Operation and maintenance data retrieval method, device and equipment based on knowledge graph, medium and program product

Abstract

The invention provides an operation and maintenance data retrieval method based on a knowledge graph, and belongs to the technical field of artificial intelligence. The operation and maintenance data retrieval method based on the knowledge graph comprises the following steps: in response to a received operation and maintenance service request, identifying a service scene corresponding to the request; calling a pre-trained weight value model, and calculating weight values of entities and / or relationships in the knowledge graph under the service scene according to entity attributes and real-time parameters involved in the service scene and the operation and maintenance service request; performing directional retrieval on the operation and maintenance service request by using the weight value; and sorting the retrieval results based on the weight values, and returning associated information meeting the expected matching degree with the service scene. The invention further provides an operation and maintenance data retrieval device and equipment based on the knowledge graph, a storage medium and a program product.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and more particularly to a knowledge graph-based operation and maintenance data retrieval method and device, equipment, medium and program product. BACKGROUND

[0002] Current enterprise operation and maintenance data management mainly adopts the following methods: operation and maintenance personnel manually retrieve information from configuration management systems, document libraries and other systems, and generate processing schemes after integration and analysis. Based on traditional databases or search engines, the results are returned by keyword matching (such as device model), and operation and maintenance personnel make operation and maintenance decisions according to the returned results. In the operation and maintenance process, operation and maintenance data naturally has the characteristics of multi-source heterogeneity and contains a large amount of unstructured information. When operation and maintenance personnel handle business, a large amount of unstructured information makes operation and maintenance personnel unable to effectively perform fast retrieval and decision-making. SUMMARY

[0003] In view of the above problems, the present application provides a knowledge graph-based operation and maintenance data retrieval method, device, equipment, medium and program product for improving operation and maintenance data retrieval efficiency.

[0004] According to a first aspect of the present application, a knowledge graph-based operation and maintenance data retrieval method is provided, comprising: in response to receiving an operation and maintenance business request, identifying a business scenario corresponding to the request; calling a pre-trained weight value model, and calculating the weight value of an entity and / or a relationship in a knowledge graph under a business scenario according to the business scenario, entity attributes and real-time parameters involved in the operation and maintenance business request; using the weight value to perform directional retrieval on the operation and maintenance business request within the knowledge graph; sorting the retrieval results based on the weight value, and returning associated information that meets the expected matching degree with the business scenario; wherein the weight value model is trained according to historical operation and maintenance data and weight value samples; and the knowledge graph is constructed according to operation and maintenance heterogeneous data sources.

[0005] According to an embodiment of the present application, the knowledge graph contains scenario nodes, and the scenario nodes have an associated mapping with entities and / or relationships.

[0006] According to an embodiment of the present application, the knowledge graph is obtained according to the following method: extracting data from operation and maintenance heterogeneous data sources, cleaning, transforming and then establishing a metadata model; extracting operation and maintenance ontology in unstructured text in operation and maintenance heterogeneous data sources; defining entities, relationships and attributes of the operation and maintenance ontology; analyzing the process relationship in the unstructured text through natural language processing technology, and completing the missing relationship between entities; storing entities, relationships and attributes based on a graph database, and establishing a knowledge graph.

[0007] According to an embodiment of the present application, the operation and maintenance ontology in unstructured text in operation and maintenance heterogeneous data sources is extracted, including: inputting the unstructured text into a pre-trained ontology extraction model to extract the operation and maintenance ontology; wherein the ontology extraction model is obtained by training according to the following method: converting each word in the unstructured text into a fixed-dimension word piece; adding a label to the word piece and inputting it into a multi-layer encoder to generate a context-related vector representation of the word piece; inputting the context-related vector representation into a fully connected layer to predict a label score corresponding to the word piece; calculating the score of a possible label sequence according to the label transition of the word piece; and maximizing the probability of the correct label sequence.

[0008] According to an embodiment of the present application, the missing relationship between entities is completed, including: first, using a pre-constructed domain knowledge rule library to complete the missing relationship; for the missing relationship that cannot be completed by the domain knowledge rule library, a machine learning model is used to predict the missing relationship; wherein the machine learning model is obtained by learning the feature mode of the existing entities and / or relationships.

[0009] According to an embodiment of the present application, the operation and maintenance business request is directionally searched, including: inputting the operation and maintenance business request into a pre-trained node embedding model to generate a node embedding vector, and directionally searching according to the node embedding vector; wherein the node embedding model is obtained by training according to the following method: determining a target node according to the operation and maintenance business request, randomly sampling a fixed number of neighbor nodes of the target node at each layer of the knowledge graph to form a hierarchical sampling structure; aggregating the features of the neighbor nodes sampled at each layer to generate aggregated features; concatenating the features of the target node at each layer with the aggregated features; and obtaining the node embedding vector of the target node through multi-layer aggregation.

[0010] The second aspect of the present application provides a knowledge graph-based operation and maintenance data retrieval device, including: a weight value calculation module, configured to identify a business scenario corresponding to a received operation and maintenance business request; call a pre-trained weight value model to calculate the weight value of an entity and / or a relationship in a knowledge graph under the business scenario according to the entity attributes and real-time parameters involved in the operation and maintenance business request; and a data retrieval module, configured to directionally search the operation and maintenance business request in the knowledge graph using the weight value; sort the search results based on the weight value, and return associated information that meets the expected matching degree of the business scenario; wherein the weight value model is obtained by training according to historical operation and maintenance data and weight value samples; and the knowledge graph is constructed according to operation and maintenance heterogeneous data sources.

[0011] The third aspect of the present application provides an electronic device, including: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0012] The fourth aspect of the present application also provides a computer readable storage medium, which stores a computer program or instructions, and the computer program or instructions are executed by a processor to realize the steps of the above method.

[0013] The fifth aspect of the present application also provides a computer program product, which comprises a computer program or instructions, and the computer program or instructions are executed by a processor to realize the steps of the above method.

[0014] Embodiments of the present application establish a knowledge graph to connect multi-source data and improve cross-source query efficiency; calculate weight values according to business scenarios, entity attributes and real-time parameters, which makes up for the semantic missing problem of traditional keyword retrieval, accurately captures implicit correlations and avoids missing key information; the weight value model dynamically adjusts the weight values in combination with real-time parameters, which can adapt to the dynamic change characteristics of operation and maintenance data, overcome the lag of the traditional static integration mode of knowledge graph, and ensure that the retrieval results reflect the current state. Through dynamic calculation and scenario-based application of weight values, efficient association, accurate retrieval and real-time adaptation of operation and maintenance data are realized, and the efficiency and quality of operation and maintenance data retrieval are comprehensively improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] The above and other objects, features and advantages of the present application will become more apparent from the following description of embodiments of the present application, taken in conjunction with the accompanying drawings, in which:

[0016] Figure 1 An application scenario diagram of a knowledge graph-based operation and maintenance data retrieval method, device, equipment, medium and program product according to embodiments of the present application is schematically shown;

[0017] Figure 2 A flowchart of a knowledge graph-based operation and maintenance data retrieval method according to embodiments of the present application is schematically shown;

[0018] Figure 3 A knowledge graph construction flowchart according to embodiments of the present application is schematically shown;

[0019] Figure 4 An ontology extraction model training flowchart according to embodiments of the present application is schematically shown;

[0020] Figure 5 A node embedding model training flowchart according to embodiments of the present application is schematically shown;

[0021] Figure 6 A structural block diagram of a knowledge graph-based operation and maintenance data retrieval device according to embodiments of the present application is schematically shown; and

[0022] Figure 7A block diagram of an electronic device suitable for implementing the knowledge graph-based operation and maintenance data retrieval method according to an embodiment of the present application is shown schematically. DETAILED DESCRIPTION

[0023] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. It should be understood, however, that the description is merely exemplary of the present application, and is not intended to limit the scope of the present application. In the following detailed description of the embodiments of the present application, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present application. However, it would be apparent to those skilled in the art that the embodiments of the present application can be practiced without these specific details. In other instances, well-known structures and functions have not been described in detail in order to avoid obscuring aspects of the present application.

[0024] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the term "includes" and tautological expressions thereof, such as "including," "includes," "include," "contains," "containing," and so on, mean the presence of stated features, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0025] All terms used herein (including technical and scientific terms) have the meanings commonly understood by one of ordinary skill in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings that are consistent with the context of the specification, and should not be interpreted in an idealized or overly formal manner.

[0026] In the case of using expressions similar to "at least one of A, B, and C, etc.", it should generally be interpreted to include any of them alone, any combination of two or more of them, and the like in a manner that makes sense in the context of the specification.

[0027] In the process of implementing the present application, the inventors found that operation and maintenance data is scattered in various heterogeneous systems, such as device account, log system, monitoring platform, etc., and there is a lack of effective correlation mechanism between various data sources. When cross-system correlation query is needed, it often relies on manual retrieval of data from different systems and integration, which not only is cumbersome to operate, but also greatly reduces the efficiency of data retrieval and analysis, making it difficult to meet the demand for quickly obtaining comprehensive operation and maintenance information.

[0028] The lack of semantic understanding capability is another prominent shortcoming of the prior art. The traditional keyword retrieval method can only perform literal matching on the text and cannot deeply understand the context logic and semantic association of the operation and maintenance business request. For example, when querying "server verification method", it is difficult to identify the implicit associated information (such as the tool required for verification, the precondition, the historical verification record, etc.) related to the method, resulting in one-sided retrieval results and missing key information, which affects the accuracy of operation and maintenance decision-making.

[0029] In scenarios with extremely high time efficiency requirements such as fault handling, the response delay problem is particularly evident. Due to the lack of automated multi-source data integration mechanism, operation and maintenance personnel need to manually collect, filter and organize relevant data from scattered data sources, which not only takes a long time, but also easily leads to information deviation due to human operation errors. This inefficient processing method directly affects the speed of fault positioning and solving, which may expand the scope of the fault and reduce the quality of operation and maintenance services. The importance of different heterogeneous data sources in the operation and maintenance business flow varies with the scene (such as "equipment account" is more important during daily inspection, and "log data" and "emergency plan" are more critical during fault troubleshooting), and the traditional integration mechanism has fixed weights for data association, which may cause redundant information to interfere with decision-making.

[0030] Embodiments of the present application provide a knowledge graph-based operation and maintenance data retrieval method, which identifies a specific business scenario according to an operation and maintenance business request, determines the weight values of entities and / or relationships in the knowledge graph according to the business scenario, and uses the weight values to retrieve and filter the retrieval results of the operation and maintenance data formed knowledge graph. By establishing a knowledge graph to connect multiple sources of data, the cross-source query efficiency is improved; the weight values are calculated according to the business scenario, entity attributes and real-time parameters, which makes up for the semantic missing problem of traditional keyword retrieval, can accurately capture implicit associations and avoid missing key information; the weight value model dynamically adjusts the weight values in combination with real-time parameters, which can adapt to the dynamic change characteristics of operation and maintenance data, overcome the lag of the traditional static integration mode of knowledge graph, and ensure that the retrieval results reflect the current state. Through dynamic calculation and scenario-based application of weight values, efficient association, accurate retrieval and real-time adaptation of operation and maintenance data are achieved, and the efficiency and quality of operation and maintenance data retrieval are comprehensively improved.

[0031] Figure 1 An application scenario diagram of a knowledge graph-based operation and maintenance data retrieval method, device, equipment, medium and program product according to an embodiment of the present application is schematically shown.

[0032] As Figure 1As shown, the application scenario 100 according to this embodiment can include a first terminal device 101, a second terminal device 102, and a third terminal device 103. A network 104 is a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and a server 105. The network 104 can include various connection types, such as wired, wireless communication links, or optical fiber cables, and the like.

[0033] A user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, and the like. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, and the like (only as examples).

[0034] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smartphones, tablet computers, laptop computers, desktop computers, and the like.

[0035] The server 105 can be a server providing various services, such as a background management server providing support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (only as an example). The background management server can analyze and process received user requests and the like, and feed back the processing results (such as web pages, information, or data, and the like obtained or generated according to user requests) to the terminal devices.

[0036] It should be noted that the knowledge graph-based operation and maintenance data retrieval method provided by the embodiments of the present application can generally be executed by the server 105. Correspondingly, the knowledge graph-based operation and maintenance data retrieval apparatus provided by the embodiments of the present application can generally be arranged in the server 105. The knowledge graph-based operation and maintenance data retrieval method provided by the embodiments of the present application can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105. Correspondingly, the knowledge graph-based operation and maintenance data retrieval apparatus provided by the embodiments of the present application can also be arranged in a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105.

[0037] It should be understood that, Figure 1The number of terminal devices, networks and servers in the figure is only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.

[0038] The following will be based on Figure 1 The described scenario, by Figures 2-5 The knowledge graph-based operation and maintenance data retrieval method according to the embodiments of the present application is described in detail.

[0039] Figure 2 The flowchart of the knowledge graph-based operation and maintenance data retrieval method according to the embodiments of the present application is schematically shown.

[0040] As Figure 2 The knowledge graph-based operation and maintenance data retrieval method of this embodiment includes operations S210-S230, which can be executed in the following manner.

[0041] In operation S210, in response to receiving an operation and maintenance service request, the corresponding business scenario is identified.

[0042] In operation S220, a pre-trained weight value model is called, and the weight value of the entity and / or relationship in the knowledge graph under the business scenario is calculated according to the entity attribute and real-time parameter involved in the business scenario and the operation and maintenance service request. The weight value model is trained according to historical operation and maintenance data and weight value samples.

[0043] In operation S230, the operation and maintenance service request is directionally searched in the knowledge graph using the weight value; the search results are sorted based on the weight value, and the associated information that meets the expected matching degree with the business scenario is returned. The knowledge graph is constructed according to the operation and maintenance heterogeneous data source.

[0044] The importance of different heterogeneous data sources in the operation and maintenance business flow changes with the scene. The embodiments of the present application divide different operation and maintenance scenes and preset core data sources for each scene. Specifically, different operation and maintenance scenes may, for example, include scenes such as "troubleshooting", "daily inspection" and "capacity planning", and each scene can be further divided into sub-scenes according to needs. For example, the "troubleshooting" scene can include the "server downtime emergency recovery" sub-scene.

[0045] In operation S210, the corresponding business scenario can be identified by using keywords, natural language processing, combining rules and pre-training models.

[0046] For example, the following operation and maintenance service requests and the corresponding business scenarios identified are as follows.

[0047] (1) When receiving the request "Server A suddenly goes down, how to quickly recover the service?", the core elements "server down" and "recover service" can be extracted by using natural language processing, and the pre-trained model is combined with the classification of operation and maintenance field scenes (such as "fault emergency handling", "daily inspection", etc.) to identify that the corresponding business scene is "server down emergency recovery".

[0048] (2) For the business request "Check the configuration change record of the firewall in the last 3 days", through semantic analysis of the keywords "firewall configuration change" and "last 3 days", it is matched to the "network configuration audit" scene, which focuses on the historical change tracing of device configuration, and is highly consistent with the query target of the request.

[0049] (3) If the request is "The CPU utilization of database B has been continuously over 90% for a long time, what could be the reason?", natural language processing will capture key information such as "CPU utilization is too high" and "reason analysis", and combined with the mapping relationship of "resource overload" and "performance troubleshooting" in the rule library, it identifies that the corresponding scene is "database performance abnormality troubleshooting".

[0050] (4) For the request "What servers' ports need to be bound for the newly deployed application C?", by analyzing the business intent of "newly deployed application" and "port binding", it is matched to the "application deployment configuration" scene, which is specifically designed to handle the associated configuration requirements of new applications and infrastructure.

[0051] In operation S220, the pre-trained weight value model is called, and the weight values of entities and / or relationships in the knowledge graph under the business scene are calculated according to the entity attributes and real-time parameters involved in the operation and maintenance business request.

[0052] The weight value model in this embodiment can be pre-trained, and by inputting the entity attributes and real-time parameters involved in the operation and maintenance request into the weight value model, the weight values of entities and / or relationships in the knowledge graph under the business scene can be calculated.

[0053] Embodiment 1, in the server down emergency recovery business scene.

[0054] The entity attributes involved in the operation and maintenance request are: server identity (Identity, abbreviated as ID), fault level (such as P0 level, highest level).

[0055] Real-time parameters: current state of server A (down), online user quantity of associated application B (more than 100,000).

[0056] Weight value model calculation: According to the pre-training rule, in the fault scenario of the core server, the basic weight value of "server A-association-emergency plan X" is 0.8; because the fault level is P0, the weight value is increased by 30% to 1.04; and because the user quantity of the associated application B exceeds the threshold, the relationship weight value of "server A-impact-application B" is increased from 0.6 to 0.8, ensuring that the emergency plan and core impact object are preferentially obtained during emergency handling.

[0057] Example 2, database performance abnormality troubleshooting business scenario.

[0058] The operation and maintenance request involves entity attributes: database ID (main database C), fault characteristics (CPU utilization 95%).

[0059] Real-time parameters: connection number of database C (500+, far exceeding the threshold 200), number of slow query in the last 30 minutes (20 times).

[0060] Weight value model calculation: the basic rule of "database C-association-slow query log" is 0.7; because the connection number exceeds the standard, the relationship weight value of "database C-dependent-connection pool configuration" is increased from 0.5 to 0.8; combined with the number of slow queries, the weight value of "database C-association-index optimization scheme" is increased from 0.6 to 0.9, so that the connection pool and index related high weight value association information are preferentially focused during troubleshooting.

[0061] In the above examples, the weight value model determines the basic weight value framework through the business scenario, and then dynamically adjusts the entity attribute (core server / P0 fault) and real-time parameter (user quantity / connection number), to ensure that the weight value of the entity and / or relationship in the knowledge graph is highly matched with the current business demand.

[0062] In operation S230, the weight value is used in the knowledge graph to perform directional retrieval on the operation and maintenance business request; the retrieval results are sorted based on the weight value, and the associated information that meets the expected matching degree of the business scenario is returned. The "expected matching degree" in this embodiment can be the top few retrieval results, for example, the top 4 retrieval results, or the first retrieval result.

[0063] Example 1, server downtime emergency recovery scenario.

[0064] Based on the calculated weight values ("Server A-Associated-Emergency Plan X" weight value 1.04, "Server A-Effect-Application B" weight value 0.8, "Server A-Historical Purchase Record" weight value 0.1), the associated data with weight value ≥ 0.7 is directly filtered through the "scenario-weight value" bidirectional index, excluding irrelevant information such as "historical purchase record" with low weight value, and retaining core data such as emergency plan X, application B, log path of server A (weight value 0.9), associated network switch (weight value 0.75), etc.

[0065] After descending order by weight value, the returned associated information is as follows: emergency plan X (weight value 1.04), directly corresponding to the core demand of "quick recovery service"; log path of server A (weight value 0.9), used to investigate the specific reason for downtime; server A-effect-application B (weight value 0.8), prompting the need to prioritize temporary switching scheme for application B; associated network switch (weight value 0.75), assisting in investigating whether the downtime is caused by network interruption. The above results all meet the expected matching degree of the "emergency recovery" scenario, ensuring that the operation and maintenance personnel can prioritize the most critical recovery steps and impact range information.

[0066] Example 2, database performance anomaly investigation scenario.

[0067] Based on the calculated weight values ("Database C-Associated-Index Optimization Scheme" weight value 0.9, "Database C-Dependence-Connection Pool Configuration" weight value 0.8, "Database C-Associated-Slow Query Log" weight value 0.7, "Database C-Historical Backup Record" weight value 0.2), the associated data with weight value ≥ 0.7 is filtered through the index, excluding "historical backup record" with low weight value, and retaining key information such as index optimization scheme, connection pool configuration, slow query log, and CPU monitoring curve of database C (weight value 0.72).

[0068] After descending order by weight value, the returned associated information is as follows: index optimization scheme (weight value 0.9), directly corresponding to the core optimization demand of "high CPU utilization"; database C-dependence-connection pool configuration (weight value 0.8), used to adjust the connection threshold to alleviate resource occupation; CPU monitoring curve of database C (weight value 0.72), assisting in locating the time node of CPU surge; slow query log (weight value 0.7), used to analyze whether there is a performance bottleneck caused by inefficient SQL. The above results all focus on the business demand of the "performance anomaly investigation" scenario, helping operation and maintenance personnel to quickly lock the optimization direction and improve the investigation efficiency.

[0069] In the knowledge graph of the embodiment, the weight value of an entity and / or a relationship is a numerical measure of "entity importance" or "relationship strength, reliability", which is used to quantify the value, correlation tightness or reliability of elements in the knowledge graph. The weight value supplements the degree information of entities and / or relationships through numericalization, enabling the knowledge graph to change from a static knowledge network to a dynamic one that can adapt to different operation and maintenance scenarios. In the embodiment, the weight value can be used not only to quickly filter high-relevance data in operation S230 (by directly locating through indexing), but also to sort the filtering results in operation S240.

[0070] The embodiment connects multiple sources of data by establishing a knowledge graph, improving cross-source query efficiency; calculates weight values according to business scenarios, entity attributes and real-time parameters, making up for the semantic missing problem of traditional keyword retrieval, accurately capturing implicit correlations and avoiding missing of key information; the weight value model dynamically adjusts the weight values in combination with real-time parameters, which can adapt to the dynamic change characteristics of operation and maintenance data, overcome the lag of the traditional static integration mode of knowledge graph, and ensure that the search results reflect the current state. Through dynamic calculation and scenario-based application of weight values, efficient correlation, accurate retrieval and real-time adaptation of operation and maintenance data are achieved, and the efficiency and quality of operation and maintenance data retrieval are improved.

[0071] In some embodiments of the present application, the knowledge graph contains scenario nodes, which have an associated mapping with entities and / or relationships.

[0072] In the embodiment, the scenario node is closely related to the operation and maintenance business request, and can be defined based on operation and maintenance task classification, and specifically can contain scenario attributes and association rules.

[0073] In the knowledge graph of the embodiment, a "scenario node" is created separately as a standardized carrier of scenarios; for example, it can be represented as: scenario { identifier: fault troubleshooting-server downtime, level: 0 level, business type: core business}.

[0074] The entity and relationship are set with "scenario-weight value" attributes as follows.

[0075] Entity level: Add a "related scenario" list to the node, for example: server node is associated with ["fault troubleshooting", "daily inspection"].

[0076] Relationship level: Add a scenario weight value dictionary to the relationship (such as "server A-dependent-application B"), for example: {"fault troubleshooting": 0.9, "daily inspection": 0.4}, which stores the weight values of the relationship in different scenarios.

[0077] Through the association mapping of the scenario node with the entity and / or the relationship (such as scenario-association-relation), a traceable link of “scenario-relation-entity” (for example, troubleshooting scenario-association-server A-relation-application B) is formed instead of the traditional “entity-relation-entity” binary structure.

[0078] The knowledge graph of the embodiment can anchor the entity and the relationship to a specific scenario, give them a scenario context, avoid information isolation, and thus improve the adaptability and application accuracy of knowledge in actual business scenarios.

[0079] In some embodiments of the present application, the knowledge graph is obtained according to the following method: data is extracted from the operation heterogeneous data source, cleaned, and transformed to establish a metadata model; the operation ontology in the unstructured text in the operation heterogeneous data source is extracted; the entity, relationship, and attribute of the operation ontology are defined; the process relationship in the unstructured text is analyzed by using natural language processing technology, and the missing relationship between entities is completed; the entity, relationship, and attribute are stored based on a graph database to establish the knowledge graph.

[0080] Figure 3 A knowledge graph construction flowchart according to an embodiment of the present application is schematically shown. As shown in Figure 3 The knowledge graph construction flowchart includes operations S310-S350.

[0081] In operation S310, data is extracted from the operation heterogeneous data source, cleaned, and transformed to establish a metadata model.

[0082] In the embodiment, the operation heterogeneous data source can include at least one of a device account book, a maintenance manual, an emergency plan, an application information account book, and a firewall configuration. By extracting data from the above data sources and establishing a unified metadata model, unified entity naming, unified attribute definition, and unified relationship type are realized. For example, “host node” and “server” are unified as “server”; at the same time, the core attributes (such as ID, status, and belonging cluster) that must be included in “server” are defined; and the relationship description between entities (such as “server-deployment-application” and “application-dependence-database”) is standardized, replacing the vague “association” and “connection” in each data source.

[0083] In operation S320, the operation ontology in the unstructured text in the operation heterogeneous data source is extracted. The embodiment can extract the operation ontology in the unstructured text in the operation heterogeneous data source by using a pre-trained model. The ontology is a core entity in the operation field and provides a basis for subsequent knowledge graph construction and association analysis. For example, the ontology can be a log path, an upstream job, and a server ID.

[0084] In operation S330, entities, relationships and attributes of the operation and maintenance ontology are defined. Through this operation, scattered operation and maintenance objects, associations and feature systems are systematized, providing a basis for subsequent automated analysis, knowledge graph construction, etc. Entities may be, for example, servers, applications and jobs, etc. Relationships may include, for example, dependency relationships, membership relationships, etc., through which associated impacts can be quickly located. Attributes may be, for example, state, version and other related information.

[0085] In operation S340, process relationships in unstructured text are analyzed through natural language processing technology, and missing relationships between entities are completed. Process relationships refer to logical relationships such as operation step order, conditional dependency, causal association, etc. implied in the text. This embodiment can first extract key entities from the text using named entity recognition technology, and then analyze the logical association between entities using a relationship extraction model. For example, from the text "When the payment service is timed out, first check the upstream order interface, and then check the database connection pool", the process relationship is parsed as: "payment service timeout" (trigger condition) - "check order interface" (step 1) - "check database connection pool" (step 2).

[0086] Unstructured text often does not record all entity relationships completely, and completing missing relationships can be based on the parsed process relationships, combined with existing knowledge, to deduce entity associations that are not explicitly mentioned in the text, making the association network between entities more complete.

[0087] In operation S350, entities, relationships and attributes are stored based on a graph database to establish a knowledge graph.

[0088] This embodiment effectively solves the data silo problem in the operation and maintenance scenario by extracting, cleaning and transforming data from operation and maintenance heterogeneous data sources and establishing a metadata model, achieving unified integration of multi-source data; fills in the implied associations between entities by analyzing process relationships and completing missing relationships with the help of natural language processing technology, improving the completeness and semantic coherence of the graph; and finally stores entities, relationships and attributes based on a graph database to establish a knowledge graph, supporting efficient knowledge storage and query, and adapting to dynamic business needs such as real-time fault troubleshooting and impact analysis in the operation and maintenance scenario.

[0089] In some embodiments of the present application, the operation and maintenance ontology in the unstructured text in the operation and maintenance heterogeneous data source is extracted, including: inputting the unstructured text into a pre-trained ontology extraction model to extract the operation and maintenance ontology. Figure 4 An ontology extraction model training flowchart according to an embodiment of the present application is schematically shown. As shown in Figure 4 The ontology extraction model is obtained according to operations S410-S450.

[0090] In operation S410, each word in the unstructured text is converted into a fixed-dimensional token. For example, each word in the unstructured text can be converted into a fixed-dimensional vector token using a 64-bit word vector model, and then special markers (such as [sentence beginning], [separator], etc.) are added before being input into the context encoding module.

[0091] In operation S420, after adding tags to the lexical units, the input is fed into the multi-layer encoder to generate context-related vector representations of the lexical units.

[0092] The context encoding module generates a context-dependent vector representation for each token through a multi-layer encoder, for example: h i =Shangxiawen(x1,x2,...,x n )[i.

[0093] Where x1,x2,...,x n The word vector representing the sentence, h i It is the semantic vector of the i-th token, and Shangxiawen refers to the operation of the context encoding module.

[0094] In operation S430, the context-related vector representation is input into the fully connected layer to predict the label score corresponding to each word.

[0095] Each token vector h output by the context encoding module i Given a fully connected layer, predict the score of each token for each tag, as shown in the following formula.

[0096]

[0097] Where W∈R n×m The sentence length is n, the number of tags is m, and s i,j This represents the emission score of the i-th token belonging to the j-th tag.

[0098] In operation S440, based on the tag transfer of the lexical, the scores of possible tag sequences are calculated (e.g., B-PER (tag name) to I-PER (tag name) is reasonable, but I-ORG (tag name) to B-PER (tag name) is not reasonable).

[0099] Maintain a state transition matrix T, T∈R m×m T k,l This represents the score at which the label is transferred from label k to label l.

[0100] Calculate the scores for all possible label sequences for a sentence X=[x1,x2] 2, ...,x n ] and the label sequence y=[y1,y2,...,yn ] is computed as follows.

[0101]

[0102] where, is the emission score, indicating how well the element x i at position i is assigned the label y i . is the transition score, indicating how reasonable it is to follow the label y i with the label y i+1 .

[0103] The highest scoring path is efficiently found by the dynamic programming algorithm, avoiding the exhaustive enumeration of all possibilities. The goal is to find the sequence of labels y that maximizes Score(y) which is given by

[0104]

[0105] The optimal path score for each label at each position is computed step by step. The score for each label for the first token = emission score + start transition score. For the ith token and label l, compute: where δ i (l) is the highest score up to the ith token with label l. The path is traced backwards from the last token.

[0106] At operation S450, the probability of the correct sequence of labels is maximized. The loss function uses the negative log-likelihood loss to maximize the probability of the correct sequence of labels, given by

[0107]

[0108] where the denominator is the sum of the scores of all possible sequences of labels. Score(y true ) represents the raw score of the model for the "correct sequence of labels y true ", and Score(y') represents the raw score of the model for any sequence of labels y'.

[0109] The embodiment converts words into fixed-dimension wordpieces and adds labels to ensure that unstructured text is standardized; a multi-layer encoder generates context-related vectors, which can capture the meaning of words in a specific context and solve the ambiguity problem of professional terms in the operation and maintenance field; the label score predicted by the full connection layer is combined with the label transfer to calculate the sequence score, which can accurately identify a single operation and maintenance entity to ensure the accuracy of the entity boundary and avoid mistakenly splitting related terms into unrelated entities; through training, the model is continuously optimized on the operation and maintenance field corpus to improve the recognition ability of professional terms. In summary, the embodiment can efficiently and accurately extract operation and maintenance entities from unstructured text and provide high-quality entities for the knowledge graph to ensure the reliability of subsequent correlation analysis and retrieval.

[0110] In some embodiments of the present application, the missing relationship between entities is completed, including: first, using a pre-constructed domain knowledge rule base to complete the missing relationship; for the missing relationship that cannot be completed by the domain knowledge rule base, a machine learning model is used to predict the missing relationship; wherein the machine learning model learns the feature mode of the existing entity and / or relationship.

[0111] Firstly, the constructed domain knowledge rule base (such as “application deployment path contains server storage path, then there is a ‘deployed on’ relationship” in the operation and maintenance field) can directly complete the explicit relationship that meets the rules based on clear business logic without complex calculation, ensuring the accuracy and completion efficiency of these relationships. Secondly, for complex implicit relationships (such as “indirect association between server CPU overload and application response delay”) that cannot be covered by the rule base, the machine learning model can mine non-explicit association rules by learning the feature mode (such as the co-occurrence frequency of the two in historical faults and the relevance of resource occupation) of the existing entity and / or relationship, making up for the insufficient coverage of the rule base for dynamic and complex scenarios.

[0112] In the embodiment, the domain knowledge rule base and the machine learning model are used to complete the missing relationship between entities, which combines the accuracy of basic relationships guaranteed by the rule base and the expansion of the range of relationship completion by machine learning, ultimately achieving efficient and comprehensive completion of the missing relationship between entities, making the knowledge graph not only reflect clear business logic but also capture implicit association rules, and better meet the actual needs of complex fields such as operation and maintenance.

[0113] In some embodiments of the present application, the operation and maintenance service request is subjected to directional retrieval, including: inputting the operation and maintenance service request into a pre-trained node embedding model to generate a node embedding vector, and performing directional retrieval according to the node embedding vector; wherein the node embedding model is obtained by training the following method: determining a target node according to the operation and maintenance service request, randomly sampling a fixed number of neighbor nodes of the target node at each layer to form a hierarchical sampling structure. The features of the neighbor nodes sampled at each layer are aggregated to generate aggregated features. The features of the target node at each layer are spliced with the aggregated features. After multi-layer aggregation, the node embedding vector of the target node is obtained.

[0114] Figure 5 The node embedding model training flowchart according to an embodiment of the present application is schematically shown. As shown in Figure 5 , the operation S510~S540 are included.

[0115] At operation S510, a target node is determined according to an operation and maintenance service request, and a fixed number of neighbor nodes of the target node are randomly sampled at each layer of a knowledge graph to form a hierarchical sampling structure.

[0116] For example, for a certain target node, a fixed number of neighbors are randomly sampled at each layer.

[0117] Layer 1: Sampling the direct neighbors of the target node.

[0118] Layer 2: Sampling the neighbors of each neighbor node at layer 1 (i.e., the second-order neighbors of the target node).

[0119] By analogy, layer k: sampling the neighbors of each neighbor node at layer k-1 (i.e., the k-order neighbors of the target node). Form a hierarchical sampling structure.

[0120] At operation S520, the features of the neighbor nodes sampled at each layer are aggregated to generate aggregated features .

[0121] When aggregating neighbor features, methods such as averaging, time-dependent or highlighting key features can be generally used, for example, the pooling aggregation method is represented as follows.

[0122]

[0123] In the formula, represents the maximum value of the calculation results of all neighbor nodes u belonging to . represents the feature vector of neighbor node u at the k-1 layer of the model; W pool represents a learnable weight matrix for linear transformation of neighbor features; b represents a bias term; represents an activation function.

[0124] Max-pooling after using a nonlinear transformation (Rectified Linear Unit, abbreviated as ReLU), which captures significant features in neighbors, where v represents the node of the current layer, and N(v) represents the sampled nodes for v. represent the features of the sampled nodes.

[0125] In operation S530, the features of the target node at each layer are spliced with the aggregated features.

[0126] The current layer node features of the target node are combined with the aggregated neighbor features as follows.

[0127]

[0128] where W k represents the learnable weight value matrix of the kth layer; and σ represents the activation function ReLU. represents the feature splicing operation.

[0129] In operation S540, after multi-layer aggregation, the node embedding vector of the target node is obtained.

[0130] Through the multi-layer stacking of S510-S530 operations, multi-order neighbor information is captured.

[0131] Layer 1: aggregate direct neighbors (1-hop).

[0132] Layer 2: aggregate two-order neighbors (2-hop).

[0133] Similarly, layer k: aggregate k-order neighbors (k-hop).

[0134] Use the cross-entropy loss function for training, and the loss function is as follows.

[0135]

[0136] where θ is the classifier parameter; is a single node in the set; represents the set of nodes with real labels in the data set; represents the real label of ; f represents the activation function; represents the inner product of the feature vector and the parameter vector.

[0137] ​In this embodiment, the node embedding model randomly samples a fixed number of neighbors for each layer of the target node during training, avoiding full graph traversal, greatly reducing computational overhead, and improving model training efficiency and subsequent retrieval response speed. Secondly, by aggregating neighbor features and concatenating each layer of the target node features, the generated node embedding vector not only contains the node's own information, but also integrates the context association in the graph, preserving the structural relationship of the knowledge graph and ensuring that the vector accurately reflects the node's semantic and association characteristics. Finally, based on the directional retrieval of the node embedding vector, the clustering characteristics of similar vectors in the vector space can be used to quickly locate the nodes related to the request, avoiding irrelevant information interference and improving the accuracy of the retrieval.

[0138] This embodiment can generate embeddings for new nodes that have not been seen in the training phase, and is suitable for dynamically changing graphs. By sampling a fixed number of neighbors layer by layer and aggregating neighbor information, full graph traversal is avoided, and computational efficiency is improved. This embodiment realizes automatic matching of fault association information, reduces manual intervention, and improves emergency response speed.

[0139] Based on the above knowledge graph-based operation and maintenance data retrieval method, the application also provides a knowledge graph-based operation and maintenance data retrieval device. The following will be combined with Figure 6 to describe the device in detail.

[0140] Figure 6 The structure block diagram of the knowledge graph-based operation and maintenance data retrieval device according to the embodiments of the application is schematically shown.

[0141] As Figure 6 shown, the knowledge graph-based operation and maintenance data retrieval device 600 of this embodiment includes a weight value calculation module 610 and a data retrieval module 620.

[0142] The weight value calculation module 610 is configured to, in response to receiving an operation and maintenance service request, identify the business scenario corresponding to the request; call a pre-trained weight value model, and calculate the weight value of the entity and / or relationship in the knowledge graph under the business scenario according to the business scenario, the entity attributes and real-time parameters involved in the operation and maintenance service request.

[0143] The data retrieval module 620 is configured to use the weight value to perform directional retrieval on the operation and maintenance service request; sort the retrieval results based on the weight value, and return the association information that meets the expected matching degree with the business scenario.

[0144] The weight value model is trained according to historical operation and maintenance data and weight value samples; and the knowledge graph is constructed according to operation and maintenance heterogeneous data sources.

[0145] According to the embodiments of the application, the knowledge graph contains scenario nodes, and the scenario nodes have an associated mapping with entities and / or relationships.

[0146] According to an embodiment of the present application, the knowledge graph is obtained according to the following method: extracting data from operation and maintenance heterogeneous data sources, cleaning, transforming, and then establishing a metadata model; extracting operation and maintenance ontology in unstructured text in the operation and maintenance heterogeneous data sources; defining entities, relationships and attributes of the operation and maintenance ontology; analyzing the process relationship in the unstructured text through natural language processing technology, and completing the missing relationship between entities; storing entities, relationships and attributes based on a graph database, and establishing a knowledge graph.

[0147] According to an embodiment of the present application, the operation and maintenance ontology in the unstructured text in the operation and maintenance heterogeneous data sources is extracted, including: inputting the unstructured text into a pre-trained ontology extraction model to extract the operation and maintenance ontology; wherein the ontology extraction model is obtained by training according to the following method: converting each word in the unstructured text into a fixed-dimensional word piece; adding a label to the word piece and inputting it into a multi-layer encoder to generate a context-related vector representation of the word piece; inputting the context-related vector representation into a fully connected layer to predict the label score corresponding to the word piece; calculating the score of the possible label sequence according to the label transition of the word piece; and maximizing the probability of the correct label sequence.

[0148] According to an embodiment of the present application, the missing relationship between entities is completed, including: first, completing the missing relationship through a pre-constructed domain knowledge rule library; for the missing relationship that cannot be completed by the domain knowledge rule library, a machine learning model is used to predict the missing relationship; wherein the machine learning model is obtained by learning the feature mode of the existing entities and / or relationships.

[0149] According to an embodiment of the present application, the data retrieval module 620 is used for directional retrieval of the operation and maintenance business request, including: inputting the operation and maintenance business request into a pre-trained node embedding model to generate a node embedding vector, and performing directional retrieval according to the node embedding vector; wherein the node embedding model is obtained by training according to the following method: determining a target node according to the operation and maintenance business request, randomly sampling a fixed number of neighbor nodes of the target node at each layer of the knowledge graph to form a hierarchical sampling structure; aggregating the features of the sampled neighbor nodes at each layer to generate aggregated features; concatenating the features of the target node at each layer with the aggregated features; and obtaining the node embedding vector of the target node through multi-layer aggregation.

[0150] According to an embodiment of the present application, any of the plurality of modules in the weight value calculation module 610 and the data retrieval module 620 can be combined in one module, or any of the modules can be split into a plurality of modules. Alternatively, at least part of the functions of one or more of the modules can be combined with at least part of the functions of the other modules, and implemented in one module. According to an embodiment of the present application, at least one of the weight value calculation module 610 and the data retrieval module 620 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on board, a system on package, an application specific integrated circuit (ASIC), or any other reasonable manner of integrating or packaging a circuit, etc. in hardware or firmware, or implemented in any one of software, hardware and firmware or in a proper combination of any of them. Alternatively, at least one of the weight value calculation module 610 and the data retrieval module 620 can be at least partially implemented as a computer program module which, when executed, can perform the corresponding functions.

[0151] Figure 7 A block diagram of an electronic device suitable for implementing the knowledge graph-based operation and maintenance data retrieval method according to an embodiment of the present application is schematically shown.

[0152] As shown in Figure 7 The electronic device 700 according to an embodiment of the present application includes a processor 701 which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 702 or loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 can include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset, and / or a special-purpose microprocessor (e.g., an application specific integrated circuit (ASIC)), etc. The processor 701 can also include an on-board memory for cache use. The processor 701 can include a single processing unit or a plurality of processing units for performing different actions of the method processes according to embodiments of the present application.

[0153] In the RAM 703, various programs and data required for the operation of the electronic device 700 are stored. The processor 701, the ROM 702 and the RAM 703 are connected to each other through a bus 704. The processor 701 performs various operations of the method processes according to embodiments of the present application by executing the programs in the ROM 702 and / or the RAM 703. It should be noted that the programs can also be stored in one or more memories other than the ROM 702 and the RAM 703. The processor 701 can also perform various operations of the method processes according to embodiments of the present application by executing the programs stored in the one or more memories.

[0154] According to an embodiment of the present application, the electronic device 700 can further include an input / output (I / O) interface 705 also connected to the bus 704. The electronic device 700 can further include one or more of the following components connected to the input / output (I / O) interface 705: an input part 706 including a keyboard, a mouse, etc.; an output part 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 708 including a hard disk, etc.; and a communication part 709 including a network interface card such as a LAN card, a modem, etc. The communication part 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as necessary. A removable medium 711 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is mounted on the drive 710 as necessary, so that a computer program read out therefrom is installed in the storage part 708 as necessary.

[0155] The present application also provides a computer readable storage medium, which can be included in the device / apparatus / system described in the above embodiments, or can exist separately without being assembled into the device / apparatus / system. The above computer readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present application.

[0156] According to an embodiment of the present application, the computer readable storage medium can be a non-volatile computer readable storage medium, which can include, but is not limited to, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any appropriate combination thereof. In the present application, the computer readable storage medium can be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present application, the computer readable storage medium can include the ROM 702 and / or the RAM 703 described above, and / or one or more memory other than the ROM 702 and the RAM 703.

[0157] The embodiments of the present application also include a computer program product, which includes a computer program containing program codes for executing the method shown in the flow chart. When the computer program product is run in a computer system, the program codes are used to make the computer system implement the knowledge graph-based operation and maintenance data retrieval method provided by the embodiments of the present application.

[0158] The above-described functions of the system / apparatus defined in the embodiments of the present application are performed when the computer program is executed by the processor 701. According to the embodiments of the present application, the system, apparatus, module, unit, etc. described above can be implemented by the computer program modules.

[0159] In one embodiment, the computer program can be stored in a tangible storage medium, such as an optical, magnetic, or other memory on a server, computer, or other computing device. In another embodiment, the computer program can be transmitted over a network, including the Internet, WAN, LAN, etc., over a wired or wireless connection, using signals, and downloaded and installed by a communication interface 709 and / or removable medium 711. The computer program code contained in the computer program can be executed using any suitable network medium, including but not limited to wireless, wireline, or any suitable combination of the above.

[0160] In such an embodiment, the computer program can be downloaded and installed from a network, using the communication interface 709 and / or removable medium 711. When the computer program is executed by the processor 701, the above-described functions of the system defined in the embodiments of the present application are performed. According to the embodiments of the present application, the system, apparatus, device, module, unit, etc. described above can be implemented by the computer program modules.

[0161] According to the embodiments of the present application, the program code for carrying out the computer program provided by the embodiments of the present application can be written in any combination of one or more programming languages, and can specifically be implemented using high-level procedural and / or object-oriented programming language, and / or assembly / machine language. The programming language includes, but is not limited to, such as Java, C++, python, "C" language, or similar programming language. The program code can be executed entirely on a user computing device, partially on a user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case involving a remote computing device, the remote computing device can be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0162] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0163] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A knowledge graph-based method for retrieving operation and maintenance data, characterized in that, include: Upon receiving an operation and maintenance service request, identify the business scenario corresponding to the request; The pre-trained weight model is invoked to calculate the weight values ​​of entities and / or relationships in the knowledge graph under the business scenario, based on the business scenario, entity attributes involved in the operation and maintenance business request, and real-time parameters. Within the knowledge graph, the weight values ​​are used to perform targeted retrieval of the operation and maintenance service requests; the retrieval results are sorted based on the weight values, and related information that meets the expected matching degree with the service scenario is returned; The weight value model is trained based on historical operation and maintenance data and weight value samples; the knowledge graph is constructed based on heterogeneous operation and maintenance data sources.

2. The method according to claim 1, characterized in that, The knowledge graph contains scene nodes, which are associated with entities and / or relationships.

3. The method according to claim 1, characterized in that, The knowledge graph was obtained using the following method: Data is extracted from heterogeneous data sources in operations and maintenance, cleaned, and transformed to establish a metadata model; Extract the operations and maintenance ontology from the unstructured text in the heterogeneous operations and maintenance data source; Define the entities, relationships, and attributes of the operation and maintenance ontology; The flow relationships in the unstructured text are analyzed using natural language processing techniques, and the missing relationships between the entities are filled in. A knowledge graph is built by storing the entities, relationships, and attributes in a graph database.

4. The method according to claim 3, characterized in that, The step of extracting the operations and maintenance ontology from the unstructured text in the heterogeneous operations and maintenance data source includes: The unstructured text is input into a pre-trained ontology extraction model to extract the operation and maintenance ontology; The ontology extraction model is trained using the following method: Convert each word in unstructured text into a fixed-dimensional word unit; After adding tags to the lexical units, they are input into a multi-layer encoder to generate context-related vector representations of the lexical units; The context-related vector representation is input into a fully connected layer to predict the label score corresponding to the word; Calculate the score of possible tag sequences based on the tag transitions of the lexical units; Maximize the probability of the correct label sequence.

5. The method according to claim 3, characterized in that, The missing relationships between entities to be filled include: First, use a pre-built domain knowledge rule base to fill in the missing relationships; For missing relationships that the domain knowledge rule base cannot complete, a machine learning model is used to predict the missing relationships; The machine learning model is obtained by learning the feature patterns of existing entities and / or relationships.

6. The method according to claim 1, characterized in that, The targeted retrieval of the operation and maintenance service requests includes: The operation and maintenance service request is input into a pre-trained node embedding model to generate a node embedding vector, and targeted retrieval is performed based on the node embedding vector; The node embedding model is trained using the following method: Based on the operation and maintenance business request, the target node is determined, and a fixed number of neighboring nodes of the target node are randomly sampled at each layer of the knowledge graph to form a hierarchical sampling structure. The features of the neighboring nodes sampled at each layer are aggregated to generate aggregated features; The features of the target node at each layer are concatenated with the aggregated features; The node embedding vector of the target node is obtained through multi-level aggregation.

7. A knowledge graph-based operation and maintenance data retrieval device, characterized in that, The device includes: The weight calculation module is used to respond to a received operation and maintenance service request, identify the business scenario corresponding to the request, call a pre-trained weight model, and calculate the weight values ​​of entities and / or relationships in the knowledge graph under the business scenario based on the business scenario, entity attributes involved in the operation and maintenance service request, and real-time parameters; and The data retrieval module is used to perform targeted retrieval of the operation and maintenance business requests within the knowledge graph using the weight values; sort the retrieval results based on the weight values; and return the associated information that meets the expected matching degree with the business scenario. The weight value model is trained based on historical operation and maintenance data and weight value samples; the knowledge graph is constructed based on heterogeneous operation and maintenance data sources.

8. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 6.

Images