SE3. Determine the exception repair strategy based on the backend monitoring data, and execute the exception repair strategy through the automated operation and maintenance script to obtain the exception handling results.
[0163] Specifically, determining the abnormality repair strategy according to the backend monitoring data includes the following steps:
[0164] SE31. Analyze backend monitoring data to determine the type of anomaly and the scope of impact;
[0165] Based on the information data source, conduct in-depth analysis of the backend monitoring data to identify the anomaly type (such as connection interruption, data consistency problem, etc.) and its impact scope (such as the number of affected users, affected modules, etc.).
[0166] SE32. Determine the anomaly repair strategy and generate a repair plan;
[0167] Determine the anomaly repair strategy: For different types of anomalies, select appropriate repair strategies (such as restarting services, data rollback, model fine-tuning, etc.). These strategies can be based on a predefined strategy library, or new strategies can be generated through artificial intelligence.
[0168] Generate a repair plan: Combine the anomaly repair strategy and the scope of anomaly impact to generate a detailed repair plan, including specific execution steps, required resources, estimated time, and other information.
[0169] The execution abnormality repair strategy includes:
[0170] SE331, Fine-tuning Natural Language Model:
[0171] Data preparation: Collect and organize relevant conversation data and key parameters of abnormalities to create a training data set.
[0172] Model fine-tuning: Use the newly generated training dataset to fine-tune the existing natural language model to improve the model's ability to handle similar anomalies.
[0173] Model deployment: Update the fine-tuned model to the production environment for testing and verification.
[0174] SE332, Update cache strategy:
[0175] Cache refresh: Based on monitoring data, identify inconsistent or expired cache entries and perform refresh or deletion operations.
[0176] Policy optimization: Adjust cache policy parameters (such as TTL time, cache capacity, etc.) to ensure that the cache system can adapt to new workloads and access patterns.
[0177] SE333, Update business database:
[0178] Data repair: Use the correct data identified in the back-end monitoring data to repair or replace erroneous records in the business database.
[0179] Consistency check: Perform a consistency check to ensure that the data in the business database remains consistent and complete after repair.
[0180] SE334, Update knowledge base:
[0181] Knowledge addition and modification: Update relevant content in the knowledge base based on repair strategies and results, including adding new FAQs, correcting existing document errors, etc.
[0182] Version control: Ensure that the updated knowledge base has clear version control and change records to ensure traceability and maintainability.
[0183] Based on the foregoing, the automated operation and maintenance script is used to fine-tune the natural language model, update the cache strategy, update the business database, and update the knowledge base; specifically, based on the fine-tuning natural language model script, the deep learning framework is run to train and fine-tune the natural language model; based on the updating cache strategy script, inconsistent or expired cache entries are identified and refreshed, and cache strategy parameters are adjusted (for example, TTL time is set, cache capacity is adjusted, etc.); based on the updating business database script, error records in the business database are repaired or replaced, including data backup, correction and recovery, and data consistency verification is performed; based on the updating knowledge base script, the content in the knowledge base is updated, new FAQs are added or existing document errors are corrected, and the knowledge base version and record change information are controlled.
[0184] The updating of the event workflow according to the exception handling result comprises the following steps:
[0185] Based on the exception handling result obtained by executing the exception repair strategy through the automated operation and maintenance script, the status information in the exception handling result is extracted; the status information includes successful processing, failed processing, in-process and archived status.
[0186] Traverse the status information of all event identifiers in the event workflow, and update the event status information to archived for successfully processed event identifiers; store the archived event identifiers from the pending event storage area of the event workflow to the archived event storage area for future audit or query use, and determine the storage time and deletion rules based on the set retention period for archived data and enterprise regulations.
[0187] The event identifiers of failed processing and in the process of processing are retained in the unresolved event storage area of the event workflow. Among them, for event identifiers that have been in a state of failed processing for a long time, the word vector representation and exception processing log corresponding to the event identifier are obtained, and the event workflow is reallocated according to the event identifier, word vector representation and exception processing log, where the exception processing log is generated during the aforementioned execution of the exception repair strategy. By reallocating the event workflow, it is possible to reduce the waste of event workflow resources and optimize the system load, and specify further processing plans for event identifiers that have failed to be processed, reduce the processing time of the overall event workflow, and improve system response speed and user satisfaction.
[0188] The present invention also provides an operation and maintenance management system based on an intelligent customer service model, comprising:
[0189] An event identification management module is used to determine the event identification of abnormal question and answer information according to the information data source;
[0190] An event workflow management module is used to determine the event workflow of abnormal question and answer information according to the information data source, and to update the event workflow according to the abnormal processing result;
[0191] A data monitoring module, used to determine a first monitoring instruction according to the information data source and word vector of the event workflow, and obtain backend monitoring data of the information data source through the first monitoring instruction;
[0192] The exception repair module is used to determine the exception repair strategy based on the backend monitoring data, and to execute the exception repair strategy through the automated operation and maintenance script to obtain the exception processing results.
[0193] In the embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the modules can be electrical, mechanical or other forms.
[0194] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0195] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.
[0196] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.
[0197] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.
Claims
1. An operation and maintenance management method based on an intelligent customer service model, characterized in that: Includes steps: Generate event identification for abnormal Q&A information and assign event workflow; Execute the event workflow to obtain the exception handling results; Update the event workflow based on the exception handling results; The method of generating an event identifier for abnormal question and answer information and assigning an event workflow comprises the following steps: Collect abnormal question and answer information of the intelligent customer service model, and obtain the information data source of the abnormal question and answer information; the information data source includes question and answer cache, business database, knowledge base and natural language model; Determine the event identifier and event workflow of abnormal question and answer information according to the information data source; the event workflow includes several event identifiers and corresponding word vector representations; The execution event workflow obtains the exception handling result, including the steps of: Determine a first monitoring instruction according to the information data source and word vector of the event workflow; Acquiring back-end monitoring data of the information data source through a first monitoring instruction; Determine the exception repair strategy based on the backend monitoring data, and execute the exception repair strategy through the automated operation and maintenance script to obtain the exception handling results; The method of determining the event identifier and event workflow of abnormal question and answer information according to the information data source comprises the following steps: Determine the event identifier of abnormal question and answer information based on the information data source; Acquire a first event workflow set according to the information data source; the first event workflow set is a set of event workflows classified according to the information data source; Obtaining a word vector representation of the abnormal question and answer information, and calculating a matching degree between the word vector representation and each event workflow in the first event workflow set; Determine the event workflow of the abnormal question and answer information according to the event identifier, word vector representation and matching degree; if there is an event workflow in the first event workflow set whose matching degree is not less than a preset matching threshold, add the event identifier and word vector representation to the event workflow with the highest matching degree; if the matching degrees of all event workflows in the first event workflow set are less than the preset matching threshold, create a new event workflow according to the event identifier and word vector representation; The step of obtaining the word vector representation of abnormal question and answer information comprises the following steps: Use the WordPiece word segmentation method to split the input abnormal question and answer information into a token sequence, insert a [CLS] tag before the token sequence, and insert a [SEP] tag at the end of the token sequence to obtain the preprocessed abnormal question and answer information; Input the preprocessed abnormal question and answer information into several Transformer layers; Extract the vector representation corresponding to the [CLS] tag from the last layer of Transformer as the word vector representation of the entire abnormal question and answer information; The calculating the matching degree between the word vector representation and each event workflow in the first event workflow set comprises the steps of: Obtaining a word vector representation of each event workflow in the first event workflow set; Calculate the cosine similarity between the word vector representation of the abnormal question and answer information and each word vector representation in the first event workflow set; The average value of the cosine similarity of each event workflow is calculated respectively, and the obtained result is recorded as the matching degree of the event workflow.
2. The operation and maintenance management method based on the intelligent customer service model according to claim 1, characterized in that: Before collecting abnormal question and answer information of the intelligent customer service model, the steps include: receiving input current questions and context information, and generating standardized questions according to the current questions and context information; Search the question and answer cache according to the standardized question. If a match is successful, the question and answer cache is recorded as the information data source. Otherwise, the business database is searched according to the preset regular matching rules and vocabulary matching rules. If the match is successful, the business database is recorded as the information data source; Otherwise, keyword matching is performed according to the standardized question to obtain the knowledge base matching result. If the match is successful, the knowledge base is recorded as the information data source; Otherwise, the natural language model is recorded as the information data source; A Prompt template is generated based on the current question, context information and information data source, and the natural language model outputs answer information according to the Prompt template and associates the answer information with the information data source.
3. The operation and maintenance management method based on the intelligent customer service model according to claim 1, characterized in that: The method of determining the first monitoring instruction according to the information data source and word vector of the event workflow comprises the steps of: Determine the target data source based on the information data source of the event workflow, and mark the abnormal word vectors in the word vector representation of the event workflow; Cluster the abnormal word vectors of the event workflow through a clustering algorithm to obtain key target parameter groups; A first monitoring instruction is determined according to a target data source and a key target parameter group.
4. The operation and maintenance management method based on the intelligent customer service model according to claim 1, characterized in that: The first monitoring instruction is a natural language instruction; the method of obtaining the back-end monitoring data of the information data source through the first monitoring instruction comprises the steps of: Parsing the first monitoring instruction to obtain a target data source and a key target parameter group in the first monitoring instruction; Generate SQL monitoring instructions based on the target data source and key target parameter group; Connect to the target data source through a database connection tool or API and send SQL monitoring instructions to the target data source; Extract backend monitoring data from the query results returned by the database.
5. The operation and maintenance management method based on the intelligent customer service model according to claim 1, characterized in that: Determining the abnormality repair strategy according to the back-end monitoring data includes the following steps: Analyze backend monitoring data to determine the type of anomaly and the scope of impact; Determine anomaly repair strategy and generate repair plan; The execution exception repair strategy includes fine-tuning the natural language model, updating the cache strategy, updating the business database and updating the knowledge base.
6. The operation and maintenance management method based on the intelligent customer service model according to claim 1, characterized in that: The updating of the event workflow according to the exception handling result comprises the following steps: Based on the exception handling result obtained by executing the exception repair strategy through the automated operation and maintenance script, extract the status information in the exception handling result; the status information includes successful processing, failed processing, processing in progress and archived status; Traverse the status information of all event identifiers in the event workflow, and update the event status information to archived for successfully processed event identifiers; store the archived event identifiers from the pending event storage area of the event workflow to the archived event storage area; for event identifiers that have been in a state of long-term processing failure, obtain the word vector representation and exception processing log corresponding to the event identifier, and reallocate the event workflow based on the event identifier, word vector representation and exception processing log.
7. An operation and maintenance management system based on an intelligent customer service model, characterized in that: The operation and maintenance management method based on the intelligent customer service model as claimed in any one of claims 1 to 6 comprises: An event identification management module is used to determine the event identification of abnormal question and answer information according to the information data source; An event workflow management module is used to determine the event workflow of abnormal question and answer information according to the information data source, and to update the event workflow according to the abnormal processing result; A data monitoring module, used to determine a first monitoring instruction according to the information data source and word vector of the event workflow, and obtain backend monitoring data of the information data source through the first monitoring instruction; The exception repair module is used to determine the exception repair strategy based on the backend monitoring data, and to execute the exception repair strategy through the automated operation and maintenance script to obtain the exception processing results.
