Multi-dimensional knowledge extraction construction method and system applied to operation and maintenance technology service

Abstract

The invention provides a multi-dimensional knowledge extraction construction method and system applied to operation and maintenance technology services, and relates to the field of operation and maintenance technology services, and the method comprises the steps: firstly obtaining an operation and maintenance service full record generated in an operation and maintenance technology service process, and then carrying out the association network construction processing of the operation and maintenance service full record, and obtaining a target association network structure. And based on the structure, identifying the core record cluster, extracting complementary operation and maintenance information, and fusing to form an operation and maintenance knowledge module set. And then performing dynamic knowledge system construction processing on the operation and maintenance knowledge module set to form a target operation and maintenance knowledge system with a self-adjustment capability. And finally, based on the target operation and maintenance knowledge system, aiming at the newly received operation and maintenance service request, generating operation and maintenance service response content containing the resource optimization suggestion, thereby effectively improving the operation and maintenance service quality and efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of operation and maintenance technical services, in particular to a multi-dimensional knowledge extraction construction method and system applied to operation and maintenance technical services. BACKGROUND

[0002] In the field of operation and maintenance technical services, with the increasing complexity and diversification of information systems, the amount of data generated in the operation and maintenance process is growing explosively. The above data covers multiple operation and maintenance scenarios, including but not limited to system troubleshooting, software update deployment, hardware device maintenance, etc. At present, operation and maintenance technical services mainly rely on manual sorting and analysis of these massive data by operation and maintenance personnel to accumulate operation and maintenance experience and form a knowledge system.

[0003] However, the existing operation and maintenance knowledge construction method has many drawbacks. On the one hand, manual data sorting is inefficient, and operation and maintenance personnel need to spend a lot of time and effort to filter and extract useful information from massive operation and maintenance service records, and key content is easily missed. On the other hand, the operation and maintenance knowledge system constructed by traditional methods is often static, lacking flexibility and adaptability. In actual operation and maintenance scenarios, the service execution process under different operation and maintenance scenarios is different, and as the system is constantly updated and changed, operation and maintenance requirements are also dynamically adjusted. The static knowledge system is difficult to update and adjust in a timely manner according to these changes, resulting in the inability to provide accurate and effective resource optimization suggestions and operation and maintenance service responses when facing new operation and maintenance service requests, thereby affecting the quality and efficiency of operation and maintenance services. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a multi-dimensional knowledge extraction construction method applied to operation and maintenance technical services, which comprises: obtaining operation and maintenance service full records generated in the operation and maintenance technical service process, the operation and maintenance service full records containing service request texts, execution operation sequences, resource call records and service effect feedback texts under multiple operation and maintenance scenarios, and each operation and maintenance scenario corresponding to at least two groups of service execution process records with differences; performing associated network construction processing on the operation and maintenance service full records to establish associated edges between each operation and maintenance service record, and obtaining a target associated network structure; based on the target associated network structure, identifying a core record cluster with association in the target associated network structure, extracting complementary operation and maintenance information from each operation and maintenance service record in the core record cluster, and fusing to form an operation and maintenance knowledge module with scene adaptability, and obtaining an operation and maintenance knowledge module set; performing dynamic knowledge system construction processing on the operation and maintenance knowledge module set to form a target operation and maintenance knowledge system with self-adjusting ability; Based on the target operation and maintenance knowledge system, an operation and maintenance service response content containing resource optimization suggestions is generated for a newly received operation and maintenance service request.

[0005] In another aspect, the embodiment of the present application also provides a multi-dimensional knowledge extraction construction system applied to operation and maintenance technical services, which comprises a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above-mentioned method.

[0006] Based on the above aspects, by comprehensively obtaining operation and maintenance service full records containing rich information in multiple operation and maintenance scenarios, performing association network construction processing on the operation and maintenance service full records, establishing the association edges between the records to form a target association network structure, the internal relationship between the records in different operation and maintenance scenarios can be presented, and the potential knowledge hidden in the massive data can be mined. Based on the target association network structure, the core record clusters are identified and the complementary operation and maintenance information is extracted, and a set of operation and maintenance knowledge modules with scene adaptability is formed by fusion, so that each knowledge module can accurately correspond to a specific operation and maintenance scenario, and the pertinence and practicality of the knowledge are improved. The operation and maintenance knowledge module set is processed by dynamic knowledge system construction, and a target operation and maintenance knowledge system with self-adjusting ability is formed, which can adjust the knowledge content in real time according to the changes of the operation and maintenance scenarios and new operation and maintenance service requests, and ensures the timeliness and accuracy of the knowledge system. Finally, based on the target operation and maintenance knowledge system, an operation and maintenance service response content containing resource optimization suggestions is generated, which effectively improves the quality and efficiency of the operation and maintenance service and reduces the operation and maintenance cost. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is the execution flow diagram of the multi-dimensional knowledge extraction construction method applied to operation and maintenance technical services provided by the embodiment of the present application.

[0008] Figure 2 is the schematic diagram of exemplary hardware and software components of the multi-dimensional knowledge extraction construction system applied to operation and maintenance technical services provided by the embodiment of the present application. DETAILED DESCRIPTION

[0009] The present application will be specifically described below in conjunction with the drawings of the specification, Figure 1 is the flow diagram of the multi-dimensional knowledge extraction construction method applied to operation and maintenance technical services provided by an embodiment of the present application, and the multi-dimensional knowledge extraction construction method applied to operation and maintenance technical services will be described in detail below.

[0010] Step S110: Obtain the operation and maintenance service full record generated in the operation and maintenance technical service process, the operation and maintenance service full record contains service request texts, execution operation sequences, resource call records and service effect feedback texts in multiple operation and maintenance scenarios, and each operation and maintenance scenario corresponds to at least two groups of service execution process records with differences.

[0011] The embodiment takes the operation and maintenance service of the smart home control of the Internet of Things as an application scenario, and the multiple operation and maintenance scenarios specifically include smart light control fault operation and maintenance, smart curtain control fault operation and maintenance, smart temperature control device fault operation and maintenance, smart security device fault operation and maintenance and multi-device linkage control fault operation and maintenance. The obtained operation and maintenance service full record covers complete service data in the above-mentioned various scenarios.

[0012] The service request text contains fault descriptions submitted by the user through the smart home APP, the voice assistant or the customer service hotline, for example, "the smart light in the living room cannot be turned on through the APP", "the smart curtain in the bedroom responds with delay", "the smart temperature controller displays a temperature inconsistent with the actual temperature" and the like. The execution operation sequence is the operation step performed by the operation and maintenance personnel for the fault, for example, for the fault that the smart light cannot be turned on, the execution operation sequence can be "check the network connection state of the device - restart the router - reset the smart light device - reconfigure the network - test the control function".

[0013] The resource call record relates to various resource information called in the operation and maintenance process, including the computing resource of the remote operation and maintenance platform, the firmware upgrade package resource of the device manufacturer, the device log query interface resource of the cloud platform and the like, for example, the instruction issuing interface of the remote operation and maintenance platform is called to send a reset instruction to the smart light device, and the latest firmware version is obtained by calling the cloud server of the device manufacturer.

[0014] The service effect feedback text is the evaluation submitted by the user after the operation and maintenance is completed, for example, "the fault has been solved, and the light control has returned to normal", "the response speed of the curtain has been improved, but it is still occasionally delayed", "the temperature controller displays normally, and the use experience is good" and the like. Each operation and maintenance scenario contains at least two groups of service execution process records with differences, for example, in the smart light control fault operation and maintenance scenario, one group of execution operation sequences contains "check the network - restart the device - reconfigure the network", and the other group contains "check the power supply - replace the device module - reconfigure the network", and the two groups of execution processes have obvious differences.

[0015] In the acquisition process, for the information containing user privacy (such as the user's home address, contact information, device use habit and the like), privacy protection processing is adopted. Specifically, the personal identification information in the service request text and the feedback text is removed through the data desensitization technology, the user device unique identification in the resource call record is encrypted and stored by using the symmetric encryption algorithm, only the authorized operation and maintenance personnel can obtain it through the decryption interface, and the privacy information is prevented from being leaked.

[0016] Step S111: Deploying operation and maintenance service record collection nodes, which are respectively connected to the operation and maintenance service request receiving system, the operation execution system, the resource scheduling system and the service evaluation feedback system.

[0017] In the Internet of Things smart home control operation and maintenance service platform, distributed collection nodes are deployed, which are respectively connected to the operation and maintenance service request receiving system, the operation execution system, the resource scheduling system and the service evaluation feedback system through TCP / IP protocol. The collection nodes have built-in data collection agent programs, which support real-time monitoring of data flow of each system and have breakpoint resume function, which can automatically supplement the recorded data during network interruption after network recovery.

[0018] The collection nodes are also configured with a data filtering module, which can filter out invalid data (such as repeatedly submitted empty request texts and operation sequences with format errors) according to preset rules to ensure the validity of the collected data. In addition, the collection nodes and each system use encrypted transmission channels, and the transmitted data is encrypted through SSL protocol to prevent data from being tampered with or stolen during transmission.

[0019] Step S112: Collecting the submission time of all service requests, the request initiation terminal identifier of all service requests and the service request text of all service requests from the operation and maintenance service request receiving system to form a request record subset.

[0020] The operation and maintenance service request receiving system receives service requests from multiple channels such as smart home APP, voice assistant and customer service hotline. The collection node extracts the submission time (accurate to seconds), request initiation terminal identifier (such as APP device ID, voice assistant device serial number, customer service hotline call number encrypted value) and service request text of each service request from the system.

[0021] For example, a request record from a smart home APP contains submission time, APP device ID and "living room smart security camera cannot record" service request text; a request record from a voice assistant contains submission time, voice assistant device serial number and "kitchen smart smoke alarm false alarm" service request text. The above information is organized according to the structure of "submission time-request initiation terminal identifier-service request text" to form a request record subset.

[0022] Step S113: Collecting the operation execution account corresponding to each service request, the execution operation sequence corresponding to each service request, the operation start time corresponding to each service request, the operation end time corresponding to each service request and the operation process log corresponding to each service request from the operation execution system to form an operation record subset.

[0023] The operation execution system provides an operation interface for operation personnel and records operation data of the whole process. The collection node collects, from the system, a unique identifier of a service request, associates a corresponding operation execution account (worker ID of the operation personnel), an execution operation sequence (stored in the form of a step list), an operation start time, an operation end time, and an operation process log (such as “2024-05-20 10:00: Start checking the network connection of the camera, display offline” and “2024-05-20 10:05: Restart the camera power supply, and the device is online”).

[0024] For example, for a service request of “the living room intelligent security camera cannot record”, the collected operation record includes the worker ID of the operation personnel, the execution operation sequence of “check the network-restart the device-check the storage setting-test the recording function”, the operation start time, the operation end time, and the detailed operation process log. The above information is associated and integrated to form an operation record subset.

[0025] Step S114: Collect, from the resource scheduling system, a resource type called in each service request execution process, a resource identifier called in each service request execution process, a resource call start time in each service request execution process, a resource call end time in each service request execution process, and a resource usage state in each service request execution process to form a resource record subset.

[0026] The resource scheduling system is responsible for managing various resources required in the operation process. The collection node collects, from the system, resource call information corresponding to each service request. The resource type includes a remote instruction issuing interface, a device log query interface, a firmware upgrade package, a cloud storage space, etc. The resource identifier is a unique identifier of each type of resource, such as the URL address of the interface, the file number of the firmware upgrade package, and the space ID of the cloud storage.

[0027] The resource call start time, end time, and usage state (such as “successful call”, “call failure”, and “resource occupation”) are also collected. For example, for the fault of the intelligent camera not being able to record, the collected resource record includes the resource type of calling the “camera log query interface”, the interface URL, the call start time, the call end time, and the usage state of “successfully obtaining the log”, the resource type of calling the “cloud storage space”, the space ID, the call time, and the usage state of “sufficient storage space”. These information constitute a resource record subset.

[0028] Step S115: Collect, from the service evaluation feedback system, an evaluation submission time after the completion of each service request, an evaluation initiation terminal identifier after the completion of each service request, a service effect feedback text after the completion of each service request, and a score information after the completion of each service request to form a feedback record subset.

[0029] The service evaluation feedback system receives the evaluation data submitted by the user after the operation and maintenance service is completed. The collection node collects the evaluation submission time, the evaluation initiation terminal identifier (consistent with the request initiation terminal identifier), the service effect feedback text, and the score information (usually a satisfaction score of 1-5) corresponding to each service request from the system.

[0030] For example, the feedback record submitted by the user after the smart light failure repair contains the evaluation submission time, the APP device ID, the feedback text “light control recovery normal, response timely”, and a score of 5; the feedback record submitted by another user for the operation and maintenance of the smart curtain contains the evaluation submission time, the voice assistant device serial number, the feedback text “curtain response still delayed”, and a score of 3. These information is integrated to form a feedback record subset.

[0031] Step S116: According to the unique identifier of the service request, the corresponding records in the request record subset, the operation record subset, the resource record subset, and the feedback record subset are associated and integrated to form each operation and maintenance service record containing complete information. All the associated and integrated operation and maintenance service records are summarized to obtain the operation and maintenance service full record.

[0032] For example, step S1161: Extract the service request unique identifier of each record in the request record subset, and use the service request unique identifier as the association key.

[0033] Each record in the request record subset contains a system-generated service request unique identifier, which is composed of a date code, a request channel code, and a random sequence. The unique identifier is extracted as the association key for associating the corresponding records of other subsets. For example, the unique identifier of a request record is “20240520-APP-001”, which is used as the key to associate the operation, resource, and feedback records.

[0034] Step S1162: In the operation record subset, all operation records corresponding to the service request unique identifier are searched. If there are multiple operation records, they are sorted according to the chronological order of the operation start time to form an operation record sequence.

[0035] In the operation record subset, all operation records corresponding to the service request unique identifier “20240520-APP-001” are searched. If the request is not solved at one time due to failure during operation and maintenance and multiple operation records are generated (such as first operation and maintenance is not successful, and second operation and maintenance supplementary operation), the above records are sorted according to the chronological order of the operation start time to form an operation record sequence, ensuring the time sequence integrity of the operation process.

[0036] Step S1163: In the resource record subset, search for all corresponding resource records according to the service request unique identifier, time-align the resource records with the operation record sequence in the order of the resource call start time of the resource records, and determine the resource call record corresponding to each operation step in the operation record sequence.

[0037] The corresponding resource record is searched for in the resource record subset based on the unique identifier of the service request. The resource records are then time-aligned with the operation record sequence in chronological order of resource call start time. For example, the resource record corresponding to the "Check device network connection status" step in the operation record sequence is "Call the network status query interface," and the resource record corresponding to the "Reset smart lighting device" step is "Call the device reset instruction interface." Timestamp matching ensures a precise correspondence between each operation step and resource call.

[0038] Step S1164: In the feedback record subset, the corresponding feedback record is searched according to the service request unique identifier. If there are multiple feedback records, the feedback record with the latest submission time is selected as the valid feedback record.

[0039] The corresponding feedback record is searched in the feedback record subset based on the unique identifier of the service request. If the user submits multiple feedbacks for the same service request (for example, if the fault is not resolved in the first feedback, the operation and maintenance personnel will submit a second feedback after re-processing), the feedback record with the latest submission time will be selected as the valid feedback record to ensure that the feedback information reflects the final service effect.

[0040] Step S1165: Integrate the fields of the request record, the sorted operation record sequence, the time-aligned resource call record, and the valid feedback record, so that each integrated record contains complete fields of the four types of information: request, operation, resource, and feedback.

[0041] The above-mentioned associated request records, operation record sequences, resource call records and valid feedback records are integrated at the field level. Each integrated record contains request fields (submission time, terminal ID, request text), operation fields (execution account, operation sequence, start and end time, process log), resource fields (resource type, ID, call time, usage status) and feedback fields (evaluation time, feedback text, score), to achieve the information integrity of a single service record.

[0042] Step S1166: Perform information integrity check on each integrated record to check whether there are any missing fields or invalid field content in each integrated record. If so, re-collect the information of the missing or invalid fields from the corresponding collection system and add it to the integrated record. The integrated record that passes the check and has complete information is determined as the operation and maintenance service record containing complete information.

[0043] The integrity of the integrated record is checked to check whether there are problems such as empty request text, missing key steps in operation sequence, unrecorded resource identification, invalid feedback text, etc. If it is found that the "resource call usage state" field of a certain integrated record is missing, the field information is re-collected from the resource scheduling system; if the "operation process log" field content is chaotic and invalid, it is re-obtained from the operation and maintenance operation execution system. The integrated record that passes the check is the operation and maintenance service record containing complete information, and all such records are aggregated to form the operation and maintenance service full record.

[0044] Step S120: performing associated network construction processing on the operation and maintenance service full record to establish associated edges between each operation and maintenance service record to obtain a target associated network structure.

[0045] For the full record of the Internet of Things smart home control operation and maintenance service, an associated network is constructed by analyzing the associativity of different service records in the request, operation, resource, and feedback four dimensions. Each operation and maintenance service record is taken as a node in the network, when two records have high associativity in multiple dimensions, an associated edge is established between them, and an associated strength value is given, and finally a target associated network structure containing nodes, associated edges and strength values is formed.

[0046] Step S121: performing problem feature extraction processing on the service request text of each operation and maintenance service record to extract problem type keywords in the service request text, fault phenomenon description phrases in the service request text, and service target expressions in the service request text to form a request feature vector of each operation and maintenance service record.

[0047] The natural language processing technology is used to extract the problem features of the service request text. First, the request text is split into independent words by the word segmentation algorithm, and then the key word and phrase are filtered out by the part-of-speech tagging, such as nouns and verbs, and then the problem type keywords, fault phenomenon description phrases and service target expressions are extracted by combining the domain dictionary (containing "smart lighting", "network connection", "unable to turn on", "response delay" and other Internet of Things smart home operation and maintenance domain words).

[0048] For example, for the service request text "the smart lighting in the living room cannot be turned on through the APP, and the remote control function needs to be restored", the extracted problem type keywords are "smart lighting", "APP control", and "remote control", the fault phenomenon description phrase is "unable to turn on through the APP", and the service target expression is "restore the remote control function". The above features are vectorized, each feature is mapped to a fixed-dimensional numerical vector by word embedding technology, and the above vectors are spliced to form the request feature vector of the service record. The request feature vector is a multi-dimensional numerical vector, and each dimension corresponds to an embedding value of a feature.

[0049] Step S122: Calculate the cosine similarity between the request feature vectors of any two operation and maintenance service records, and determine the cosine similarity as the problem similarity of the service request texts of the two operation and maintenance service records.

[0050] For any two operation and maintenance service records (denoted as record A and record B), the request feature vectors of the two records are obtained. The process of calculating the cosine similarity is as follows: first, calculate the dot product of the two vectors, that is, multiply the corresponding dimension values and then sum them up; then, calculate the module length of the two vectors, that is, the square root of the sum of squares of the dimension values; finally, divide the dot product result by the product of the two module lengths to obtain the cosine similarity value. The range of the cosine similarity value is between 0 and 1, and the closer to 1, the higher the problem similarity of the service request texts of the two records.

[0051] For example, the cosine similarity calculated from the request feature vector of record A and the request feature vector of record B is 0.85, indicating that the service requests of the two records are highly similar in problem type, fault phenomenon, and service target, and may both be remote control failures of intelligent lighting. The cosine similarity is directly determined as the problem similarity of the service request texts of the two records.

[0052] Step S123: Perform step-by-step processing on the execution operation sequence of each operation and maintenance service record, split the execution operation sequence into multiple continuous operation step units, count the number of steps that are completely consistent in the operation step units of any two operation and maintenance service records, and determine the ratio of the number of steps to the total number of steps in the two operation and maintenance service records as the step coincidence degree of the execution operation sequence of the two operation and maintenance service records.

[0053] The execution operation sequence of each operation and maintenance service record is split into independent operation step units according to the logical order of the operations. For example, the execution operation sequence "check device network connection status - restart router - reset intelligent lighting device - re-networking - test control function" is split into five step units: "check device network connection status", "restart router", "reset intelligent lighting device", "re-networking", and "test control function".

[0054] For any two records, count the number of steps that are completely consistent in the step units of the two records. For example, the step units of record C are [step 1, step 2, step 3, step 4], and the step units of record D are [step 1, step 2, step 5, step 6], and the number of completely consistent steps is 2. The total number of steps of the two records is the sum of the number of steps of record C and the number of steps of record D, i.e., 4+4=8. The step coincidence degree is the ratio of 2 to 8, i.e., 0.25. If the number of steps of the two records is different, the step coincidence degree is still calculated according to the above method to ensure that the step coincidence degree can reflect the similarity of the operation sequence.

[0055] Step S124: Type classification processing is performed on the resource call records of each operation and maintenance service record, and the resource call types are divided into hardware resource call, software resource call and network resource call. The number of resource types commonly called by any two operation and maintenance service records is counted, and the ratio of the number of common resource types to the total number of resource types in the two operation and maintenance service records is determined as the type correlation degree of the resource call records of the two operation and maintenance service records.

[0056] The resource call records are classified by type. The hardware resource call includes calling device debugging tools, hardware detection devices, etc. The software resource call includes calling firmware upgrade packages, operation and maintenance management software, cloud platform interfaces, etc. The network resource call includes calling remote network diagnosis services, bandwidth resources, etc. For example, a record calls “device log query interface” (software resource) and “remote network diagnosis service” (network resource), and the resource call type is software resource call and network resource call.

[0057] For any two records, the number of commonly called resource types is counted. For example, the resource call type of record E is [software resource, network resource], the resource call type of record F is [hardware resource, software resource], and the number of commonly called resource types is 1. The total number of resource types of the two records is 2+2=4, and the type correlation degree is the ratio of 1 to 4, i.e. 0.25. If a record calls three types of resources and another record calls two types of resources, and there are two common types, then the number of common types is 2, the total number is 3+2=5, and the type correlation degree is 2 / 5.

[0058] Step S125: Sentiment analysis processing is performed on the service effect feedback text of each operation and maintenance service record to determine the evaluation tendency of the service effect feedback text as positive evaluation, neutral evaluation or negative evaluation, and each evaluation tendency is mapped to a preset continuous numerical score. For any two operation and maintenance service records, the similarity of the evaluation tendency numerical scores is calculated, and the similarity is determined as the evaluation convergence of the service effect feedback text of the two operation and maintenance service records.

[0059] The service effect feedback text is processed using a sentiment analysis model. The sentiment analysis model identifies sentiment words and semantic tendencies in the text through training to determine the evaluation tendency. Positive evaluation such as “fault has been solved, use normally”, neutral evaluation such as “fault has been partially solved, still needs to be observed”, and negative evaluation such as “fault has not been improved after operation and maintenance”.

[0060] The three evaluation tendencies are respectively mapped to preset continuous numerical scores, for example, positive evaluation is mapped to a higher numerical range, neutral evaluation is mapped to a middle numerical range, and negative evaluation is mapped to a lower numerical range. When calculating the evaluation convergence of two records, first, the numerical scores of the two records are obtained, then the absolute value of the score difference of the two records is calculated, and the ratio of the absolute value to the score range is subtracted by 1 to obtain the evaluation convergence value. The evaluation convergence value is also between 0 and 1, and the closer the value is to 1, the more consistent the service effect feedback evaluation tendencies of the two records are.

[0061] For example, the evaluation tendency of record G is positive evaluation, and the mapped numerical score is a higher value; the evaluation tendency of record H is also positive evaluation, and the mapped numerical score is a similar higher value, the absolute value of the score difference of the two records is small, and the calculated evaluation convergence value is close to 1, indicating that the evaluation tendencies of the two records are highly consistent. If the evaluation tendency of record I is negative evaluation (mapped to a lower numerical score), and the evaluation tendency of record J is positive evaluation (mapped to a higher numerical score), the absolute value of the score difference of the two records is large, and the evaluation convergence value is close to 0, indicating that the evaluation tendencies are significantly different.

[0062] Step S126: When the problem similarity, step coincidence degree, type association degree and evaluation convergence of any two operation and maintenance service records all exceed the corresponding preset threshold, an association edge is established between the two operation and maintenance service records, and the standardized weighted average value of the problem similarity, step coincidence degree, type association degree and evaluation convergence is determined as the association strength value of the association edge.

[0063] The problem similarity, step coincidence degree, type association degree and evaluation convergence are respectively set with preset thresholds, which are determined based on historical data statistics and business requirements of the Internet of Things smart home operation and maintenance service. For example, the problem similarity threshold is set to a higher value, the step coincidence degree threshold is set to a medium value, the type association degree threshold is set to a basic value, and the evaluation convergence threshold is set to a medium value.

[0064] For any two operation and maintenance service records, if the problem similarity exceeds the corresponding threshold, and the step coincidence degree, type association degree and evaluation convergence all exceed the corresponding preset thresholds, it is determined that the two records have strong association, and an association edge is established between the two records. After establishing the association edge, the association strength value needs to be calculated. First, the four indexes are standardized, and the index values are uniformly converted to the standard interval of 0 to 1 (if the index value is already in the interval, no processing is needed).

[0065] Subsequently, different weights are assigned to the four normalized indexes, and the weights are set according to the influence degree of each index on the association of the operation and maintenance service, for example, the problem similarity weight is the highest, the step coincidence degree and the type association degree are the second, and the evaluation convergence weight is relatively low. The product of the four indexes and the respective weights is calculated, and the sum of the four products is added to obtain the association strength value of the association edge, which reflects the closeness of the association between the two records.

[0066] Step S127: All established association edges and corresponding operation and maintenance service records are combined to form a target association network structure containing node attributes, edge attributes and association strength values.

[0067] Each operation and maintenance service record is taken as a node in the target association network structure, and the node attributes include the service request text, the execution operation sequence, the resource call record, the service effect feedback text and the type of the operation and maintenance scene of the record. All the association edges established in step S126 are included in the network, and the edge attributes include the identification of the two node records corresponding to the association edge, the original values and the standardized values of the four indexes for establishing the association, and other information.

[0068] At the same time, the association strength value of each association edge is bound to the corresponding association edge to form a complete target association network structure. The target association network structure can be stored and visually displayed through a graph database, each node in the graph database corresponds to an operation and maintenance service record, and each edge corresponds to the association relationship and the association strength between the records, which facilitates the subsequent analysis of the association of the nodes in the network and the identification of the core record cluster.

[0069] Step S130: Based on the target association network structure, a core record cluster with association in the target association network structure is identified, complementary operation and maintenance information is extracted from each operation and maintenance service record in the core record cluster, and a operation and maintenance knowledge module with scene adaptability is formed by fusion to obtain a set of operation and maintenance knowledge modules.

[0070] Based on the target association network structure, the core record cluster with close association is identified by analyzing the association strength and the network topology between the nodes. For each core record cluster, the complementary information of different records is mined, integrated and fused to form a knowledge module adapted to a specific operation and maintenance scene, and finally all the knowledge modules are summarized to form a set.

[0071] Step S131: The number of association edges of each node in the target association network structure is calculated, and the nodes with the number of association edges exceeding a preset edge number threshold are marked as core nodes.

[0072] Each node in the target association network structure is traversed to count the number of association edges connected to each node, which reflects the association frequency of the operation and maintenance service record corresponding to the node and other records. The preset edge number threshold is set according to the overall distribution of the number of association edges of the nodes in the network, for example, the sum of the average value and the standard deviation of the number of association edges of all nodes is taken as the threshold.

[0073] Nodes with the number of association edges exceeding the preset threshold are marked as core nodes. The operation and maintenance service records corresponding to these nodes usually involve high-frequency or typical operation and maintenance scenarios, have close association with other records, and are the center nodes constituting the core record cluster. For example, a node corresponding to the operation record of "smart lighting network connection fault" has a large number of association edges, exceeds the preset threshold, and is marked as a core node.

[0074] Step S132: Taking each core node as the center, all nodes connected by the association edges of the core node are traversed to form an initial node set centered on the core node, and the ratio of the total number of association edges between all nodes in the initial node set to the theoretical maximum number of association edges of all nodes in the initial node set is calculated. The ratio is determined as the association density of the initial node set.

[0075] Taking each core node as the starting point, the nodes connected by all association edges of the core node are obtained by depth-first traversal or breadth-first traversal, and the nodes and the core node together constitute an initial node set. For example, the core node A connects nodes B, C, and D, and the initial node set is {A, B, C, D}.

[0076] The total number of association edges in the initial node set is calculated, that is, the number of association edges established between any two nodes in the set. At the same time, the theoretical maximum number of association edges of the set is calculated, and the calculation formula is the number of nodes multiplied by (the number of nodes-1) and then divided by 2. The formula is based on the theoretical situation that an association edge can be established between any two nodes. The association density is the ratio of the actual total number of association edges to the theoretical maximum number of association edges, which reflects the close association degree of the nodes in the initial node set.

[0077] Step S133: The initial node set with an association density exceeding a preset density threshold is determined as a core record cluster, and each core record cluster corresponds to a type of operation and maintenance scenario.

[0078] The preset density threshold is set according to the scenario characteristics of the operation and maintenance service and the overall association level of the association network, for example, it is set to a moderately high value to ensure that the nodes in the core record cluster have strong internal association. The initial node set with an association density exceeding the threshold is determined as a core record cluster, and the initial node set with an association density not reaching the threshold is split or merged into other clusters.

[0079] The operation and maintenance service records corresponding to the nodes in each core record cluster belong to the same operation and maintenance scene, for example, the core record cluster composed of the core node "intelligent light network connection fault" and its associated nodes corresponds to the "intelligent light control fault operation and maintenance" scene; the cluster composed of the core node "intelligent temperature controller temperature display abnormality" and its associated nodes corresponds to the "intelligent temperature control device fault operation and maintenance" scene.

[0080] Step S134: performing information complementarity analysis processing on each operation and maintenance service record in each core record cluster, identifying the differential fault description of the service request text in different operation and maintenance service records, the complementary operation steps of the execution operation sequence in different operation and maintenance service records, and the alternative resource type of the resource call record in different operation and maintenance service records.

[0081] Step S1341: clustering the problem feature vectors of the service request texts of all operation and maintenance service records in the core record cluster to obtain multiple problem feature clustering clusters, each problem feature clustering cluster corresponding to a type of core fault problem.

[0082] The request feature vectors of all records in the core record cluster are clustered by using a clustering algorithm. The clustering algorithm calculates the distance between vectors, and vectors with close distances are classified into the same cluster. For example, the request feature vectors of "living room intelligent light cannot be opened by APP", "bedroom intelligent light APP control failure", and "study room intelligent light remote control non-response" are clustered into the same cluster, corresponding to the "intelligent light APP remote control fault" core fault problem; the vectors of "intelligent light local key non-response" and "intelligent light not bright after power on" are clustered into another cluster, corresponding to the "intelligent light local control fault" core fault problem.

[0083] The records in each problem feature clustering cluster are around the same core fault problem, and different clustering clusters correspond to different core fault problems.

[0084] Step S1342: performing fault description disassembly on the service request text of each operation and maintenance service record in each problem feature clustering cluster, extracting the fault occurrence environment in each service request text, the fault manifestation in each service request text, and the fault impact range in each service request text, identifying the differential description content of the same core fault problem in different service request texts, and forming a differential fault description list.

[0085] For each problem feature clustering cluster, the service request text of each record in the cluster is disassembled. For example, in the "smart lighting APP remote control failure" clustering cluster, the request text of a record "living room smart lighting cannot be turned on through the APP when the WiFi signal is weak" is disassembled, and the fault occurrence environment is "weak WiFi signal", the fault manifestation is "cannot be turned on through the APP", and the fault impact range is "living room smart lighting"; the request text of another record "bedroom smart lighting cannot be remotely controlled after APP version update" is disassembled, and the fault occurrence environment is "after APP version update", the fault manifestation is "cannot be remotely controlled", and the fault impact range is "bedroom smart lighting".

[0086] By comparing the disassembled results of different records in the cluster, different description contents are identified, such as fault occurrence environments "weak WiFi signal", "after APP version update", and "after router restart", and fault impact ranges involving smart lighting in different rooms. The above different contents are sorted to form a differential fault description list, which comprehensively covers different scenes of the same core fault problem.

[0087] Step S1343: Perform sequence alignment processing on the operation step units of the execution operation sequences of all operation and maintenance service records in the core record cluster, and construct a standard operation step framework based on the logical relationship of the operation steps.

[0088] The step units of the execution operation sequences of all records in the core record cluster are collected, and a sequence alignment algorithm is used to process the above step units. The sequence alignment algorithm aligns similar steps in different sequences to the same position according to the semantic similarity and logical relationship of the operation steps. For example, for the execution operation sequence of "smart lighting APP control failure", "check WiFi connection" and "check network status" are aligned to the same position, and "restart router" and "reset network" are aligned to the same position.

[0089] Based on the aligned step units, a standard operation step framework is constructed according to the logical order of "troubleshooting-fault location-fault repair-function test". For example, the standard operation step framework is "check device network connection-check APP version and permission-restart network device-reset smart lighting device-reconfigure network-test APP control function", which covers the general step flow of handling such faults.

[0090] Step S1344: Identify the additional operation steps in the execution operation sequences of each operation and maintenance service record that are not included in the standard operation step framework, and determine whether the additional operation steps can solve the special cases not covered in the standard operation step framework. The additional operation steps that can solve the special cases are determined as complementary operation steps.

[0091] Traverse the execution operation sequence of each record in the core record cluster, compare with the standard operation step framework, and identify additional operation steps not included in the framework. For example, the operation sequence of a certain record includes steps such as "uninstall and reinstall APP" and "clear APP cache data", which are not included in the standard framework.

[0092] Determine whether these additional steps are for special cases, such as "APP cache exception causing control failure" and "APP installation file damage causing function exception", which are not covered by the standard framework. If the additional steps can effectively solve these special cases, they are determined as complementary operation steps. For example, "uninstall and reinstall APP" can solve the APP file damage problem and is determined as a complementary operation step; "clear APP cache data" can solve the cache exception problem and is also determined as a complementary operation step.

[0093] Step S1345: Classify and count the resource types of the resource call records of all operation and maintenance service records in the core record cluster, obtain the call frequency of each resource type, and determine the resource types with call frequency exceeding the preset frequency threshold as core resource types.

[0094] Statistically analyze the resource call types of all records in the core record cluster, calculate the call frequency of each resource type, i.e., the number of times the resource of this type is called by different records. For example, "device log query interface" is called multiple times, "remote instruction issuing interface" is called multiple times, "firmware upgrade package" is called multiple times, and "hardware detection tool" is called less frequently.

[0095] The preset frequency threshold is set according to the overall distribution of resource calls, and the resource types with call frequency exceeding the threshold are determined as core resource types. For example, the call frequencies of "device log query interface", "remote instruction issuing interface", and "firmware upgrade package" exceed the threshold, and they are determined as core resource types. These resources are key resources for handling the corresponding operation and maintenance scene faults of the cluster.

[0096] Step S1346: Identify other resource types called in the operation and maintenance service records calling the core resource types, determine whether the other resource types can complete the same function when the core resource types are unavailable, and determine the other resource types that can complete the same function as alternative resource types.

[0097] Find all records calling the core resource types and extract other non-core resource types called in these records. For example, some records calling "device log query interface" (core resource) also call "local device log reading tool"; some records calling "remote instruction issuing interface" (core resource) also call "device local instruction input tool".

[0098] Determine whether these other resource types can replace the core resources to complete the same function when the core resources are unavailable. For example, when the "device log query interface" cannot be called due to network failure, the "local device log reading tool" can read the device log through local connection to complete the same log query function, and is therefore determined as a replacement resource type; when the "remote instruction issuing interface" is unavailable, the "device local instruction input tool" can send instructions to the device through local operation, and is determined as a replacement resource type.

[0099] Step S1347: Organize the differentiated fault description list, complementary operation steps, and replacement resource types to form the information complementarity analysis result.

[0100] Integrate the differentiated fault description list generated in step S1342, the complementary operation steps determined in step S1344, and the replacement resource types identified in step S1346, and organize them according to the structure of "core fault problem-differentiated description-standard operation framework-complementary steps-core resources-replacement resources" to form the information complementarity analysis result.

[0101] Step S135: Extract the differentiated fault description from each operation and maintenance service record to form a scene fault feature set, extract the complementary operation steps from each operation and maintenance service record to form a complete operation sequence, and extract the replacement resource types from each operation and maintenance service record to form a resource alternative list.

[0102] Extract all the contents of the differentiated fault description list from the information complementarity analysis result, and classify them according to the core fault problem to form a scene fault feature set. For example, the fault feature set of the "smart lighting APP control failure" scene includes "weak WiFi signal leading to control failure", "APP version problem leading to control failure", and "network device failure leading to control failure".

[0103] Extract the complementary operation steps and supplement them to the standard operation step framework to form a complete operation sequence. For example, based on the standard framework "check network-restart device-reconfigure network-test function", supplement the complementary steps "uninstall and reinstall APP" and "clear cache" to form a complete operation process that covers general and special cases.

[0104] Extract the replacement resource types and classify them according to the corresponding core resource types to form a resource alternative list. For example, the alternative resource for "device log query interface" is "local device log reading tool", and the alternative resource for "remote instruction issuing interface" is "local instruction input tool", which clearly lists the replacement solutions for each core resource.

[0105] Step S136: The scene fault feature set, complete operation flow sequence and resource alternative list are associated and integrated, the resource calling rule corresponding to each step in the complete operation flow sequence and the fault handling plan corresponding to each step in the complete operation flow sequence are supplemented, and an operation and maintenance knowledge module containing scene features, operation flow, resource configuration and fault plan is formed.

[0106] The scene fault feature set, complete operation flow sequence and resource alternative list are associated and integrated, the resource calling rule corresponding to each step in the complete operation flow sequence and the fault handling plan corresponding to each step in the complete operation flow sequence are supplemented, and an operation and maintenance knowledge module containing scene features, operation flow, resource configuration and fault plan is formed.

[0107] The fault handling plan provides a processing scheme for possible abnormal situations of each step, for example, in the "restart router" step, if the network still does not recover after restarting, the plan is "check router hardware state - replace router - reconfigure network"; in the "reconfigure network" step, if the network fails, the plan is "check device network configuration mode - reset device network configuration state - reinitiate network configuration".

[0108] The integrated content forms a complete operation and maintenance knowledge module, which contains four core parts: scene fault feature set (clear fault type in the scene), complete operation flow sequence (clear processing flow), resource configuration (clear resource calling rule and alternative resource) and fault plan (clear abnormal handling scheme), and has scene adaptability.

[0109] Step S137: Repeat the above processing for each core record cluster to obtain multiple operation and maintenance knowledge modules, and aggregate the operation and maintenance knowledge modules to form an operation and maintenance knowledge module set.

[0110] For each core record cluster identified in the target associated network structure, repeat the processing procedures of steps S134 to S136, that is, sequentially perform information complementarity analysis, extract a scene fault feature set, a complete operation flow sequence and a resource alternative list, associate and integrate and supplement resource calling rules and fault handling plans, and form an operation and maintenance knowledge module corresponding to each cluster.

[0111] For example, for the "intelligent curtain control fault operation and maintenance" core record cluster, a knowledge module containing curtain control fault features, operation flow, resource configuration and plan is formed; for the "intelligent security device fault operation and maintenance" cluster, a corresponding security device operation and maintenance knowledge module is formed. All these knowledge modules are aggregated and stored according to the operation and maintenance scene type to form an operation and maintenance knowledge module set.

[0112] Step S140: Dynamic knowledge system construction processing is performed on the set of operation and maintenance knowledge modules to form a target operation and maintenance knowledge system with self-adjusting capability.

[0113] Based on the set of operation and maintenance knowledge modules, an initial multi-dimensional knowledge system is constructed. By recording the actual application data and user feedback of the knowledge modules, the classification framework, module classification attribution and module content of the knowledge system are regularly adjusted to enable the knowledge system to have self-adjusting capability, and finally a target operation and maintenance knowledge system is formed.

[0114] Step S141: The classification framework of the operation and maintenance knowledge system is initially set, and the classification framework of the operation and maintenance knowledge system includes three levels of scene categories, scene sub-categories and scene fine categories. Each level is divided into multiple classification tags according to the business attributes of the operation and maintenance service.

[0115] The initially constructed classification framework includes three levels. The scene categories are divided according to the device types of the Internet of Things smart home control, including five large category tags of "smart lighting device operation and maintenance", "smart curtain device operation and maintenance", "smart temperature control device operation and maintenance", "smart security device operation and maintenance" and "multi-device linkage operation and maintenance".

[0116] The scene sub-categories are divided under the scene categories according to the fault types, for example, the "smart lighting device operation and maintenance" category includes four sub-category tags of "control function failure", "power failure", "network connection failure" and "firmware failure"; the "smart temperature control device operation and maintenance" category includes three sub-category tags of "temperature display failure", "adjustment function failure" and "network failure".

[0117] The scene fine categories are divided under the scene sub-categories according to the specific fault manifestations, for example, the "smart lighting device operation and maintenance-control function failure" sub-category includes three fine category tags of "APP remote control failure", "local key control failure" and "voice control failure"; the "smart lighting device operation and maintenance-network connection failure" sub-category includes three fine category tags of "WiFi connection failure", "unstable network signal" and "IP address conflict". Each level of the classification framework has a clear definition of business attributes to ensure that the subsequent knowledge modules can be accurately classified.

[0118] Step S142: Each operation and maintenance knowledge module in the set of operation and maintenance knowledge modules is assigned to the corresponding scene fine category under the classification framework of the operation and maintenance knowledge system according to the matching degree of the scene fault feature set of the operation and maintenance knowledge module and each level classification tag in the classification framework of the operation and maintenance knowledge system, to form an initial multi-dimensional knowledge system.

[0119] For each module in the set of operation and maintenance knowledge modules, extract the key information such as core fault type and specific fault performance in the scene fault feature set of the module. Calculate the matching degree of these key information and the scene category, scene subcategory and scene subclass labels in the classification framework. The matching degree is determined based on the coincidence degree and semantic similarity of feature keywords.

[0120] For example, the scene fault feature set of a certain operation and maintenance knowledge module is "smart lighting APP remote control failure, frequent failure when WiFi signal is weak", the core fault type is "control function failure", and the specific fault performance is "APP remote control failure". The calculation shows that the module has the highest matching degree with "smart lighting device operation and maintenance" (scene category), "control function failure" (scene subcategory) and "APP remote control failure" (scene subclass), and is therefore assigned to the "smart lighting device operation and maintenance-control function failure-APP remote control failure" subclass.

[0121] All operation and maintenance knowledge modules are assigned to the corresponding scene subclass according to the above method, and each scene subclass contains at least one operation and maintenance knowledge module. Thus, an initial multidimensional knowledge system is formed, which clearly organizes the operation and maintenance knowledge of different scenes in a hierarchical structure.

[0122] Step S143: Record the usage scenario frequency of each operation and maintenance knowledge module in actual application and the cross-scenario reuse rate of each operation and maintenance knowledge module. The usage scenario frequency is the number of operation and maintenance scenarios corresponding to the operation and maintenance knowledge module when it is called. The cross-scenario reuse rate is the ratio of the number of times the operation and maintenance knowledge module is called in non-initially assigned scenarios to the total number of calls of the operation and maintenance knowledge module.

[0123] After the initial multidimensional knowledge system is put into actual use, a usage log recording function is set for each operation and maintenance knowledge module. The usage scenario frequency is determined by counting the number of different operation and maintenance scenarios corresponding to the module when it is called. For example, a module is initially assigned to the "smart lighting APP remote control failure" scenario, and in actual application, it is called not only in this scenario but also in the "smart curtain APP remote control failure" and "smart temperature control APP remote control failure" scenarios, so its usage scenario frequency is 3.

[0124] The calculation of the cross-scenario reuse rate requires first counting the total number of calls of the module, and then counting the number of calls in non-initially assigned scenarios. The ratio of the two is the cross-scenario reuse rate. For example, the total number of calls of a module is several times, and the number of calls in non-initially assigned scenarios is several times. The cross-scenario reuse rate is the ratio of the two. If the module is only called in the initially assigned scenario, the cross-scenario reuse rate is 0.

[0125] Step S144: Collect the adaptation feedback of each operation and maintenance knowledge module after being applied in the new scene, which includes the applicability score of the operation and maintenance knowledge module operation process, the matching score of the operation and maintenance knowledge module resource configuration, and the effectiveness score of the operation and maintenance knowledge module fault plan.

[0126] When the operation and maintenance knowledge module is applied in the new operation and maintenance scene, the adaptation feedback questionnaire is pushed to the operation and maintenance personnel and users through the operation and maintenance service platform. The adaptation feedback questionnaire includes three core dimension score items: the applicability score of the operation process, which is used to evaluate whether the operation steps in the module are suitable for the fault handling of the new scene; the matching score of the resource configuration, which is used to evaluate whether the resource calling scheme in the module matches the resource demand of the new scene; and the effectiveness score of the fault plan, which is used to evaluate whether the abnormal handling plan in the module is effective in the new scene.

[0127] The score adopts a fixed scoring standard, and the operation and maintenance personnel and users complete the score according to the actual application experience. After the system collects all the feedback scores, the average score of each dimension is calculated as the adaptation feedback result of the module in the new scene. The adaptation feedback of each module is associated with the corresponding application scene information.

[0128] Step S145: When the cross-scene reuse rate of any operation and maintenance knowledge module exceeds the preset reuse threshold, the common characteristics of the multiple scenes to which the operation and maintenance knowledge module is adapted are analyzed, a scene subclass or a scene subclass corresponding to the common characteristics is added in the classification framework of the operation and maintenance knowledge system, and the operation and maintenance knowledge module is simultaneously assigned to the added classification and the original classification of the operation and maintenance knowledge module.

[0129] The preset reuse threshold is set according to the application demand of the knowledge system and the general situation of module reuse. When the cross-scene reuse rate of a certain operation and maintenance knowledge module exceeds the threshold, it indicates that the module has strong cross-scene applicability, and the common characteristics of all the scenes to which the module is adapted need to be analyzed.

[0130] For example, a module initially assigned to "smart lighting APP remote control fault" has a cross-scene reuse rate exceeding the threshold, and the scenes to which the module is adapted include "smart lighting APP remote control fault", "smart curtain APP remote control fault", and "smart temperature control APP remote control fault". The common characteristics of these scenes are "smart device APP remote control fault", and they all belong to different device type categories.

[0131] Based on the common characteristics, a scene subclass "APP remote control fault" is added in the knowledge system classification framework. The subclass does not belong to a specific scene category, but exists as a shared subclass across categories. The module is simultaneously assigned to the added "APP remote control fault" subclass and the original "smart lighting device operation and maintenance-control function fault-APP remote control fault" subclass, realizing the multi-classification attribution of the module and improving the knowledge reuse efficiency.

[0132] Step S146: When the adaptation feedback score of any one operation and maintenance knowledge module is lower than the preset score threshold, the reason for being lower than the preset score threshold is analyzed. If the operation process of the operation and maintenance knowledge module is missing, the operation steps of the operation and maintenance knowledge module in the new scene are supplemented. If the resource configuration of the operation and maintenance knowledge module is not matched, the resource candidate list of the operation and maintenance knowledge module is updated to form an adjusted operation and maintenance knowledge module.

[0133] The preset score threshold is used to judge the adaptation effect of the operation and maintenance knowledge module in the new scene. When the adaptation feedback score of the module is lower than the threshold, the module adjustment process is started. First, the specific reason for the low score is analyzed by checking the operation records of the operation and maintenance personnel, the specific description of the user feedback and the application log of the module to locate the root cause of the problem.

[0134] If the reason is that the operation process is missing, that is, the operation steps in the module cannot cover the fault handling requirements of the new scene, for example, when a certain module is applied in the "smart curtain APP remote control fault" scene, it is found that the operation step of "checking the curtain motor drive state" is missing, which leads to incomplete fault diagnosis. The step is supplemented in the complete operation flow sequence of the module, and the execution conditions and operation specifications of the step are specified.

[0135] If the reason is that the resource configuration is not matched, that is, the resource calling scheme in the module cannot meet the resource requirements of the new scene, for example, when a certain module is applied in the "smart temperature control APP remote control fault" scene, the "remote log query interface" recommended by the module cannot adapt to the log format of the brand temperature controller. The resource candidate list of the module is updated, the "dedicated log query interface" suitable for the brand temperature controller is added, and the resource calling rules are adjusted to preferentially call the dedicated interface. After adjustment, a new operation and maintenance knowledge module is formed to replace the original module for use.

[0136] Step S147: According to the usage scene frequency of all operation and maintenance knowledge modules, the cross-scene reuse rate of all operation and maintenance knowledge modules and the adaptation feedback of all operation and maintenance knowledge modules, the hierarchical structure of the classification framework of the operation and maintenance knowledge system, the classification labels of the classification framework of the operation and maintenance knowledge system and the classification attribution of all operation and maintenance knowledge modules are adjusted regularly to form a target operation and maintenance knowledge system with self-adjusting capability.

[0137] Step S1471: Set an adjustment period. In each adjustment period, the number of operation and maintenance knowledge modules under each scene subcategory in the classification framework of the operation and maintenance knowledge system is counted. If the number of operation and maintenance knowledge modules under any one scene subcategory exceeds the preset module number threshold, the scene subcategory is split into multiple new scene subcategories, and each new scene subcategory corresponds to different sub-scene characteristics of operation and maintenance knowledge modules.

[0138] A fixed adjustment period is set, for example, knowledge system adjustment is performed once a month. In each adjustment period, the number of operation and maintenance knowledge modules under each subcategory in the classification framework is counted. A preset module number threshold is set according to the business coverage of the subcategory and the knowledge management requirement. When the number of modules under a subcategory exceeds the threshold, it indicates that the business scope of the subcategory is too wide and needs to be split.

[0139] For example, the number of modules under the subcategory “smart security device operation and maintenance-camera failure-recording function failure” exceeds the threshold, and the sub-scene features corresponding to these modules include “insufficient storage space causes recording failure”, “recording format incompatibility causes storage failure”, and “camera hardware failure causes recording interruption”. The subcategory is split into three new subcategories: “insufficient storage space causes recording failure”, “recording format incompatibility causes storage failure”, and “camera hardware failure causes recording interruption”, and corresponding operation and maintenance knowledge modules are assigned to each new subcategory.

[0140] Step S1472: If the number of operation and maintenance knowledge modules under any one subcategory is less than the preset module number threshold, and the operation and maintenance knowledge module scene feature similarity between the subcategory and the adjacent subcategory exceeds the preset similarity threshold, the subcategory is merged into the adjacent subcategory.

[0141] For subcategories with module numbers less than the preset module number threshold, the scene feature similarity of the operation and maintenance knowledge modules under the subcategory and the adjacent subcategory (subcategories with similar business attributes) is calculated. The scene feature similarity is calculated based on the coincidence degree of the fault feature set, operation process, resource configuration, and other multi-dimensional information of the modules.

[0142] If the similarity exceeds the preset similarity threshold, it indicates that the business attributes of the two subcategories are highly similar and can be merged. For example, the number of modules under the subcategories “smart lighting device operation and maintenance-power failure-socket contact failure” and “smart lighting device operation and maintenance-power failure-power line aging” are both less than the threshold, and the scene feature similarity of the two subcategories exceeds the threshold. The two subcategories are merged into the subcategory “smart lighting device operation and maintenance-power failure-power supply line abnormality”, and the modules under the original two subcategories are uniformly assigned to the new subcategory.

[0143] Step S1473: The average usage scene frequency of operation and maintenance knowledge modules under each classification label in the classification framework of the operation and maintenance knowledge system is counted. If the average usage scene frequency of operation and maintenance knowledge modules under any one classification label is less than the preset frequency threshold, the classification label is deleted, and the operation and maintenance knowledge modules under the classification label are reassigned to other related classification labels in the classification framework of the operation and maintenance knowledge system.

[0144] Calculate the average value of the usage scenario frequency of all operation and maintenance knowledge modules under each classification label (including scene major category, subcategory, and fine category) in the classification framework. If the average usage scenario frequency under a certain classification label is lower than the preset frequency threshold, it indicates that the business scenario corresponding to the label has a low usage frequency, and it does not need to be retained separately and should be deleted.

[0145] For example, the average usage scenario frequency of the modules under the label “Smart lighting device operation and maintenance - firmware failure - old version compatibility failure” is lower than the threshold, the label is deleted, and the modules under it are reanalyzed for scene features and assigned to related labels such as “Smart lighting device operation and maintenance - firmware failure - firmware version failure”, ensuring that the modules can still be effectively retrieved and called.

[0146] Step S1474: For each operation and maintenance knowledge module, recalculate the matching degree of the scene fault feature set of the operation and maintenance knowledge module and each level classification label in the classification framework of the current operation and maintenance knowledge system, and adjust the classification attribution of the operation and maintenance knowledge module according to the new matching degree, so that the operation and maintenance knowledge module is assigned to the classification with the highest matching degree.

[0147] Since the labels of the classification framework may be added, merged, or deleted, the matching degree of each operation and maintenance knowledge module and each level label of the current classification framework needs to be recalculated. The matching degree calculation method is consistent with step S142, which is based on the keyword overlap and semantic similarity between the scene fault feature set and the label.

[0148] According to the new matching degree result, adjust the classification attribution of the module. For example, a certain module is originally assigned to the fine category “Smart window curtain device operation and maintenance - network failure - WiFi connection failure”, and after the adjustment of the classification framework, a new cross-major-category subcategory “Smart device network failure - WiFi connection failure” is added. After recalculating the matching degree, it is found that the matching degree of the module with the new subcategory is higher than that of the original fine category, so the classification attribution of the module is adjusted to the new subcategory, while the original fine category attribution is retained, realizing multi-classification management.

[0149] Step S1475: Record the classification framework structure of the operation and maintenance knowledge system after each adjustment, the classification attribution change of each operation and maintenance knowledge module after each adjustment, and the adjustment basis after each adjustment, forming an adjustment log.

[0150] All operations in the knowledge system adjustment process are recorded in detail, including the changes in the classification framework structure (added, merged, deleted labels and level adjustment), the adjustment of the classification attribution of each operation and maintenance knowledge module (original attribution, new attribution), and the specific basis for the adjustment (such as module usage scenario frequency, cross-scenario reuse rate, adaptation feedback score, etc.).

[0151] The adjustment log is stored indexed by time stamp, supporting tracing of historical adjustment records. By analyzing the adjustment log, the evolution law of the knowledge system can be summarized, and finally a target operation and maintenance knowledge system with self-adjusting ability is formed.

[0152] Step S150: Based on the target operation and maintenance knowledge system, a resource optimization suggestion is generated for a newly received operation and maintenance service request.

[0153] When a new Internet of Things smart home control operation and maintenance service request is received, a matching operation and maintenance knowledge module is retrieved from the target operation and maintenance knowledge system, and a resource allocation scheme is optimized in combination with the resource constraint conditions of the request to generate a service response content containing an operation flow, resource suggestion, and fault contingency, which is pushed to the operation and maintenance personnel and user terminal to guide the operation and maintenance service.

[0154] Step S151: Receive a new operation and maintenance service request, and perform scene feature extraction processing on the new operation and maintenance service request to extract fault description text in the new operation and maintenance service request, available resource types in the new operation and maintenance service request, and service target requirements in the new operation and maintenance service request, forming a request scene feature vector of the new operation and maintenance service request.

[0155] A new service request is received through an operation and maintenance service request receiving system, and the service request can come from a smart home APP, a voice assistant, or a customer service hotline. The request content is extracted for scene features, including fault description text such as “the smart window curtain in the bedroom cannot be controlled through the APP, and the WiFi signal is normal”; available resource types provided by the user, such as “local debugging tools and router management background can be called”; and service target requirements, such as “restore the APP control function of the window curtain within 1 hour”.

[0156] The extracted fault description text, available resource types, and service target requirements are vectorized. The same word embedding technology as in step S121 is used to map each feature to a fixed-dimensional numerical vector, and the above vectors are spliced to form a request scene feature vector that can fully reflect the scene attributes of the new request.

[0157] Step S152: Calculate the similarity between the request scene feature vector of the new operation and maintenance service request and the scene fault feature set of each operation and maintenance knowledge module in the target operation and maintenance knowledge system, and determine the operation and maintenance knowledge module with a similarity exceeding a preset scene similarity threshold as a candidate knowledge module.

[0158] Step S1521: convert the scene fault feature set of each operation and maintenance knowledge module into a feature vector form to obtain a knowledge module feature vector of each operation and maintenance knowledge module, the dimension of the knowledge module feature vector of each operation and maintenance knowledge module being consistent with the dimension of the request scene feature vector of the new operation and maintenance service request.

[0159] For each operation and maintenance knowledge module in the target operation and maintenance knowledge system, the scene fault feature set thereof is extracted, and the same word embedding technology and vector splicing method as the request scene feature vector are used to convert it into a knowledge module feature vector.

[0160] For example, the request scene feature vector has several dimensions, and each knowledge module feature vector is adjusted to the same dimension, wherein the dimensions of the vector part corresponding to the fault description, the vector part corresponding to the resource type, and the vector part corresponding to the service target are respectively consistent with the request vector, so as to ensure feature alignment.

[0161] Step S1522: calculate the Euclidean distance between the request scene feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module, and calculate the cosine similarity between the request scene feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module.

[0162] When calculating the Euclidean distance, first, the square of the difference between the numerical values of the corresponding dimensions of the request scene feature vector and the knowledge module feature vector is calculated, then the sum of all square values is calculated, and finally the square root of the sum is taken to obtain the Euclidean distance value, which is smaller, indicating that the two vectors are more similar.

[0163] The method for calculating the cosine similarity is consistent with that of step S122, which is obtained by dividing the vector dot product by the product of the lengths of the two vectors, and the value closer to 1 indicates a higher similarity. The two similarity indicators are calculated simultaneously, which can measure the matching degree of the request and the module from different angles and improve the matching accuracy.

[0164] Step S1523: multiply the normalized result of the Euclidean distance between the request scene feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module by the Euclidean distance weight, multiply the cosine similarity between the request scene feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module by the cosine similarity weight, and add them to obtain a comprehensive similarity value.

[0165] First, the Euclidean distance is normalized to convert it to a value between 0 and 1, and the conversion method is to subtract the current Euclidean distance from the maximum Euclidean distance, and then divide it by the difference between the maximum and minimum Euclidean distances, so that the larger the normalized Euclidean distance value is, the higher the vector similarity is.

[0166] The Euclidean distance normalization result and the cosine similarity are respectively assigned weights, and the weights are set according to historical matching effects, for example, the weight of the cosine similarity is higher than the weight of the Euclidean distance. The normalized Euclidean distance is multiplied by its weight, and the cosine similarity is multiplied by its weight, and the sum of the two is the comprehensive similarity value, which comprehensively reflects the matching degree of the request and the module.

[0167] Step S1524: Mark the operation and maintenance knowledge module with a comprehensive similarity value exceeding the scenario similarity threshold value as a selected knowledge module.

[0168] The preset scenario similarity threshold value is set based on the matching accuracy of historical operation and maintenance services, and the operation and maintenance knowledge module with a comprehensive similarity value exceeding the threshold value is marked as a selected knowledge module. For example, in the comprehensive similarity calculation result of a new request, the comprehensive similarity values of three modules exceed the threshold value, and the three modules are all marked as selected knowledge modules and enter the next step of screening.

[0169] Step S1525: For all selected knowledge modules, the average score of the adaptation feedback of each selected knowledge module in historical applications is counted, the selected knowledge module with an average score exceeding a preset score threshold value is determined as a candidate knowledge module, and the candidate knowledge modules are sorted in descending order of comprehensive similarity values to form a candidate knowledge module sequence.

[0170] The historical adaptation feedback scores of each selected knowledge module are extracted, and the average value of all scores is calculated. The preset score threshold value is used to screen modules with good adaptation effects, and the selected knowledge module with an average score exceeding the threshold value is determined as a candidate knowledge module.

[0171] The candidate knowledge modules are sorted in descending order of comprehensive similarity values to form a candidate knowledge module sequence. The modules with higher rankings in the sequence have higher matching degrees with the new request and better adaptation effects, and are preferentially used as knowledge bases for service response.

[0172] Step S153: Extract the resource constraint conditions in the new operation and maintenance service request, and the resource constraint conditions in the new operation and maintenance service request include available resource type restrictions, resource quantity restrictions, and resource call permission restrictions.

[0173] The resource constraint conditions are extracted from the new operation and maintenance service request. The available resource type restriction refers to the range of resource types that can be called in the user or operation and maintenance environment, for example, "only local resources can be called, and cloud platform interfaces cannot be accessed"; the resource quantity restriction refers to the upper limit of the number of available resources, for example, "only one debugging device is available"; and the resource call permission restriction refers to the range of resource call permissions of the operation and maintenance personnel, for example, "no device firmware upgrade permission".

[0174] Step S154: Analyzing the resource candidate list of each candidate knowledge module, judging whether the resource type in the resource candidate list of the candidate knowledge module meets the available resource type limit in the new operation and maintenance service request, judging whether the resource quantity requirement in the resource candidate list of the candidate knowledge module is within the available quantity range in the new operation and maintenance service request, and judging whether the resource call in the resource candidate list of the candidate knowledge module meets the permission limit in the new operation and maintenance service request.

[0175] The resource candidate list of each candidate knowledge module is analyzed one by one. First, the resource type in the list is compared with the available resource type limit of the new request. For example, if the available resource type limit of the new request is "only local resources", and the resource candidate list of a certain candidate knowledge module contains "cloud platform log query interface" (non-local resource), it is determined that the resource type does not meet the limit; if the list contains "local device debugging tool" and "local router management interface" (both are local resources), it is determined that the type limit is met.

[0176] Then, it is checked whether the resource quantity requirement is within the available quantity range. If the resource quantity limit of the new request is "only one debugging device available", a certain operation step of a certain candidate knowledge module requires "two debugging devices to run simultaneously", it is determined that the resource quantity requirement exceeds the range; if the step requires "a single debugging device to complete", it is determined that the quantity limit is met.

[0177] Finally, it is verified whether the resource call meets the permission limit. If the resource call permission limit of the new request is "no firmware upgrade permission", the resource candidate list of a certain candidate knowledge module contains "firmware upgrade package call" operation, it is determined that the resource call does not meet the permission limit; if the list only contains "device state query" and "parameter adjustment" operations that do not require upgrade permission, it is determined that the permission limit is met.

[0178] Step S155: For the resource type that meets the resource constraint condition in the new operation and maintenance service request, the priority of the resource type in the resource candidate list of the candidate knowledge module is maintained; for the resource type that does not meet the resource constraint condition in the new operation and maintenance service request, the resource type is eliminated from the resource candidate list of the candidate knowledge module, and a substitute resource type that meets the resource constraint condition in the new operation and maintenance service request and the call rule of the substitute resource type are supplemented.

[0179] For each candidate knowledge module, the resource type that meets the constraint condition in the resource candidate list is maintained in its original call priority order. For example, the "local device log reading tool" has a higher priority than the "local parameter detection tool" in the original resource candidate list of a certain module, and both of them meet the constraint condition, so the priority is maintained.

[0180] For resource types that do not meet the constraint conditions, they are directly deleted from the list. For example, resource types such as "cloud platform instruction issuing interface" and "firmware upgrade package resource" that do not meet the "local resource limit" and "no upgrade permission" are removed.

[0181] Subsequently, alternative resource types that meet the constraint conditions are supplemented. After removing the "cloud platform log query interface", a local resource that can achieve the same log query function, such as "device local serial port log export tool", is supplemented, and the calling rules of the alternative resource are clearly stated: "in the operation step 'get device running log', the local serial port log export tool is started first, the device is connected to the debugging terminal through USB, the log export instruction is sent, and the log file is saved to the local directory".

[0182] Step S156: According to the adjusted resource candidate list of the candidate knowledge module, the complete operation sequence of the candidate knowledge module is optimized, and the resource calling scheme corresponding to each operation step in the complete operation sequence of the candidate knowledge module is adjusted, so that the complete operation sequence of the candidate knowledge module matches the resource constraint conditions in the new operation and maintenance service request.

[0183] According to the adjusted resource candidate list, the complete operation sequence of the candidate knowledge module is optimized. If the resource corresponding to a step in the original sequence has been removed and replaced by a supplemented alternative resource, the operation description of the step is adjusted. For example, the original step "call the cloud platform log query interface to get the device log" is optimized to "use the local serial port log export tool to get the device log", and specific operation details such as tool connection and instruction sending are supplemented.

[0184] If there are associated steps that depend on resources that do not meet the constraint conditions in the original sequence, the step logic needs to be redesigned. For example, the original sequence "test function after upgrading firmware" cannot be executed due to "no upgrade permission", and is optimized to "test function after adjusting existing parameters of the device", and the calling scheme of the "local parameter adjustment tool" is matched.

[0185] After optimization, the logical coherence of the operation sequence is ensured, each step of resource calling meets the constraint conditions of the new request, and the operation order meets the operation and maintenance specifications of the Internet of Things smart home device, such as "first check basic faults, then perform deep debugging" and "first hardware detection, then software configuration".

[0186] Step S157: Extract the complete operation sequence in the adjusted candidate knowledge module, the resource candidate list in the adjusted candidate knowledge module, and the fault preplan in the adjusted candidate knowledge module, combine the service target requirements of the new operation and maintenance service request, supplement the execution priority suggestion of the complete operation sequence of the candidate knowledge module and the resource allocation optimization scheme of the candidate knowledge module, and form the operation and maintenance service response content.

[0187] Extract core content from the adjusted candidate knowledge module: complete operation flow sequence (such as "check device power - connect local debugging tool - export device log - analyze log to locate fault - adjust device parameter - test control function"), adjusted resource candidate list (such as "local debugging tool, serial log export tool, parameter adjustment software") and fault plan (such as "when log export fails, reconnect debugging tool after restarting device").

[0188] Combine the service target requirements of the new request (such as "restore the APP control function of the bedroom smart curtain within 1 hour"), supplement the execution priority suggestion: set "connect local debugging tool - export device log" as high priority, require to complete within 10 minutes; set "analyze log - adjust parameter" as medium priority, require to complete within 30 minutes; set "test control function" as low priority, require to complete within the remaining time.

[0189] At the same time, supplement the resource allocation optimization scheme: suggest that the only available local debugging tool be allocated to the "export device log" step first, and then used for the "parameter adjustment" step; if the debugging tool occupies too much time, the built-in debugging function of the smart terminal can be temporarily used as an auxiliary. Integrate the above content into structured operation and maintenance service response content, present it in a clear hierarchical structure, so as to facilitate the quick understanding and execution of operation and maintenance personnel.

[0190] Step S158: Send the operation and maintenance service response content to the terminal device that initiates the new operation and maintenance service request, and record the calling scenario of the candidate knowledge module this time and the adaptation adjustment content of the candidate knowledge module this time.

[0191] Through the message push interface of the operation and maintenance service platform, send the operation and maintenance service response content to the terminal device that initiates the new request, such as the user's smart home APP and the display terminal of the voice assistant. The sending format is adapted according to the terminal type, the APP end adopts the form of combination of text and picture, and the voice assistant end adopts the form of voice broadcast + text summary.

[0192] At the same time, record the candidate knowledge module identifier, the operation and maintenance scenario of the new request (such as "bedroom smart curtain APP control fault - local resource constraint - no upgrade permission") and the adaptation adjustment content (such as "eliminate cloud platform resources, supplement serial log export tool and call rules, optimize operation sequence priority") in the module call log of the target operation and maintenance knowledge system. The above records will serve as an important basis for subsequent knowledge system adjustment and module optimization, ensuring that the target operation and maintenance knowledge system can continuously adapt to diversified operation and maintenance scenario requirements.

[0193] Figure 2An exemplary hardware and software components of the multi-dimensional knowledge extraction system 100 for operation and maintenance technical service provided by some embodiments of the present application are shown in the schematic diagram. For example, the processor 120 can be used in the multi-dimensional knowledge extraction system 100 for operation and maintenance technical service and used to execute the functions in the present application.

[0194] For example, the multi-dimensional knowledge extraction system 100 for operation and maintenance technical service can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, a ROM, or a RAM, or any combination thereof. Exemplarily, the multi-dimensional knowledge extraction system 100 for operation and maintenance technical service can also include program instructions stored in a ROM, a RAM, or other types of non-transitory storage media, or any combination thereof. The methods of the present application can be implemented according to the program instructions. The multi-dimensional knowledge extraction system 100 for operation and maintenance technical service also includes an I / O interface 150 between the computer and other input and output devices.

[0195] In addition, the present application also provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the multi-dimensional knowledge extraction method for operation and maintenance technical service is implemented.

[0196] It should be noted that, in order to simplify the description of the present application and help to understand one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A multi-dimensional knowledge extraction and construction method applied to operation and maintenance technical services, characterized in that: The method comprises: Obtain the full set of operation and maintenance service records generated during the operation and maintenance technical service process, which includes service request texts, execution operation sequences, resource call records, and service effect feedback texts under multiple types of operation and maintenance scenarios, and each type of operation and maintenance scenario corresponds to at least two sets of different service execution process records; Performing association network construction processing on the full set of operation and maintenance service records, establishing association edges between each operation and maintenance service record, and obtaining a target association network structure; Based on the target association network structure, identifying a core record cluster with association in the target association network structure, extracting complementary operation and maintenance information from each operation and maintenance service record in the core record cluster, fusing them to form a scenario-adaptive operation and maintenance knowledge module, and obtaining an operation and maintenance knowledge module set; Performing dynamic knowledge system construction processing on the operation and maintenance knowledge module set to form a target operation and maintenance knowledge system with self-adjustment capability; Based on the target operation and maintenance knowledge system, an operation and maintenance service response content including resource optimization suggestions is generated for a newly received operation and maintenance service request.

2. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 1 is characterized in that: The process of performing association network construction on the full set of operation and maintenance service records, establishing association edges between the operation and maintenance service records, and obtaining a target association network structure includes: Perform problem feature extraction on the service request text of each operation and maintenance service record, extracting problem type keywords, fault phenomenon description phrases, and service goal statements from the service request text to form a request feature vector for each operation and maintenance service record; Calculate the cosine similarity between the request feature vectors of any two operation and maintenance service records, and determine the cosine similarity as the problem similarity of the service request texts of the two operation and maintenance service records; Perform step decomposition processing on the execution operation sequence of each operation and maintenance service record, splitting the execution operation sequence into multiple continuous operation step units, counting the number of completely identical steps in the operation step units of any two operation and maintenance service records, and determining the ratio of the number of steps to the total number of steps in the two operation and maintenance service records as the step overlap of the execution operation sequences of the two operation and maintenance service records; Classify the resource call records of each operation and maintenance service record into hardware resource calls, software resource calls, and network resource calls. Count the number of resource types commonly called by any two operation and maintenance service records. The ratio of the number of common resource types to the total number of resource types in the two operation and maintenance service records is used as the type correlation between the resource call records of the two operation and maintenance service records. Perform sentiment analysis on the service effect feedback text of each operation and maintenance service record to determine whether the evaluation tendency of the service effect feedback text is positive, neutral, or negative. Each evaluation tendency is mapped to a preset continuous numerical score. For any two operation and maintenance service records, calculate the similarity of the evaluation tendency numerical scores of the two, and determine this similarity as the evaluation convergence of the service effect feedback texts of the two operation and maintenance service records; When the problem similarity, step overlap, type association, and evaluation convergence of any two operation and maintenance service records exceed corresponding preset thresholds, an association edge is established between the two operation and maintenance service records, and the standardized weighted average of the problem similarity, step overlap, type association, and evaluation convergence is determined as the association strength value of the association edge; All established association edges and corresponding operation and maintenance service records are taken as nodes and combined to form a target association network structure containing node attributes, edge attributes and association strength values.

3. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 1 is characterized in that: Based on the target association network structure, a core record cluster with association in the target association network structure is identified, complementary operation and maintenance information is extracted from each operation and maintenance service record in the core record cluster, and integrated to form an operation and maintenance knowledge module with scenario adaptability, thereby obtaining an operation and maintenance knowledge module set, including: Calculating the number of associated edges of each node in the target associated network structure, and marking nodes whose number of associated edges exceeds a preset edge number threshold as core nodes; Taking each core node as the center, traverse all nodes connected by the core node's associated edges to form an initial node set centered on the core node. Calculate the ratio of the total number of associated edges between all nodes in the initial node set to the theoretical maximum number of associated edges of all nodes in the initial node set, and determine this ratio as the association density of the initial node set. The initial node set whose association density exceeds the preset density threshold is determined as the core record cluster, and each core record cluster corresponds to a type of operation and maintenance scenario; Perform information complementarity analysis on each operation and maintenance service record in each core record cluster to identify differentiated fault descriptions in the service request texts of different operation and maintenance service records, complementary operation steps in the execution sequence of different operation and maintenance service records, and alternative resource types in the resource call records of different operation and maintenance service records; Extract differentiated fault descriptions from each operation and maintenance service record to form a scenario fault feature set, extract complementary operation steps from each operation and maintenance service record to form a complete operation process sequence, and extract alternative resource types from each operation and maintenance service record to form a resource alternative list; The scenario fault feature set, the complete operation process sequence, and the resource alternative list are associated and integrated, and the resource call rules corresponding to each step in the complete operation process sequence and the fault response plan corresponding to each step in the complete operation process sequence are supplemented to form an operation and maintenance knowledge module that includes scenario features, operation processes, resource configuration, and fault response plans; The above process is repeated for each core record cluster to obtain multiple operation and maintenance knowledge modules, which are then aggregated to form an operation and maintenance knowledge module set.

4. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 3 is characterized in that: The information complementarity analysis and processing of each operation and maintenance service record in each core record cluster is performed to identify the differentiated fault descriptions of the service request texts in different operation and maintenance service records, the complementary operation steps of the execution operation sequences in different operation and maintenance service records, and the alternative resource types of the resource call records in different operation and maintenance service records, including: Cluster the problem feature vectors of the service request texts of all operation and maintenance service records in the core record cluster to obtain multiple problem feature clusters, each of which corresponds to a type of core fault problem; Decompose the fault descriptions of the service request texts in the operation and maintenance service records within each problem feature cluster, extract the fault occurrence environment, fault manifestation, and fault impact scope of each service request text, identify the differentiated descriptions of the same core fault problem in different service request texts, and form a differentiated fault description list; Align the operation step units of the execution operation sequence of all operation and maintenance service records in the core record cluster, and build a standard operation step framework based on the logical relationship of the operation steps; Identify additional operation steps in the execution operation sequence of each operation and maintenance service record that are not included in the standard operation procedure framework, determine whether the additional operation steps can solve special situations not covered by the standard operation procedure framework, and determine the additional operation steps that can solve special situations as complementary operation steps; Classify and count the resource types of resource call records of all operation and maintenance service records in the core record cluster to obtain the call frequency of each resource type, and determine the resource type with a call frequency exceeding the preset frequency threshold as the core resource type; Identify other resource types that are simultaneously called in the operation and maintenance service records that call the core resource type, determine whether the other resource types can complete the same function when the core resource type is unavailable, and determine the other resource types that can complete the same function as alternative resource types; The differentiated fault description list, complementary operation steps and alternative resource types are organized to form the information complementarity analysis results.

5. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 1 is characterized in that: The dynamic knowledge system construction process is performed on the set of operation and maintenance knowledge modules to form a target operation and maintenance knowledge system with self-adjustment capability, including: Initially setting a classification framework for the operation and maintenance knowledge system, which includes three levels: scenario categories, scenario subcategories, and scenario subcategories. Each level is divided into multiple classification labels based on the business attributes of the operation and maintenance service. Assigning each operation and maintenance knowledge module in the operation and maintenance knowledge module set to a corresponding scenario subcategory in the classification framework of the operation and maintenance knowledge system based on the matching degree between the scenario fault feature set of the operation and maintenance knowledge module and the classification labels of each level in the classification framework of the operation and maintenance knowledge system, thereby forming an initial multi-dimensional knowledge system; Record the usage scenario frequency of each operation and maintenance knowledge module in actual applications and the cross-scenario reuse rate of each operation and maintenance knowledge module. The usage scenario frequency is the number of operation and maintenance scenarios corresponding to the operation and maintenance knowledge module being called. The cross-scenario reuse rate is the ratio of the number of times the operation and maintenance knowledge module is called in non-initial allocation scenarios to the total number of times the operation and maintenance knowledge module is called. Collect adaptation feedback of each operation and maintenance knowledge module after its application in a new scenario. The adaptation feedback includes the applicability score of the operation and maintenance knowledge module's operation process, the matching score of the operation and maintenance knowledge module's resource configuration, and the effectiveness score of the operation and maintenance knowledge module's fault plan. When the cross-scenario reuse rate of any operation and maintenance knowledge module exceeds the preset reuse threshold, the common features of multiple scenarios adapted by the operation and maintenance knowledge module are analyzed, and a new scenario subcategory or scenario subcategory corresponding to the common features is added to the classification framework of the operation and maintenance knowledge system. The operation and maintenance knowledge module is simultaneously assigned to the new classification and the original classification of the operation and maintenance knowledge module; When the adaptation feedback score of any operation and maintenance knowledge module is lower than the preset score threshold, the reason for the score being lower than the preset score threshold is analyzed. If the operation process of the operation and maintenance knowledge module is missing, the operation steps of the operation and maintenance knowledge module in the new scenario are supplemented. If the resource configuration of the operation and maintenance knowledge module is mismatched, the resource candidate list of the operation and maintenance knowledge module is updated to form an adjusted operation and maintenance knowledge module. Based on the usage scenario frequency of all operation and maintenance knowledge modules, the cross-scenario reuse rate of all operation and maintenance knowledge modules, and the adaptation feedback of all operation and maintenance knowledge modules, the hierarchical structure of the classification framework of the operation and maintenance knowledge system, the classification labels of the classification framework of the operation and maintenance knowledge system, and the classification attribution of all operation and maintenance knowledge modules are adjusted regularly to form a target operation and maintenance knowledge system with self-adjustment capabilities.

6. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 5 is characterized in that: According to the usage scenario frequency of all operation and maintenance knowledge modules, the cross-scenario reuse rate of all operation and maintenance knowledge modules, and the adaptation feedback of all operation and maintenance knowledge modules, the hierarchical structure of the classification framework of the operation and maintenance knowledge system, the classification labels of the classification framework of the operation and maintenance knowledge system, and the classification attribution of all operation and maintenance knowledge modules are regularly adjusted, including: Set an adjustment cycle. During each adjustment cycle, count the number of operation and maintenance knowledge modules under each scenario subcategory in the classification framework of the operation and maintenance knowledge system. If the number of operation and maintenance knowledge modules under any scenario subcategory exceeds the preset module number threshold, split the scenario subcategory into multiple new scenario subcategories. Each new scenario subcategory corresponds to different sub-scenario characteristics of the operation and maintenance knowledge module. If the number of operation and maintenance knowledge modules under any scenario subcategory is lower than the preset module number threshold, and the similarity of the scenario characteristics of the operation and maintenance knowledge modules of the scenario subcategory and the adjacent scenario subcategory exceeds the preset similarity threshold, then the scenario subcategory will be merged into the adjacent scenario subcategory; Count the average usage frequency of the operation and maintenance knowledge modules under each classification label in the classification framework of the operation and maintenance knowledge system. If the average usage frequency of the operation and maintenance knowledge modules under any classification label is lower than the preset frequency threshold, delete the classification label and reallocate the operation and maintenance knowledge modules under the classification label to other relevant classification labels in the classification framework of the operation and maintenance knowledge system. For each operation and maintenance knowledge module, recalculate the matching degree between the scenario fault feature set of the operation and maintenance knowledge module and the classification labels of each level in the classification framework of the current operation and maintenance knowledge system. Adjust the classification of the operation and maintenance knowledge module based on the new matching degree so that the operation and maintenance knowledge module is assigned to the classification with the highest matching degree. Record the classification framework structure of the operation and maintenance knowledge system each time it is adjusted, the changes in the classification and attribution of the operation and maintenance knowledge modules each time it is adjusted, and the basis for each adjustment to form an adjustment log.

7. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 1 is characterized in that: The generating of an operation and maintenance service response including resource optimization suggestions for a newly received operation and maintenance service request based on the target operation and maintenance knowledge system includes: Receive a new operation and maintenance service request, perform scenario feature extraction on the new operation and maintenance service request, extract the fault description text in the new operation and maintenance service request, the available resource type in the new operation and maintenance service request, and the service target requirement in the new operation and maintenance service request, to form a request scenario feature vector for the new operation and maintenance service request; Calculating the similarity between the request scenario feature vector of the new operation and maintenance service request and the scenario fault feature set of each operation and maintenance knowledge module in the target operation and maintenance knowledge system, and determining the operation and maintenance knowledge modules whose similarity exceeds a preset scenario similarity threshold as candidate knowledge modules; Extracting resource constraints in the new operation and maintenance service request, where the resource constraints in the new operation and maintenance service request include available resource type restrictions, resource quantity restrictions, and resource call permission restrictions; Analyze the resource candidate list of each candidate knowledge module, determine whether the resource type in the resource candidate list of the candidate knowledge module meets the available resource type restriction in the new operation and maintenance service request, determine whether the resource quantity requirement in the resource candidate list of the candidate knowledge module is within the available quantity range in the new operation and maintenance service request, and determine whether the resource call in the resource candidate list of the candidate knowledge module meets the permission restriction in the new operation and maintenance service request; For resource types that meet the resource constraints in the new operation and maintenance service request, the priority of the resource type in the resource candidate list of the candidate knowledge module is retained; for resource types that do not meet the resource constraints in the new operation and maintenance service request, they are removed from the resource candidate list of the candidate knowledge module, and alternative resource types that meet the resource constraints in the new operation and maintenance service request and the calling rules of the alternative resource types are supplemented; Based on the adjusted candidate knowledge module's resource candidate list, optimize the candidate knowledge module's complete operation process sequence and adjust the resource call plan corresponding to each operation step in the candidate knowledge module's complete operation process sequence so that the candidate knowledge module's complete operation process sequence matches the resource constraints in the new operation and maintenance service request; Extract the complete operation process sequence, resource alternative list and fault contingency plan from the adjusted candidate knowledge module, and combine them with the service goal requirements of the new operation and maintenance service request to supplement the execution priority recommendation of the complete operation process sequence of the candidate knowledge module and the resource allocation optimization plan of the candidate knowledge module to form the operation and maintenance service response content; The operation and maintenance service response content is sent to the terminal device that initiates the new operation and maintenance service request, and the calling scenario of the candidate knowledge module and the adaptation adjustment content of the candidate knowledge module are recorded.

8. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 7 is characterized in that: The calculating the similarity between the request scenario feature vector of the new operation and maintenance service request and the scenario fault feature set of each operation and maintenance knowledge module in the target operation and maintenance knowledge system, and determining the operation and maintenance knowledge modules whose similarity exceeds a preset scenario similarity threshold as candidate knowledge modules, includes: Converting the scenario fault feature set of each operation and maintenance knowledge module into a feature vector form to obtain a knowledge module feature vector of each operation and maintenance knowledge module, wherein the dimension of the knowledge module feature vector of each operation and maintenance knowledge module is consistent with the dimension of the request scenario feature vector of the new operation and maintenance service request; Calculating the Euclidean distance between the request scenario feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module, and calculating the cosine similarity between the request scenario feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module; Multiply the normalized result of the Euclidean distance between the request scenario feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module by the Euclidean distance weight, and multiply the cosine similarity between the request scenario feature vector of the new operation and maintenance service request and the knowledge module feature vector of each operation and maintenance knowledge module by the cosine similarity weight, and add the two to obtain a comprehensive similarity value; The operation and maintenance knowledge modules whose comprehensive similarity values ​​exceed the scenario similarity threshold are marked as candidate knowledge modules; For all candidate knowledge modules, the average adaptation feedback score of each candidate knowledge module in historical applications is counted, and the candidate knowledge modules with an average score exceeding the preset score threshold are determined as candidate knowledge modules. The candidate knowledge modules are sorted in descending order according to the comprehensive similarity value to form a candidate knowledge module sequence.

9. The multi-dimensional knowledge extraction and construction method for operation and maintenance technical services according to claim 1 is characterized in that: The full record of operation and maintenance services generated during the acquisition of operation and maintenance technical services includes: Deploy operation and maintenance service record collection nodes, which are respectively connected to the operation and maintenance service request acceptance system, the operation and maintenance operation execution system, the resource scheduling system and the service evaluation feedback system; Collect the submission time of all service requests, the request initiating terminal identifier of all service requests, and the service request text of all service requests from the operation and maintenance service request acceptance system to form a request record subset; Collect the operation execution account corresponding to each service request, the execution operation sequence corresponding to each service request, the operation start time corresponding to each service request, the operation end time corresponding to each service request, and the operation process log corresponding to each service request from the operation and maintenance operation execution system to form an operation record subset; Collecting from the resource scheduling system the resource type, resource identifier, start time, end time, and usage status of each resource called during the execution of each service request, to form a resource record subset. Collect the evaluation submission time after each service request is completed, the evaluation initiating terminal identifier after each service request is completed, the service effect feedback text after each service request is completed, and the score information after each service request is completed from the service evaluation feedback system to form a feedback record subset; According to the unique identifier of the service request, the corresponding records in the request record subset, operation record subset, resource record subset and feedback record subset are associated and integrated to form an operation and maintenance service record containing complete information. All the associated and integrated operation and maintenance service records are summarized to obtain the full record of the operation and maintenance service.

10. A multi-dimensional knowledge extraction and construction system applied to operation and maintenance technical services, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the multi-dimensional knowledge extraction and construction method applied to operation and maintenance technical services as described in any one of claims 1 to 9 above.

Images