Method, system, device and storage medium for constructing vertical domain knowledge graph

Abstract

The present invention provides a method, system, device and storage medium for constructing a vertical domain knowledge graph, the method comprising: obtaining business data in a unified data model of a power grid; sorting and classifying business rules in the business data; describing relevant business rules in the form of functions to obtain business rule functions; embedding business rule functions into a neural network, and fusing them with a decision tree to form a text data stream classification model; arranging business rules to form rule parameters, forming an activation function to replace the activation function in the original extraction algorithm of entities and relationships, and obtaining an entity relationship extraction algorithm of fusion rules; based on the entity relationship extraction algorithm of fusion rules, extracting entity relationships from a unified data model, and constructing a knowledge graph based on the entity relationship extraction results. By linking common entities and related businesses in different businesses, ensuring information interconnection between different business graphs, it is helpful to improve the construction and application effects of vertical domain knowledge graphs.

Description

Technical Field

[0001] The present invention belongs to the technical field of vertical domain knowledge graph construction and application, and in particular relates to a vertical domain knowledge graph construction method, system, device and storage medium. Background Art

[0002] The unified data model is an enterprise data model built based on object-oriented modeling technology. It is a data physical model built by combining the core business needs of the corresponding enterprise with the "business demand driven top-down" and "current situation driven bottom-up" modes, which includes data from multiple sub-business areas. Although the unified data model collects data in a unified manner, the correlation between data is weak, making it difficult to conduct a through analysis of data from multiple business areas. In recent years, knowledge graphs, as a management and application technology for knowledge data, have broken through the link barriers between business data through the entity-relationship-attribute triple form, providing an important method for knowledge management of unified data models. Therefore, the unified data model is transformed into knowledge application by combining knowledge graph technology, so as to establish a unified data service with semantic interoperability, realize efficient and intelligent query of data, and ensure cross-business data penetration, which is conducive to the efficient application and intelligent analysis of text data in the unified data model.

[0003] Currently, when constructing knowledge graphs for business data in a unified information model, people often start from the algorithm technology itself and pay less attention to business rules. Therefore, in the process of knowledge graph construction, there are problems such as complicated knowledge systems and difficulty in entity relationship extraction results to meet business rules. Summary of the invention

[0004] In order to improve the current problems in knowledge graph construction, the present invention provides a method, system, device and storage medium for constructing a vertical domain knowledge graph. The method improves the domain knowledge graph construction technology based on a unified data model by designing a text data classification and entity relationship extraction method that introduces business rules, thereby enhancing the coupling between the knowledge graph and the business in the unified data model.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for constructing a vertical domain knowledge graph includes the following steps:

[0007] Obtain business data in the unified data model of the power grid;

[0008] Sort out the business rules in the business data, express the business rules with functions to obtain business rule mapping functions;

[0009] The rule mapping function is embedded in the neural network and fused with the decision tree to form a text data stream classification model, which can divert text data into different business types; entities and relationships are extracted from the text data classified by the text data stream classification model to obtain an entity relationship extraction algorithm;

[0010] Combined with the business rule mapping function, the rule coefficient is obtained, and the rule coefficient is introduced into the activation function of the entity relationship extraction algorithm to form an entity and relationship extraction algorithm that integrates business rules.

[0011] The entity relationship extraction algorithm based on fusion rules extracts entity relationships from the unified data model of the power grid, and constructs a knowledge graph based on the entity relationship extraction results.

[0012] As a further improvement of the present invention, the combing of the business rules in the business data specifically includes:

[0013] Analyze the rules and constraints of different business text data in the unified data model of the power grid, sort out the rules that need to be followed in building the corpus relationship network, and list the rule sets;

[0014] The rule set is sorted out and divided into mechanism rules, binding rules and attribute rules. Among them, mechanism rules are often connected with the line and plant name after the location; binding rules are expressions in the text with nouns + numbers; attribute rules are related words in the field of power transmission in the text;

[0015] Count different types of rules, integrate and describe the business rules of text data.

[0016] As a further improvement of the present invention, the business rule mapping function obtained by expressing the business rule with a function specifically includes:

[0017] Convert the mechanisms and conditional constraints in business rules into functional forms;

[0018] Among them, the mechanism rule: different element types in the text are x, y, z, and the mechanism rules satisfied by them are expressed by the functional relationship:

[0019] n(x)=a*f(x)+b*g(y) (2)

[0020] Among them, n(x) represents the numerical representation of the adjacent elements after the location, f(x) and g(y) represent the numerical representation of the line and the plant station respectively, and a and b represent the probability coefficients of the connection with the line and the plant station after x respectively;

[0021] Constraint rules: If there are different element types x and y in the text, the constraints are expressed as:

[0022] R2: f(x)∈{y|a<y<b} (3)

[0023] Among them, f(x) is the numerical calculation method of the correlation between x and y, and the relationship between different elements is constrained by constraining its numerical range with y;

[0024] Attribute rule: A single or multiple elements in the text are contained in a certain range, and its constraints are expressed as:

[0025] R3: x, y, z∈{f(a), g(a)} (4)

[0026] Among them, f(a), g(a) represent the range boundaries defined by element a in different function forms;

[0027] Different types of rules are represented functionally, and then a rule set is established:

[0028] ∑R: {R1, R2,...,Rn} (5)

[0029] Among them, ∑R represents a set of rule functions of a certain type, and R1, R2, ..., Rn represent rule functions of the same type.

[0030] As a further improvement of the present invention, the rule mapping function is embedded in the neural network and fused with the decision tree to form a text data stream classification model, specifically:

[0031] Based on the integration of the rule function set, a new function is constructed as the activation function of the neural network neuron. The Relu activation function is expressed as:

[0032] Max(0,∑(R i (x)*a i )) (6)

[0033] Among them, R i is a rule function in a certain class of rules, a i is the rule function weight, x is the input text function representation variable;

[0034] The activation function integrated into the rule is obtained through the above formula, and neural networks for classifying different businesses are constructed respectively. They are combined with decision trees to form different neural networks as nodes of decision trees. The input data stream is classified layer by layer to obtain a text data stream classification model.

[0035] As a further improvement of the present invention, the specific method of the entity relationship extraction algorithm for obtaining the fusion rule is:

[0036] Combined with the business rule mapping function, sort out the rule mechanism coefficient, rule constraint coefficient, and sample boundary coefficient;

[0037] The business rule mechanism coefficient is to form rule parameters by arranging business rules and calculate the rule coefficient:

[0038]

[0039] Among them, α i is the rule mechanism coefficient, is the fitting value of the rule mechanism function, n(x) is the actual value, is the average value;

[0040] The specific calculation method of the business rule constraint coefficient is:

[0041] β i =∑(f(x) i *k) (8)

[0042] Among them, β i is the rule constraint coefficient, f(x) i Calculate the value of the constraint function for an element, k is 1 if it satisfies the constraint condition, otherwise it is zero;

[0043] The specific calculation method of the sample boundary coefficient is:

[0044] γ i =∑(∏f k (x j )) (9)

[0045] Among them, γ i is the sample boundary coefficient, x j For a single element in the text, f k (x j ) is the range calculation result of a certain element;

[0046] The rule mechanism coefficient, rule constraint coefficient, and sample boundary coefficient are introduced into the activation function of the long short-term memory network to replace the activation function in the entity relationship extraction algorithm, and the entity relationship extraction algorithm with fusion rules is obtained:

[0047]

[0048] Among them, σ r (z) is the activation function integrated into the rule, α i is the rule mechanism coefficient, β i is the rule constraint coefficient, γ i is the sample boundary coefficient.

[0049] As a further improvement of the present invention, the construction of the knowledge graph based on the entity relationship extraction results is specifically as follows:

[0050] After extracting entities and relationships from the text data stream of the unified data model, we obtain entity relationships similar to the rule types, and then obtain entity-relationship combination units, including mechanistic relationships, constraint relationships, and dependency relationships. Different entity nodes in the graph are connected through multiple relationships to form a knowledge graph containing multiple entity-relationship combination units.

[0051] As a further improvement of the present invention, the knowledge graph constructed based on the unified data model is divided into blocks based on sub-businesses, and association relationships are established between sub-businesses based on common entities or related businesses.

[0052] A vertical domain knowledge graph construction system, comprising:

[0053] An acquisition unit, used for acquiring business data in a unified data model of a power grid;

[0054] A business rule sorting unit is used to sort out the business rules in the business data and express the business rules with functions to obtain business rule mapping functions;

[0055] The entity relationship establishment unit is used to embed the rule mapping function into the neural network and fuse it with the decision tree to form a text data stream classification model to divert the text data into different business types; extract entities and relationships from the text data classified by the text data stream classification model to obtain an entity relationship extraction algorithm;

[0056] The entity-relationship fusion unit is used to combine the business rule mapping function to obtain the rule coefficient, introduce the rule coefficient into the activation function of the entity-relationship extraction algorithm, and form an entity-relationship extraction algorithm that integrates business rules.

[0057] The knowledge graph construction unit extracts entity relationships from the unified data model of the power grid based on the entity relationship extraction algorithm with fusion rules, and constructs a knowledge graph based on the entity relationship extraction results.

[0058] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for constructing a vertical field knowledge graph when executing the computer program.

[0059] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for constructing a vertical field knowledge graph.

[0060] The beneficial effects of the present invention are embodied in:

[0061] The present invention effectively integrates and extracts the rules in the text data in the unified information model, and proposes a functional representation method, so that business rules can be effectively integrated into the algorithm. The combined model of neural network and decision tree based on rule function improvement improves the limitation that the text data flow in the unified information model is difficult to automatically divide according to business rules. By integrating the entity and relationship extraction algorithm of business rules, the consistency between the knowledge graph entity relationship and business rules is improved. By linking common entities and related businesses in different businesses, ensuring the information interconnection between different business graphs, it helps to improve the construction and application effects of vertical field knowledge graphs. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a schematic diagram of the process of constructing a vertical field knowledge graph according to a preferred embodiment of the present invention;

[0063] Figure 2 Extracting business rules from the unified data model of the power grid for step one;

[0064] Figure 3 Describe the business rules in step 2 in functional form;

[0065] Figure 4 Embed the rules of step 3 into the neural network activation function and combine them with the decision tree to form data stream classification;

[0066] Figure 5 A classification model for text data streams;

[0067] Figure 6 Arrange the rule parameters for step 4 and propose entity relationship extraction with fusion rules;

[0068] Figure 7 For step 5, build a knowledge graph based on the entity relationship extraction algorithm integrated with rules;

[0069] Figure 8 This is an example of a knowledge graph that contains multiple types of relationships;

[0070] Fig. 9 This is an example of the association between sub-business graphs of the knowledge graph based on a unified data model;

[0071] Fig.10 A schematic diagram of the structure of a vertical domain knowledge graph construction system in a preferred embodiment of the present invention;

[0072] Fig.11 It is a schematic diagram of the structure of an electronic device according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0073] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0074] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention.

[0075] The present invention belongs to the construction and application technology of vertical field knowledge graph, such as Figure 1 As shown, the present invention provides a method for constructing a vertical field knowledge graph based on a unified data model.

[0076] First, according to the business data in the unified data model, the rules contained in it are sorted and classified, such as the unified data model of the power grid.

[0077] Secondly, based on the rule logic, design the functional representation of the rules.

[0078] The rule mapping function is again embedded in the activation function of the neural network model and combined with the decision tree to form a text data stream diversion model to divert the text into different business types.

[0079] Finally, an activation function that introduces rule coefficients is designed to form an entity and relationship extraction algorithm that integrates business rules, and a knowledge graph is constructed based on the entity relationship extraction results.

[0080] The knowledge graph is constructed mainly through the following five steps.

[0081] This paper proposes a vertical field knowledge graph construction method based on a unified data model, which will improve the knowledge application technology level and application effect of the unified data model in various industries from three aspects:

[0082] First, based on the text data in the unified data model, business rules are sorted and extracted, and then expressed in a functional manner, which solves the problem of difficulty in analyzing business rules through mathematical means.

[0083] Secondly, by using neural networks as decision tree nodes, a business segmentation model for the same data model is constructed, which improves the efficiency of vertical data business segmentation and enables a deep integration of the original text corpus with business characteristics.

[0084] Third, business rules are introduced into the construction of the knowledge graph to form an entity relationship extraction algorithm that integrates the rules, so that the constructed knowledge graph is more in line with the business specifications of the vertical field and better supports the application of vertical field knowledge.

[0085] The method of the present invention specifically comprises the following steps:

[0086] Obtain business data in the unified data model of the power grid;

[0087] Sort out the business rules in the business data, express the business rules with functions to obtain business rule mapping functions;

[0088] The rule mapping function is embedded in the neural network and fused with the decision tree to form a text data stream classification model, which can divert text data into different business types; entities and relationships are extracted from the text data classified by the text data stream classification model to obtain an entity relationship extraction algorithm;

[0089] Combined with the business rule mapping function, the rule coefficient is obtained, and the rule coefficient is introduced into the activation function of the entity relationship extraction algorithm to form an entity and relationship extraction algorithm that integrates business rules.

[0090] The entity relationship extraction algorithm based on fusion rules extracts entity relationships from the unified data model of the power grid, and constructs a knowledge graph based on the entity relationship extraction results.

[0091] Taking the electric power field as an example, the present invention achieves the above-mentioned purpose through the following technical solutions:

[0092] The knowledge graph is constructed mainly through the following five steps.

[0093] Step 1: Sort out the business rules for extracting text data from the unified data model of the power grid.

[0094] Figure 2 Step 1: Extract business rules from the unified data model of the power grid

[0095] (1) Analyze the rules and constraints of text data in different businesses, sort out the rules that need to be followed in building the corpus relationship network, and list the rule sets. For example:

[0096] Rule 1: The location is often connected with the line, plant and station name;

[0097] Rule 2: The text is mainly expressed in terms of nouns + numbers;

[0098] Rule 3: The relevant words such as "pole tower", "steel pipe pole", "angle steel tower" in the text belong to the field of power transmission.

[0099] The rule set is sorted out, and the rules are divided into mechanism rules, constraint rules, and attribute rules. Mechanism rules refer to the rules that determine the association and driving methods between different elements, such as the location is often connected with the line and plant name; constraint rules determine the existence of elements, such as the text is mainly expressed in the form of nouns + numbers; attribute rules determine the subordinate scope of elements, such as the relevant words such as "pole tower", "steel pipe pole", "angle steel tower" in the text belong to the field of power transmission, etc. Different types of rules are counted, and their main conditions are integrated and described.

[0100] Step 2: Organize the mechanisms and conditional constraints in the business rules and describe the relevant rules in the form of functions.

[0101] Figure 3 The business rules of step 2 are described in functional form.

[0102] The mechanisms and conditional constraints in business rules are converted into functional forms, making it easier to integrate the rules with machine learning algorithms. Taking mechanism rules, constraint rules, and attribute rules as examples, they are described in functional form as follows:

[0103] Mechanism rules: There are different types of elements in the text, x, y, and z. The mechanism rules that they satisfy can be expressed by functional relationships:

[0104] R1: F(x, y, z) = a (1)

[0105] Among them, R1 represents rule 1, and function F stipulates the mathematical relationship between elements x, y, and z. The relationship between them can be determined according to the specific situation. For the example of "location is often connected with line, plant and station name", its function form can be expressed as:

[0106] n(x)=a*f(x)+b*g(y) (2)

[0107] Among them, n(x) represents the numerical representation of the adjacent elements after the location, f(x) and g(y) represent the numerical representations represented by the line and the plant respectively, and a and b represent the probability coefficients of the connection with the line and the plant respectively after x.

[0108] Constraint rules: Different elements in the text need to meet certain constraints. For example, if there are different element types x and y in the text, the constraints can be expressed as:

[0109] R2: f(x)∈{y|a<y<b} (3)

[0110] Among them, f(x) is the relevant numerical calculation method of x and y. By constraining its numerical range with y, the relationship between different elements is constrained.

[0111] Attribute rule: A single or multiple elements in a text are contained in a certain range, and its constraints can be expressed as:

[0112] R3: x, y, z (4)

[0113] ∈{f(a),g(a)}

[0114] Among them, f(a) and g(a) represent the range boundaries defined by element a in the form of different functions.

[0115] As mentioned above, different types of rules can be represented functionally to establish a rule set:

[0116] ∑R: {R1, R2,...,Rn} (5)

[0117] Step 3: Embed the rules into the neural network and fuse them with the decision tree to form a text data stream classification model. Figure 4 Step 3: The rules are embedded in the neural network activation function and combined with the decision tree to form data stream classification.

[0118] Based on the integration of rule function sets, a new function is constructed as the activation function of the neural network neurons, and the network is ensured to meet the training requirements of the gradient descent method, so that the neural network can classify and process the data stream according to the business rules. Taking the Relu type activation function as an example, when a certain type of business rule function is embedded in the activation function, the Relu activation function can be expressed as follows:

[0119] Max(0,∑(R i (x)*a i )) (6)

[0120] Among them, R i is a rule function in a certain class of rules, a i is the rule function weight, and x is the input text function representation variable.

[0121] The above formula is used to derive the activation function that incorporates the rules. For different business types, the number and types of rules included are also different. We can build neural networks for classifying different businesses separately and combine them with decision trees to form different neural networks as nodes of the decision tree. We can classify the input data stream layer by layer and finally obtain a classification model for the business type described by the data stream.

[0122] Taking the text classification in the electric power field as an example, the text classification process in this step is explained.

[0123] Figure 5 A classification model for text data streams.

[0124] Figure 5In the example, the text field "power transmission line fault tripping" is vectorized and input into the neural network + decision tree classification model in step 3. From top to bottom, the decision tree nodes are passed through for business division. If the business classification is not obtained at the current level node, the next level node is entered for classification. In this way, the text classification is finally obtained. In this example, "power transmission line fault tripping" belongs to the power transmission business, so the classification result can be obtained in the neural network NN1 of the first level node.

[0125] Step 4: Organize business rules to form rule parameters, form activation functions to replace the activation functions in the original entity and relationship extraction algorithms, and propose an entity relationship extraction algorithm that integrates rules.

[0126] Figure 6 Step 4: Arrange rule parameters and propose entity relationship extraction based on fusion rules

[0127] Entities and relationships are extracted from the classified text data in a specific business field. Combined with the business rule functions in different fields sorted out in step 1, parameters such as rule mechanism coefficients, rule constraint coefficients, and sample boundary coefficients are sorted out to improve the activation function of neurons in the long-short memory network. The improvement process is as follows:

[0128] Rule mechanism coefficient:

[0129]

[0130] Among them, α i is the rule mechanism coefficient, is the fitting value of the rule mechanism function, n(x) is the actual value, It is the average value, which can reflect the degree of consistency between business data and rules.

[0131] Rule constraint coefficient:

[0132] β i =∑(f(x) i *k) (8)

[0133] Among them, β i is the rule constraint coefficient, f(x) i Calculates the value of the constraint function for an element, k is 1 if it satisfies the constraint, otherwise it is zero.

[0134] Sample boundary coefficient:

[0135] γ i =∑(∏f k (x j )) (9)

[0136] Among them, γ i is the sample boundary coefficient, x jFor a single element in the text, f k (x j ) is the result of calculating the range of an element.

[0137] After obtaining the rule mechanism coefficient, rule constraint coefficient, and sample boundary coefficient, they are introduced into the activation function of the long short-term memory network. The improved activation function is as follows:

[0138]

[0139] Among them, σ r (z) is the activation function integrated into the rule, α i is the rule mechanism coefficient, β i is the rule constraint coefficient, γ i is the sample boundary coefficient.

[0140] Rule coefficients are introduced into the forget gate, input gate and output gate of the long short-term memory network to effectively combine the algorithm with the rules.

[0141] Step 5: Based on the entity relationship extraction algorithm of fusion rules, entities and relationships are extracted from the unified data model, and then a knowledge graph is constructed.

[0142] Figure 7 Step 5: Build a knowledge graph based on the rule-based entity relationship extraction algorithm

[0143] After extracting entities and relationships from the unified data model text data stream, similar to the types of rules, there are also many types of relationships between entities, such as mechanism relationships, constraint relationships, and attribute relationships. In turn, entity-relationships also have multiple combination units. Different entity nodes in the graph are connected through multiple relationships to form a knowledge graph containing multiple entity-relationship combination units. Taking the power field as an example, the knowledge graph sample containing mechanism relationships, constraint relationships, and attribute relationships is as follows:

[0144] Figure 8 This is an example of a knowledge graph containing multiple types of relationships.

[0145] For a unified data model in a certain field, it often contains data within the scope of multiple sub-businesses. Therefore, the knowledge graph built based on the unified data model is often divided into blocks based on sub-businesses. At the same time, sub-businesses establish associations based on shared entities or related businesses. Therefore, the knowledge graph built based on the unified data model is presented in different sub-businesses in the following form:

[0146] Fig. 9 This is an example of the association between sub-business graphs of a knowledge graph based on a unified data model.

[0147] Therefore, when constructing a knowledge graph based on a unified data model, the present invention, on the one hand, extracts entities and relationships based on the data of each sub-business, and then saves the sub-business knowledge data through the knowledge graph; on the other hand, it links the common entities and association relationships between the knowledge graphs of each sub-business to achieve the interconnection of knowledge graph data in each business.

[0148] At this point, through the processing of steps one to five, a vertical domain knowledge graph construction technology based on a unified data model is formed.

[0149] The business rules of the data in this article in the unified data model are sorted out and represented functionally according to the inherent logic of the rules. The rule function is introduced into the neuron activation function, and the text data stream in the statistical data model is classified according to the business rule logic. The business rule function is sorted to form rule parameters, and the activation function is constructed to replace the activation function in the original extraction algorithm of entities and relationships, so that the knowledge graph construction is more in line with business rules and the effect of knowledge application in the unified data model in vertical fields is improved.

[0150] The advantages of the present invention are:

[0151] First, a functional representation method for text data business rules in a unified data model is proposed. The business rules of text data in the unified model are sorted out and integrated, and different types of rules such as mechanism rules, constraint rules, and attribute rules are extracted. Then, the mechanism and condition constraints in the business rules are analyzed, and the relevant rules are described in the form of functions;

[0152] Second, design a fusion model of neural network and decision tree to realize the classification of text data stream in the unified data model. Based on the business rules expressed in functional form, organize and form a new activation function of the neural network, and design a decision tree model with neural network as nodes to realize the classification of business text data;

[0153] Third, design a knowledge graph construction technology that integrates unified data model business rules. Based on the business rules expressed in a functional way, further organize and form parameters such as rule mechanism coefficients, rule constraint coefficients, and sample boundary coefficients, improve the activation function of the long and short-term memory network neurons used for entity and relationship extraction, and form a knowledge graph construction technology in the power field that integrates business rules.

[0154] like Fig.10 As shown, another object of the present invention is to propose a vertical field knowledge graph construction system, comprising:

[0155] An acquisition unit, used for acquiring business data in a unified data model of a power grid;

[0156] A business rule sorting unit is used to sort out the business rules in the business data and express the business rules with functions to obtain business rule mapping functions;

[0157] The entity relationship establishment unit is used to embed the rule mapping function into the neural network and fuse it with the decision tree to form a text data stream classification model to divert the text data into different business types; extract entities and relationships from the text data classified by the text data stream classification model to obtain an entity relationship extraction algorithm;

[0158] The entity-relationship fusion unit is used to combine the business rule mapping function to obtain the rule coefficient, introduce the rule coefficient into the activation function of the entity-relationship extraction algorithm, and form an entity-relationship extraction algorithm that integrates business rules.

[0159] The knowledge graph construction unit extracts entity relationships from the unified data model of the power grid based on the entity relationship extraction algorithm with fusion rules, and constructs a knowledge graph based on the entity relationship extraction results.

[0160] The business rule arrangement unit specifically includes:

[0161] The rule set arrangement unit is used to analyze the rules and constraints of different business text data in the unified data model of the power grid, sort out the rules that need to be followed in building the corpus relationship network, and list the rule sets;

[0162] The rule set classification unit is used to sort out the rule set and divide the rules into mechanism rules, constraint rules, and attribute rules; among them, the mechanism rules are often connected with the line and plant name after the location; the constraint rules are the expressions of nouns + numbers in the text; the attribute rules are the words related to the field of power transmission in the text;

[0163] The business rule integration unit is used to count different types of rules and integrate and describe the business rules of the text data.

[0164] like Fig.11 As shown, the third object of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the vertical field knowledge graph construction method when executing the computer program.

[0165] The method comprises the following steps:

[0166] Obtain business data in the unified data model of the power grid;

[0167] Sort out and classify the business rules in the business data; organize the mechanisms and conditional constraints in the business rules, and describe the relevant business rules in the form of functions to obtain business rule functions;

[0168] The business rule function is embedded in the neural network and fused with the decision tree to form a text data stream classification model; the business rules are sorted to form rule parameters, and the activation function is formed to replace the activation function in the original entity and relationship extraction algorithm to obtain an entity relationship extraction algorithm with fusion rules;

[0169] Based on the fusion rule-based entity relationship extraction algorithm, entity relationships are extracted from the unified data model, and a knowledge graph is constructed based on the entity relationship extraction results.

[0170] The fourth object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the steps of the vertical field knowledge graph construction method.

[0171] The method comprises the following steps:

[0172] Obtain business data in the unified data model of the power grid;

[0173] Sort out and classify the business rules in the business data; organize the mechanisms and conditional constraints in the business rules, and describe the relevant business rules in the form of functions to obtain business rule functions;

[0174] The business rule function is embedded in the neural network and fused with the decision tree to form a text data stream classification model; the business rules are sorted to form rule parameters, and the activation function is formed to replace the activation function in the original entity and relationship extraction algorithm to obtain an entity relationship extraction algorithm with fusion rules;

[0175] Based on the fusion rule-based entity relationship extraction algorithm, entity relationships are extracted from the unified data model, and a knowledge graph is constructed based on the entity relationship extraction results.

[0176] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0177] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0178] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0179] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a vertical domain knowledge graph, characterized in that: The following steps are involved: Obtain business data in the unified data model of the power grid; Sort out the business rules in the business data, express the business rules with functions to obtain business rule mapping functions; The rule mapping function is embedded in the neural network and fused with the decision tree to form a text data stream classification model, which can divert text data into different business types; entities and relationships are extracted from the text data classified by the text data stream classification model to obtain an entity relationship extraction algorithm; Combined with the business rule mapping function, the rule coefficient is obtained, and the rule coefficient is introduced into the activation function of the entity relationship extraction algorithm to form an entity relationship extraction algorithm that integrates business rules; The entity relationship extraction algorithm based on the fusion of business rules is used to extract entity relationships from the unified data model of the power grid, and a knowledge graph is constructed based on the entity relationship extraction results; The combing of the business rules in the business data specifically includes: Analyze the rules and constraints of different business text data in the unified data model of the power grid, sort out the rules that need to be followed in building the corpus relationship network, and list the rule sets; The rule set is sorted out and divided into mechanism rules, binding rules and attribute rules. Among them, the mechanism rules are connected with the line and plant name after the location; the binding rules are the expressions of nouns + numbers in the text; the attribute rules are the words related to the field of power transmission in the text; Count different types of rules, integrate and describe the business rules of text data; The business rule mapping function obtained by expressing the business rule with a function specifically includes: Convert the mechanisms and conditional constraints in business rules into functional forms; Among them, the mechanism rules: different element types in the text have , then the mechanism rules satisfied between them are expressed by the functional relationship: in, The numerical representation of the adjacent elements after the location, f(x), g(y) represent the numerical representation of the line and plant station respectively. They represent the probability coefficients of connection with the line and plant values ​​after x respectively; Binding rules: Different element types in the text have , then its constraints are expressed as: in, It is a method for calculating the related values ​​of x and y, and constrains the relationship between different elements by constraining the value range of x and y; Attribute rule: A single or multiple elements in the text are contained in a certain range, and its constraints are expressed as: in, Representation elements Range boundaries defined in different functional forms; Different types of rules are represented functionally, and then a rule set is established: Among them, ∑R represents a set of rule functions of a certain type, and R1, R2, ..., Rn represent rule functions of the same type.

2. The method according to claim 1, characterized in that: The rule mapping function is embedded in the neural network and integrated with the decision tree to form a text data stream classification model, specifically: Based on the integration of the rule function set, a new function is constructed as the activation function of the neural network neuron. The Relu activation function is expressed as: in, R i is a rule function in a class of rules, a i is the rule function weight, x is the input text function representation variable; The activation function integrated into the rule is obtained through the above formula, and neural networks for classifying different businesses are constructed respectively. They are combined with decision trees to form different neural networks as nodes of decision trees. The input data stream is classified layer by layer to obtain a text data stream classification model.

3. The method according to claim 1, characterized in that: The specific method of the entity relationship extraction algorithm for obtaining the fusion rule is as follows: Combined with the business rule mapping function, sort out the rule mechanism coefficient, rule constraint coefficient, and sample boundary coefficient; The business rule mechanism coefficient is to form rule parameters by arranging business rules and calculate the rule coefficient: in, is the rule mechanism coefficient, is the fitting value of the rule mechanism function, is the actual value, is the average value; The specific calculation method of the business rule constraint coefficient is: in, is the rule constraint coefficient, f(x) i Calculate the value of the constraint function for an element, k is 1 if it satisfies the constraint condition, otherwise it is zero; The specific calculation method of the sample boundary coefficient is: in, is the sample boundary coefficient, x j For a single element in the text, f k (x j ) is the range calculation result of a certain element; The rule mechanism coefficient, rule constraint coefficient, and sample boundary coefficient are introduced into the activation function of the long short-term memory network to replace the activation function in the entity relationship extraction algorithm, and the entity relationship extraction algorithm with fusion rules is obtained: in, is the activation function that incorporates the rule, is the rule mechanism coefficient, is the rule constraint coefficient, is the sample boundary coefficient.

4. The method according to claim 1, characterized in that: The construction of the knowledge graph based on the entity relationship extraction results is specifically as follows: After extracting entities and relationships from the text data stream of the unified data model, we obtain entity relationships similar to the rule types, and then obtain entity-relationship combination units, including mechanistic relationships, constraint relationships, and dependency relationships. Different entity nodes in the graph are connected through multiple relationships to form a knowledge graph containing multiple entity-relationship combination units.

5. The method according to claim 4, characterized in that: The knowledge graph built based on a unified data model is divided into blocks based on sub-businesses, and relationships are established between sub-businesses based on shared entities or related businesses.

6. A vertical domain knowledge graph construction system, implementing the vertical domain knowledge graph construction method according to any one of claims 1 to 5, characterized in that: include: An acquisition unit, used for acquiring business data in a unified data model of a power grid; A business rule sorting unit is used to sort out the business rules in the business data and express the business rules with functions to obtain business rule mapping functions; The entity relationship establishment unit is used to embed the rule mapping function into the neural network and fuse it with the decision tree to form a text data stream classification model to divert the text data into different business types; extract entities and relationships from the text data classified by the text data stream classification model to obtain an entity relationship extraction algorithm; The entity relationship fusion unit is used to combine the business rule mapping function to obtain the rule coefficient, introduce the rule coefficient into the activation function of the entity relationship extraction algorithm, and form an entity relationship extraction algorithm that integrates business rules. The knowledge graph construction unit extracts entity relationships from the unified data model of the power grid based on the entity relationship extraction algorithm that integrates business rules, and constructs a knowledge graph based on the entity relationship extraction results.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the vertical field knowledge graph construction method as described in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing a vertical field knowledge graph as described in any one of claims 1 to 5 is implemented.

Images