Knowledge graph construction method and system of cable structured process library

Abstract

The invention discloses a knowledge graph construction method and system for a cable structured process library, and the method comprises the following steps: carrying out the knowledge extraction of the cable structured process library, and carrying out the entity recognition of an extracted text, and obtaining a knowledge text; performing relation type analysis on the knowledge text to obtain a relation type of the cable structured entity; performing cable structured knowledge subject-object concept distinguishing on discrete data in the knowledge text, and encoding the distinguished discrete data to obtain encoded text data; performing text preprocessing on the encoded text data to obtain processed text data; performing ternary modeling on the cable structuring process based on the processed text data and the relationship type, and mapping an obtained entity relationship triple into graph data; and establishing nodes and relationships based on the encoded text data and graph data, and constructing a knowledge graph of the cable structuring process library.

Description

Technical Field

[0001] The present invention belongs to the field of cable processing technology and knowledge graph technology, and specifically relates to a method and system for constructing a knowledge graph of a cable structured process library. Background Art

[0002] In the process of cable processing, digital production is a necessary direction and research content. The digital production process can more comprehensively and truly reflect the process and status of the target in the processing process. The structured process library can more intuitively display the process flow of the cable, and facilitate the query, storage and classification of various related information, improve the efficiency of processing, and provide information support for intelligent decision-making.

[0003] With the advent of the intelligent era, the structured process library shows important application value. The structured process library can improve the efficiency of supervision and processing by displaying comprehensive relevant information.

[0004] The cable processing process will see a lot of changes in business requirements, and it is very difficult to extract and improve information by relying on a structured process library. Flexible data models can adapt to changing business needs. While expressing entities and their relationships in the real world, knowledge graphs can easily analyze and adjust complex relationship networks. Summary of the invention

[0005] The present invention aims to solve the deficiencies of the prior art and provides the following solutions:

[0006] A method for constructing a knowledge graph of a cable structured process library comprises the following steps:

[0007] Extract knowledge from the cable structured process library and perform entity recognition on the extracted text to obtain knowledge text;

[0008] Performing relationship type analysis on the knowledge text to obtain the relationship type of the cable structured entity;

[0009] Distinguishing the concepts of cable structured knowledge subject and object for the discrete data in the knowledge text, and encoding the differentiated discrete data to obtain encoded text data;

[0010] Performing text preprocessing on the encoded text data to obtain processed text data;

[0011] Performing triple modeling on the cable structured process based on the processed text data and the relationship type, and mapping the obtained entity relationship triples into graph data;

[0012] Nodes and relationships are established based on the encoded text data and the graph data to construct a knowledge graph of the cable structured process library.

[0013] Preferably, the method for obtaining the knowledge text includes:

[0014] Construct an entity recognition model combining the first BiLSTM layer with the CRF layer;

[0015] Using the first BiLSTM layer to extract knowledge from the cable structured process library to obtain extracted knowledge;

[0016] The CRF layer is used to perform entity recognition on the extracted knowledge to obtain the knowledge text.

[0017] Preferably, the method for obtaining the relationship type of the cable structured entity comprises:

[0018] Constructing a Self-ATT-BiLSTM neural network, wherein the Self-ATT-BiLSTM neural network includes: an input layer, a word vector layer, a second BiLSTM layer, a Self-ATT layer, and an output layer;

[0019] The input layer receives the knowledge text;

[0020] The word vector layer maps each word in the knowledge text into a word vector of fixed length;

[0021] The second BiLSTM layer calculates the word vector to obtain a high-level word vector feature;

[0022] The Self-ATT layer calculates the weight of each word, merges the word vector features of each time step, and multiplies the word vector features and the corresponding weight to obtain a text vector;

[0023] The output layer calculates the text vector to obtain the relationship type between entities.

[0024] Preferably, the method for obtaining the encoded text data includes:

[0025] Preprocessing the discrete data in the knowledge text, and using a named entity recognition model to recognize named entities in the preprocessed text;

[0026] Analyzing the discrete data after named entity recognition using dependency syntactic analysis to determine the relationship between the subject, verb, and object in the discrete data;

[0027] The discrete data is encoded based on the common features of the subject and the tool data of the subject and the object and the attribute data of the object before and after the object is changed, so as to obtain the encoded text data.

[0028] Preferably, the method for obtaining the processed text data comprises:

[0029] Performing a word segmentation operation on the selected text in the encoded text data to obtain word segments, and marking the part of speech of the word segments;

[0030] Based on the word segmentation and the part of speech, selectively identifying entity words with specific meanings in the encoded text data;

[0031] The recognized text is subjected to dependency syntactic analysis to obtain the processed text data.

[0032] Preferably, the method for triple modeling includes: combining dependency syntax principles and Chinese grammar heuristic rules, combining the entity set in the processed text data and the relationship type, to obtain the entity relationship triple.

[0033] Preferably, the method for constructing the knowledge graph of the cable structured process library includes:

[0034] The graph data is stored in Neo4j and nodes and relationships are established to construct the main framework of the cable process knowledge graph;

[0035] The encoded text data is stored in Neo4j, and the feature code is stored in the database along with the entity in the form of attributes to generate nodes;

[0036] The name information and attribute information of the nodes are traversed, duplicate nodes are searched and merged, and the construction of the knowledge graph is completed.

[0037] The present invention also provides a knowledge graph construction system for a cable structured process library, the system applying any of the above methods, including: an extraction and recognition module, a type analysis module, an encoding module, a preprocessing module, a mapping module and a graph construction module;

[0038] The extraction and recognition module is used to extract knowledge from the cable structured process library and perform entity recognition on the extracted text to obtain knowledge text;

[0039] The type analysis module is used to perform relationship type analysis on the knowledge text to obtain the relationship type of the cable structured entity;

[0040] The encoding module is used to distinguish the cable structured knowledge subject and object concepts of the discrete data in the knowledge text, and encode the discrete data after distinction to obtain encoded text data;

[0041] The preprocessing module is used to perform text preprocessing on the encoded text data to obtain processed text data;

[0042] The mapping module performs triple modeling on the cable structured process based on the processed text data and the relationship type, and maps the obtained entity relationship triples into graph data;

[0043] The graph construction module establishes nodes and relationships based on the encoded text data and the graph data to construct a knowledge graph of the cable structured process library.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention obtains the conditional probability distribution of another set of output sequences under the condition of a set of input sequences by performing knowledge extraction and text recognition on the cable structured process library, and can add constraints to the predicted labels finally generated; the relationship type of the cable structured process library is obtained by performing syntactic dependency analysis on the collected knowledge text; the discrete data is distinguished based on the subject and object concepts in the design and processing, and the discrete data is encoded to obtain the standardized encoding of the cable usage information and manufacturing characteristics; the text preprocessing of the process text data of the cable processing is performed, so that the cable structured process library has the effect of selectively identifying entity words with specific meanings in the text; the cable structured process design is triple modeled through the knowledge graph, so that the cable structured process library obtains the ability of mapping the process into graph data; by adding discrete data and triple data to be stored in Neo4j, the knowledge graph of the cable structured process library with highly correlated discrete data and text data content is successfully established. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0047] Figure 1 The figure is a schematic diagram of a method flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Embodiment 1

[0051] In this embodiment, if Figure 1 As shown, a method for constructing a knowledge graph of a cable structured process library includes the following steps:

[0052] S1. Extract knowledge from the cable structured process library and perform entity recognition on the extracted text to obtain knowledge text.

[0053] The method for obtaining the knowledge text includes: constructing an entity recognition model combining the first BiLSTM layer and the CRF layer; using the first BiLSTM layer to extract knowledge from the cable structured process library to obtain extracted knowledge; using the CRF layer to perform entity recognition on the extracted knowledge to obtain the knowledge text.

[0054] In this embodiment, a model combining BiLSTM and CRF is used for entity recognition, and an entity recognition model combining the first BiLSTM layer and the CRF layer is constructed. The first BiLSTM layer combines information in both the forward and backward directions to model the cable structured process library sequence, to help the model better understand the sequence data, thereby performing knowledge extraction. The CRF layer, as a conditional random field, is used to capture the dependencies between labels in the knowledge graph construction task, thereby improving the performance of the model in the sequence labeling task.

[0055] Specifically, the first BiLSTM layer uses bidirectional information capture to simultaneously consider the information before and after the current moment, and fully understands the input sequence. The first BiLSTM layer can process the long-term dependencies in the sequence to avoid gradient vanishing or gradient explosion. The first BiLSTM layer extracts features from the cable structured knowledge data. The CRF layer obtains the optimal label sequence by considering the dependencies between adjacent labels and joint learning, and establishes label dependencies. The CRF layer calculates the probability of the entire label sequence through global normalization, ensuring that the optimal solution sequence is globally optimal. The CRF layer converts the label prediction problem into a sequence-level text entity recognition problem by solving the label bias problem.

[0056] S2. Perform relationship type analysis on the knowledge text to obtain the relationship type of the cable structured entity.

[0057] The method for obtaining the relationship type of the cable structured entity includes: constructing a Self-ATT-BiLSTM neural network, the Self-ATT-BiLSTM neural network includes: an input layer, a word vector layer, a second BiLSTM layer, a Self-ATT layer and an output layer; the input layer receives the knowledge text; the word vector layer maps each word in the knowledge text to a word vector of a fixed length; the second BiLSTM layer calculates the word vector to obtain a high-level word vector feature; the Self-ATT layer calculates the weight of each word, merges the word vector features of each time step, and multiplies the word vector features and the corresponding weights to obtain a text vector; the output layer calculates the text vector to obtain the relationship type between entities.

[0058] In this embodiment, a Self-ATT-BiLSTM neural network is constructed and used to extract the relationship between entities. The Self-ATT model is an improved version of the attention mechanism model, which integrates the word vector features of each time step and fully considers the semantic and grammatical connections between different words in the sentence, thereby generating a feature vector that can better reflect the context connection. The feature matrix E∈R output by the second BiLSTM layer s*2u It is used as the input of the self-attention mechanism. Where s is the length of the input sequence and u is the size of the second BiLSTM hidden layer. The self-attention weight vector a∈R s , and then the entity relationship probability distribution p is obtained.

[0059] a=Softmax(v att [tanh(E)]J)

[0060] p = Softmax[WLtanh(aE)] T+ b T e

[0061] Among them, v att ∈R 2u , W∈R c*2u , b∈R cis a trainable parameter of the self-attention layer; J is a transformation matrix for processing nonlinear input; L is a learnable parameter matrix; e is the relevant embedding vector of the input; c is the number of relationship types that need to be output, and there are 7 relationship types in this embodiment, that is, c=7; tanh(·) is a hyperbolic tangent activation function. Specifically, the Self-ATT-BiLSTM neural network includes: an input layer, a word vector layer, a second BiLSTM layer, a Self-ATT layer, and an output layer; the input layer receives the knowledge text; the word vector layer maps each word in the knowledge text to a word vector of a fixed length; the second BiLSTM layer calculates the word vector to obtain a high-level word vector feature; the Self-ATT layer calculates the weight of each word, merges the word vector features of each time step, and multiplies the word vector features and the corresponding weights to obtain a text vector; the output layer calculates the text vector to obtain the relationship type between entities.

[0062] S3. Distinguish the concepts of subject and object of cable-structured knowledge for discrete data in the knowledge text, and encode the distinguished discrete data to obtain encoded text data.

[0063] The method for obtaining encoded text data includes: preprocessing discrete data in knowledge text, and using a named entity recognition model to recognize named entities in the preprocessed text; using dependency syntactic analysis to analyze the discrete data after named entity recognition, and determine the relationship between the subject, verb and object in the discrete data; encoding the discrete data based on the common characteristics of the tool data of the subject and its implementation behavior and the attribute book data of the object and its change sequence, to obtain encoded text data.

[0064] In this embodiment, the text is firstly subjected to preprocessing operations such as word segmentation and stop word removal for subsequent analysis; the named entities (such as cable model, cable parameters, manufacturers, etc.) in the cable structured text are identified using a named entity recognition (NER) model or tool; dependency syntactic analysis is performed to capture the grammatical relationship between different words, especially the subject-verb-object relationship, and the relationship between the subject (subject), verb and object (object) in the corresponding cable knowledge sentence is determined according to the dependency syntactic analysis results. The identity of the subject and object is further established by combining linguistic rules and context, and the position relationship between verbs and nouns is used for judgment. Then, the discrete data is encoded based on the common features of the subject and the tools and means of implementing the behavior, the object and its attributes in order of change. The cutting processing data uses a 4-bit encoding, in which the first to fourth bits from left to right represent the production factor characteristics: processing means, equipment feature type, processing accuracy and cable parameters; the encoding of other subjects and implementation behavior tools is also similar. The information of cables and their accessories is encoded in 8 bits, where the 1st to 4th bits from left to right define the usage information of the cable, including domestic and international identification, application scenarios, conductivity and production period; and the 5th to 8th bits define the manufacturing characteristics of the cable, including component types, processing characteristics, accuracy levels and processing dimensions; the encoding method for other objects and changes in their attributes also follows similar rules; after the encoding is completed, the encoded text data is obtained.

[0065] S4. Perform text preprocessing on the encoded text data to obtain processed text data.

[0066] The method for obtaining processed text data includes: performing a word segmentation operation on selected text in the encoded text data to obtain word segments, and marking the part of speech of the word segments; based on the word segments and the part of speech, selectively identifying entity words with specific meanings in the encoded text data; performing dependency syntactic analysis on the identified text to obtain processed text data.

[0067] Text preprocessing is an important step in processing cable technology textual knowledge. In this embodiment, it mainly goes through three stages: word segmentation and part-of-speech tagging, named entity, and dependency syntactic analysis.

[0068] Specifically, firstly, the selected text is segmented. Chinese sentences can be regarded as an ordered set S consisting of Chinese characters [e] and words [E] elements. {∑E} is a subset of S. Single-character elements [e] can be combined into words [E] in a unidirectional and orderly manner. However, there is no obvious dividing boundary between words {∑E} in the sentence, so selective segmentation is performed according to the commonly used Chinese standards. Separators are marked at the boundaries of the segmented words to obtain an ordered word set {∑E}. The selection criteria for word [E] are mainly summarized as the upper and lower limits on the number of words, the fit between the word and the referenced professional vocabulary, etc. After segmentation, the part of speech of the word sequence is identified and marked in sequence with each word. Secondly, named entity recognition is based on word segmentation and part of speech tagging, and selectively identifies entity words with specific meanings in the text. Finally, dependency syntactic analysis is performed to decompose the dependency relationship between the word parts of speech {∑E} in the sentence around "subject, predicate, object, attributive, adverbial, complement", and judge and identify the core verb of the sentence. As the central word that dominates other words, the core verb is not dominated by other words. There is a direct dependency relationship between the words separated in the sentence structure, forming a dependency pair; in the dependency relationship, one is the dominant word, which is called the parent node in the syntactic analysis process, and the other is the subordinate word, which is called the child node. The dependency relationship is represented by a unidirectional dependency arc, from the subordinate word to the dominant word. According to the definition of dependency syntax, the syntactic dependency analysis is performed on the first sentence of the bogie process text example overview. After preprocessing, the processed text data is obtained.

[0069] S5. Perform triple modeling on the cable structured process based on the processed text data and relationship types, and map the obtained entity relationship triples into graph data.

[0070] The method for triple modeling includes: combining dependency syntax principles and Chinese grammar heuristic rules, combining entity sets and relationship types in processed text data, and obtaining entity-relationship triples.

[0071] In this embodiment, the triple expression format is: (head, label, tail), head and tail represent the head entity and tail entity respectively, head and tail belong to the entity set; label belongs to the relationship set, and the entity content is reflected as a node in the knowledge graph. Combining the dependency syntax principle and the heuristic rules of Chinese grammar, the entity and relationship of the triple are formed. It can be obtained by screening the parent-child nodes of the dependency syntax analysis, and the relationship can be abbreviated as: (relationship expression = adverbial * verb + complement? object?). Among them, * means that the adverbial does not appear or appears any number of times; + means that the verb appears at least once or any number of times;? Indicates that the complement and object appear once or not; construct the entity relationship triple, in the dependency syntax analysis result, according to the preposition and predicate verb of the adverbial structure, the noun indirectly points to the noun in the form of "subject, predicate and object", based on this grammatical logic, the constructed triple entity and relationship are screened based on the length of the word, excluding too short or too long content and punctuation, and the entity relationship triple can be constructed.

[0072] S6. Establish nodes and relationships based on the encoded text data and graph data, and construct a knowledge graph of the cable structured process library.

[0073] The method for constructing the knowledge graph of the cable structured process library includes: storing graph data in Neo4j and establishing nodes and relationships to construct the main framework of the cable process knowledge graph; storing the encoded text data in Neo4j, storing the feature encoding in the form of attributes with the entity in the library to generate nodes; traversing the name information and attribute information of the node, searching and merging the duplicate nodes to complete the construction of the knowledge graph.

[0074] In this embodiment, the graph database is a data structure that uses graphics to store data, and its main components are nodes, relationships, and attributes. The knowledge graph represents knowledge through a triple point cloud graph constructed by entities and relationships, stores the graph data processed by dependency syntax analysis in Neo4j to establish nodes and relationships, and constructs the main framework of the cable process knowledge graph. Then, the text data content encoded in the encoding system is imported into Neo4j, and the feature encoding is stored in the library with the entity in the form of attributes to generate nodes; after the text data is processed by dependency syntax analysis, some important content nodes such as "cable" will appear frequently, and the discrete data is highly correlated with the text data content, so there will be repeated content nodes in the graph, so the name information and attribute information of the node are traversed, and the repeated nodes are searched and merged, and the feature encoding results and the dependency syntax analysis results can be combined based on the repeated node content. At this point, the knowledge graph construction is completed.

[0075] Embodiment 2

[0076] In this embodiment, a knowledge graph construction system for a cable structured process library includes: an extraction and recognition module, a type analysis module, an encoding module, a preprocessing module, a mapping module, and a graph construction module;

[0077] The extraction and recognition module is used to extract knowledge from the cable structured process library and perform entity recognition on the extracted text to obtain knowledge text.

[0078] The type analysis module is used to perform relationship type analysis on the knowledge text to obtain the relationship type of the cable structured entity.

[0079] The encoding module is used to distinguish the concepts of cable structured knowledge subject and object for discrete data in the knowledge text, and encode the distinguished discrete data to obtain encoded text data.

[0080] The preprocessing module is used to perform text preprocessing on the encoded text data to obtain processed text data.

[0081] The mapping module performs triple modeling on the cable structured process based on the processed text data and relationship types, and maps the obtained entity relationship triples into graph data.

[0082] The graph construction module establishes nodes and relationships based on the encoded text data and graph data to construct a knowledge graph of the cable structured process library.

[0083] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A knowledge graph construction method for a cable structured process library, characterized in that: The following steps are involved: Extract knowledge from the cable structured process library and perform entity recognition on the extracted text to obtain knowledge text; Performing relationship type analysis on the knowledge text to obtain the relationship type of the cable structured entity; Performing cable structured knowledge subject and object concept distinction on the discrete data in the knowledge text, and encoding the discrete data after distinction to obtain encoded text data; Performing text preprocessing on the encoded text data to obtain processed text data; Performing triple modeling on the cable structured process based on the processed text data and the relationship type, and mapping the obtained entity relationship triples into graph data; Nodes and relationships are established based on the encoded text data and the graph data to construct a knowledge graph of the cable structured process library.

2. The method for constructing a knowledge graph of a cable structured process library according to claim 1, characterized in that: The method for obtaining the knowledge text includes: Construct an entity recognition model that combines the first BiLSTM layer with the CRF layer; Using the first BiLSTM layer to extract knowledge from the cable structured process library to obtain extracted knowledge; The CRF layer is used to perform entity recognition on the extracted knowledge to obtain the knowledge text.

3. The method for constructing a knowledge graph of a cable structured process library according to claim 1, characterized in that: The method for obtaining the relationship type of the cable structured entity includes: Constructing a Self-ATT-BiLSTM neural network, wherein the Self-ATT-BiLSTM neural network includes: an input layer, a word embedding layer, a second BiLSTM layer, a Self-ATT layer, and an output layer; The input layer receives the knowledge text; The word vector layer maps each word in the knowledge text into a word vector of fixed length; The second BiLSTM layer calculates the word vector to obtain high-level word vector features; The Self-ATT layer calculates the weight of each word, merges the word vector features of each time step, and multiplies the word vector features by the corresponding weight to obtain a text vector; The output layer calculates the text vector to obtain the relationship type between entities.

4. The method for constructing a knowledge graph of a cable structured process library according to claim 1, characterized in that: The method for obtaining the encoded text data includes: Preprocessing discrete data in the knowledge text, and using a named entity recognition model to identify named entities in the preprocessed text; Analyzing the discrete data after named entity recognition using dependency parsing to determine the relationship between the subject, verb, and object in the discrete data; The discrete data is encoded based on the common features of the subject and the tool and means data for implementing the behavior and the object and the attribute book data of the order of change to obtain the encoded text data.

5. The method for constructing a knowledge graph of a cable structured process library according to claim 1, characterized in that: The method for obtaining the processed text data includes: Performing a word segmentation operation on the selected text in the encoded text data to obtain word segments, and marking the part of speech of the word segments; Based on the word segmentation and the part of speech, selectively identifying entity words with specific meanings in the encoded text data; Dependency syntactic analysis is performed on the recognized text to obtain the processed text data.

6. The method for constructing a knowledge graph of a cable structured process library according to claim 1, characterized in that: The method for triple modeling includes: combining dependency syntax principles and Chinese grammar heuristic rules, combining the entity set in the processed text data and the relationship type, and obtaining the entity relationship triple.

7. The method for constructing a knowledge graph of a cable structured process library according to claim 1, characterized in that: The method for constructing the knowledge graph of the cable structured process library includes: The graph data is stored in Neo4j and nodes and relationships are established to construct the main framework of the cable process knowledge graph; The encoded text data is stored in Neo4j, and the feature code is stored in the database along with the entity in the form of attributes to generate a node; The name information and attribute information of the nodes are traversed, duplicate nodes are searched and merged, and the construction of the knowledge graph is completed.

8. A knowledge graph construction system for a cable structured process library, the system applying the method according to any one of claims 1 to 7, characterized in that: include: Extraction and recognition module, type analysis module, encoding module, preprocessing module, mapping module and graph construction module; The extraction and recognition module is used to extract knowledge from the cable structured process library and perform entity recognition on the extracted text to obtain knowledge text; The type analysis module is used to perform relationship type analysis on the knowledge text to obtain the relationship type of the cable structured entity; The encoding module is used to distinguish the cable structured knowledge subject and object concepts of the discrete data in the knowledge text, and encode the discrete data after distinction to obtain encoded text data; The preprocessing module is used to perform text preprocessing on the encoded text data to obtain processed text data; The mapping module performs triple modeling on the cable structured process based on the processed text data and the relationship type, and maps the obtained entity relationship triples into graph data; The graph construction module establishes nodes and relationships based on the encoded text data and the graph data to construct a knowledge graph of the cable structured process library.

Images