An information extraction and knowledge graph construction system and method

Abstract

The application discloses an information extraction and knowledge graph construction system and method, belongs to the technical field of data processing, and aims to solve the technical problems of how to improve the accuracy and completeness of information extraction and how to efficiently and accurately construct a knowledge graph. The system comprises a data preprocessing module, an entity recognition module, an event extraction module, a knowledge graph construction module and a knowledge representation and retrieval module.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a system and method for information extraction and knowledge graph construction. Background Technology

[0002] In the field of natural language processing, there are many fundamental technologies and algorithms, such as part-of-speech tagging, syntactic analysis, semantic role labeling, entity recognition, relation extraction, and event extraction. Knowledge graph construction involves knowledge representation, entity recognition, relation extraction, and graph databases.

[0003] The existing constraints and limitations mainly lie in methodological limitations and construction complexity. Traditional natural language processing methods may be limited by rules and pattern matching, making it difficult to handle semantic and contextual complexity, resulting in inaccurate and incomplete extracted information. Building knowledge graphs requires complex tasks such as entity recognition, relation extraction, and knowledge representation, consuming significant human and time resources, and the accuracy and consistency of the results are sometimes difficult to guarantee.

[0004] Improving the accuracy and completeness of information extraction, and constructing knowledge graphs efficiently and accurately are technical problems that need to be solved. Summary of the Invention

[0005] The technical objective of this invention is to address the above-mentioned shortcomings by providing an information extraction and knowledge graph construction system and method to solve the technical problems of how to improve the accuracy and completeness of information extraction and how to efficiently and accurately construct knowledge graphs.

[0006] In a first aspect, the present invention provides an information extraction and knowledge graph construction system, comprising:

[0007] The data preprocessing module provides word segmentation, part-of-speech tagging, and syntactic analysis services. The word segmentation service divides a continuous text sequence into discrete words or tags. The part-of-speech tagging service determines the part of speech or word class for each word. The syntactic analysis service determines the grammatical relationships between words in a sentence to obtain the syntactic structure.

[0008] The entity recognition module provides entity feature extraction and entity label prediction services. The entity feature extraction service learns the contextual information and semantic features of entities and learns long-term dependencies and local features in text sequences to obtain entity features. The entity label prediction service predicts the entity label for each word based on the dependencies between labels and the learned contextual features of entities.

[0009] The event extraction module provides event feature extraction, event template matching, event classification and extraction, and event relationship modeling services. The event feature extraction service extracts key features of events; the event template matching service identifies and extracts specific types of events based on predefined event templates; the event classification and extraction service classifies and extracts events by learning from labeled events; and the event relationship modeling service constructs relationships between events based on temporal order, logical relationships, and semantic connections in the text.

[0010] The knowledge graph construction module provides data model definition services, data storage services, data update and import services, and visualization and interaction services. The data model definition service defines entities, relationships, and attributes, as well as the organizational relationships between entities, relationships, and attributes, based on a graph structure. The data storage service uses a graph database as the storage engine for the knowledge graph. The data update and import service provides a data update and import interface that supports updating the knowledge graph in a full or incremental manner. The visualization and interaction service allows users to browse and navigate the knowledge graph based on the data model and graph database, through a graphical interface or visualization tools.

[0011] The knowledge representation and retrieval module provides knowledge representation services, knowledge retrieval services, similarity calculation services, and reasoning extension services. The knowledge representation service uses graph representation learning techniques to vectorize entities and relationships in a knowledge graph. The knowledge retrieval service supports users in querying and filtering data through query languages ​​or API interfaces to obtain entities and relationships. The similarity calculation service matches potentially related new knowledge based on the similarity between entities or relationships. The reasoning extension service analyzes the logical relationships and semantic connections between entities and relationships to discover new entities and relationships from the knowledge graph.

[0012] Preferably, the data preprocessing module is configured with a word segmentation model based on a hidden Markov model, which provides word segmentation services.

[0013] The data preprocessing module is configured with a part-of-speech tagging model based on the maximum entropy model, which provides part-of-speech tagging services.

[0014] The data preprocessing module is configured with a syntactic analysis model built by a statistical constituent syntactic analyzer. The syntactic analysis model provides syntactic analysis services. The working mode of the syntactic analysis model is as follows: the sentence is divided into multiple words, and the word structure and grammatical relations are analyzed to obtain the syntactic structure.

[0015] Preferably, the entity feature extraction service is used to learn the contextual information and semantic features of entities through multiple features, including part-of-speech tags, word forms, words in the context window, and bag-of-words models;

[0016] The entity feature extraction module is equipped with an entity recognition model built on recurrent neural networks and convolutional neural networks. The entity recognition model learns long-term dependencies and local features in the text sequence to obtain the entity features.

[0017] The entity recognition module is configured with an entity recognition model based on a conditional random field model, which provides entity label prediction services.

[0018] Preferably, the event extraction module is used to extract key features of an event from multiple features. The key features of the event include verbs, noun phrases, time phrases, and parts of speech. By learning the key features of the event, key factors in the event are identified. The key factors include actions, participants, and time.

[0019] The event template describes the relationships between the various elements of the event;

[0020] The event extraction module is equipped with an event classification and extraction model based on a recurrent neural network, which provides event classification and extraction services.

[0021] As a preferred approach, for graph-structured data models, entities are represented as nodes in the graph, relationships as edges, and attributes as properties of nodes and edges.

[0022] Preferably, the knowledge retrieval service supports querying based on entity attributes, relationship types, and event times as conditions to obtain entities and relationships that meet the conditions;

[0023] The reasoning extension service is used to analyze the logical relationships and semantic connections between entities and relations through logical reasoning or graph algorithms, and to discover new entities and relations from the knowledge graph.

[0024] In a second aspect, the present invention provides an information extraction and knowledge graph construction method, which extracts information and constructs a knowledge graph using the information extraction and knowledge graph construction system as described in any one of the first aspects, the method comprising the following steps:

[0025] Data preprocessing: The continuous text sequence is segmented into discrete words or tags by word segmentation service, the part-of-speech tagging service determines the part of speech or word class of each word by part-of-speech tagging service, and the syntactic relationship between words in the sentence is determined by syntactic analysis service to obtain the syntactic structure;

[0026] Entity recognition: The entity feature extraction service learns the contextual information and semantic features of entities, and learns long-term dependencies and local features in the text sequence to obtain entity features. Based on the dependencies between labels, the entity label prediction service learns the contextual features of entities to predict the entity label of each word.

[0027] Event Extraction: Key features of events are extracted through event feature extraction service. Based on predefined event templates, specific types of events are identified and extracted through event template matching service. Events are classified and extracted by learning labeled events and based on event classification and extraction service. Relationships between events are constructed based on the time sequence, logical relationship and semantic connection in the text through event relationship modeling service.

[0028] Knowledge graph construction: Based on graph structure, define entities, relations and attributes, as well as the organizational relationships between entities, relations and attributes through data model definition services. Use graph database as the storage engine of knowledge graph, and provide data update import interface for graph database. The data update import interface supports updating knowledge graph in full or incremental way. Based on data model and graph database, browse and navigate knowledge graph through graphical interface or visualization tools.

[0029] Knowledge Representation and Retrieval: Based on graph representation learning technology, entities and relationships in the knowledge graph are vectorized. Through the knowledge retrieval service, users can query and filter data through query language or API interface to obtain entities and relationships. Based on the similarity between entities or relationships, new knowledge with potential associations and similarities is matched. Through the reasoning extension service, the logical relationships and semantic connections between entities and relationships are analyzed to discover new entities and relationships from the knowledge graph.

[0030] As a preferred approach, for data preprocessing, word segmentation services are provided through a word segmentation model built on a Hidden Markov Model.

[0031] Part-of-speech tagging services are provided through a part-of-speech tagging model built on the maximum entropy model.

[0032] Syntactic analysis services are provided through a syntactic analysis model built using a statistical constituent syntactic analyzer.

[0033] As a preferred approach, for entity extraction, the contextual information and semantic features of the entity are learned through multiple features, including part-of-speech, word form, words in the context window, and bag-of-words model;

[0034] The entity recognition model built on recurrent neural networks and convolutional neural networks performs the following: learning long-term dependencies and local features in text sequences to obtain entity features;

[0035] Provide entity label prediction services through an entity recognition model built on a conditional random field model;

[0036] For event extraction, key features of the event are extracted based on multiple features. The key features of the event include verbs, noun phrases, time phrases, and parts of speech. By learning the key features of the event, key factors in the event are identified. Key factors include actions, participants, and time.

[0037] The event template describes the relationships between the various elements of the event;

[0038] Event classification and extraction services are provided by configuring an event classification and extraction model based on recurrent neural networks.

[0039] As a preferred approach, for graph-structured data models, entities are treated as nodes in the graph, relationships as edges, and attributes as properties of nodes and edges.

[0040] For knowledge representation and retrieval, queries are performed based on entity attributes, relation types, and event times as conditions, and entities and relations that meet the conditions are obtained through knowledge retrieval services;

[0041] By analyzing the logical relationships and semantic connections between entities and relations through logical reasoning or graph algorithms, new entities and relations can be discovered from the knowledge graph.

[0042] The information extraction and knowledge graph construction system and method of the present invention have the following advantages:

[0043] 1. Automated information extraction: It can automatically extract structured information from a large amount of text. Compared with traditional manual processing methods, the system can greatly improve the efficiency and accuracy of information extraction. Through automated information extraction, useful knowledge and information can be quickly obtained from massive amounts of data.

[0044] 2. Knowledge Graph Construction and Representation: Knowledge graphs are constructed using extracted entity, relationship, and event information to represent knowledge in a graphical structure. Knowledge graphs can more intuitively display the relationships between entities, helping users to better understand and explore knowledge. By establishing a structured knowledge graph, scattered information can be integrated into a unified framework, providing a comprehensive and accurate knowledge representation.

[0045] 3. Knowledge Retrieval and Reasoning: Users can perform efficient knowledge retrieval and reasoning. The system provides powerful query functions, which can perform complex search operations based on entity attributes, relationship types, event times, and other conditions to help users quickly obtain the knowledge they need. At the same time, the system supports reasoning based on knowledge graphs, which can discover new knowledge and associations by analyzing the logical relationships and semantic connections between entities and relationships.

[0046] 4. Knowledge Application and Intelligent Services: Based on knowledge representation services, knowledge retrieval services, similarity calculation services, and reasoning extension services, various knowledge applications and intelligent services can be developed. Knowledge graphs can provide a rich knowledge base for search engines, recommendation systems, intelligent question answering, etc., improving user experience and service quality. By combining knowledge graphs with other artificial intelligence technologies, more intelligent and personalized knowledge services can be achieved. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] The invention will be further described below with reference to the accompanying drawings.

[0049] Figure 1 This is a flowchart of an information extraction and knowledge graph construction method according to Example 2. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0051] This invention provides an information extraction and knowledge graph construction system and method to address the technical problems of how to improve the accuracy and completeness of information extraction and how to efficiently and accurately construct knowledge graphs.

[0052] Example 1:

[0053] The present invention provides an information extraction and knowledge graph construction system, comprising a data preprocessing module, an entity recognition module, an event extraction module, a knowledge graph construction module, and a knowledge representation and retrieval module.

[0054] The data preprocessing module provides word segmentation, part-of-speech tagging, and syntactic analysis services. The word segmentation service divides a continuous text sequence into discrete words or tags. The part-of-speech tagging service determines the part of speech or word class for each word. The syntactic analysis service determines the grammatical relationships between words in a sentence to obtain the syntactic structure.

[0055] In this embodiment, the data preprocessing module is configured with a word segmentation model based on a Hidden Markov Model, which provides word segmentation services; simultaneously, it is configured with a part-of-speech tagging model based on a maximum entropy model, which provides part-of-speech tagging services; and it is also configured with a syntactic analysis model based on a statistical constituent syntactic analyzer, which provides syntactic analysis services. The working mode of the syntactic analysis model is as follows: it divides the sentence into multiple words and analyzes the word structure and grammatical relations to obtain the syntactic structure.

[0056] Correspondingly, in this embodiment, the data preprocessing module can perform word segmentation, part-of-speech tagging, and syntactic analysis operations.

[0057] Word segmentation is the process of dividing a continuous text sequence into discrete words or tags. This embodiment uses a Hidden Markov Model (HMM) as the word segmentation algorithm model. A Hidden Markov Model (HMM) is a commonly used statistical model for modeling and analyzing sequential data. This model consists of a Markov chain and an observation sequence. There are two key components in an HMM: Hidden State: Represents the state within the system that is not directly observed. Each hidden state has an associated observation. Hidden states can be discrete, represented as symbols or labels. Observation Sequence: Represents the data sequence that we can directly observe. There is a certain correlation between the observation sequence and the hidden state. The core assumption of the HMM is the Markov property: the probability distribution of the current state depends only on the previous state and is independent of earlier states. This assumption is called the first-order Markov property. Using a Hidden Markov Model as the word segmentation algorithm model can strengthen contextual dependencies and compensate for error propagation. Hidden Markov Models (HMMs) consider contextual information, meaning the segmentation result of the current word depends on the words preceding and following it. This contextual dependency helps resolve polysemy and ambiguity, improving segmentation accuracy. Since Chinese words lack explicit boundary markers, incorrect segmentation results can impact subsequent processing tasks. HMMs, by considering the global context during segmentation, can reduce error propagation and improve overall segmentation quality.

[0058] Part-of-speech tagging is the process of determining the part of speech or word class for each word. This embodiment uses the maximum entropy model as the part-of-speech tagging algorithm model. The maximum entropy model is a statistical model used for classification and prediction. It is based on the principle of maximum entropy in information theory. The maximum entropy model has been widely used in natural language processing, machine learning, and statistics. The core idea of ​​the maximum entropy model is to select the model with the highest entropy among the probability distributions that satisfy certain constraints as the optimal model, given some constraints. Entropy represents the degree of uncertainty or disorder of a probability distribution. The maximum entropy principle states that in the absence of other prior knowledge, the most uncertain model should be selected to maintain the consistency and robustness of the model. The maximum entropy model can select appropriate features according to different languages ​​and application scenarios, and allows multiple features to be combined for modeling. This flexibility allows the maximum entropy model to make full use of contextual information, lexical information, and other linguistic features to improve the accuracy of part-of-speech tagging. In part-of-speech tagging tasks, the tagging of each word usually depends on the tags of other words in its context. The maximum entropy model can solve the problems of polysemy and ambiguity by considering global contextual information. It can capture the transition probabilities between parts of speech, thus inferring the tag of each word more accurately. The maximum entropy model has good interpretability, providing the weight and contribution of each feature in the model. This makes the model's output interpretable and understandable, facilitating model debugging and improvement.

[0059] Syntactic analysis is the process of determining the grammatical relationships between words in a sentence. This embodiment uses a statistical constituent parser as the syntactic analysis model. The constituent parser can perform structured analysis of sentences, dividing them into words and phrases, and determining the hierarchical and dependency relationships between them. This analysis can provide a deep understanding of sentences, revealing their constituent parts and the grammatical relationships between them. The statistical constituent parser uses a probabilistic model and training data for grammatical parsing, learning and inference to determine the most probable syntactic structure. Such analysis helps to understand the grammatical rules and semantic meaning of sentences, thus supporting natural language understanding and generation tasks. The constituent parser can provide richer contextual information to improve the performance of language models. By revealing the phrase structure and dependency relationships in sentences, the parser can provide more accurate contextual representations for language model generation and prediction.

[0060] The entity recognition model provides entity feature extraction and entity label prediction services. The entity feature extraction service learns the contextual information and semantic features of entities, and learns long-term dependencies and local features in text sequences to obtain entity features. The entity label prediction service predicts the entity label for each word based on the dependencies between labels and the learned contextual features of entities.

[0061] In this embodiment, the entity feature extraction service is used to learn the contextual information and semantic features of entities through multiple features. Entity features include part-of-speech, word form, words within the context window, and bag-of-words model. At the same time, the entity feature extraction module is configured with an entity recognition model built based on recurrent neural networks and convolutional neural networks. The entity recognition model learns long-term dependencies and local features in the text sequence to obtain the entity features. Meanwhile, the entity recognition module is configured with an entity recognition model built based on a conditional random field model. The entity recognition model provides entity label prediction services.

[0062] Correspondingly, the entity feature extraction module can provide operations such as feature extraction and entity label prediction.

[0063] In entity recognition, feature extraction is a crucial step. On the one hand, this embodiment uses a variety of features to capture the contextual information and semantic features of entities. These features include part-of-speech tags, word forms, words within the context window, bag-of-words models, etc. By comprehensively utilizing these features, the accuracy and robustness of entity recognition can be improved. On the other hand, in order to further improve the performance of entity recognition, this embodiment introduces recurrent neural networks (RNNs) and convolutional neural networks (CNNs) from deep learning models. By constructing an entity recognition model using the above models, long-term dependencies and local features in text sequences can be learned, and entity recognition can be performed through end-to-end training.

[0064] For entity label prediction, this embodiment uses the classic Conditional Random Field (CRF) model as the entity label prediction model. The CRF model can take into account the interdependencies between labels and predict the entity label of each word by learning contextual features. The CRF model is widely used in entity recognition tasks and has good performance.

[0065] The event extraction module provides event feature extraction, event template matching, event classification and extraction, and event relationship modeling services. The event feature extraction service extracts key features of events, the event template matching service identifies and extracts specific types of events based on predefined event templates, the event classification and extraction service classifies and extracts events by learning from labeled events, and the event relationship modeling service builds relationships between events based on the time sequence, logical relationships, and semantic connections in the text.

[0066] In this embodiment, the event extraction module is used to extract key features of an event from multiple features. These key features include verbs, noun phrases, time phrases, and parts of speech. By learning these key features, the module identifies key factors within the event, including actions, participants, and time. The event template describes the relationships between the various elements of the event. Simultaneously, the event extraction module is configured with an event classification and extraction model based on a recurrent neural network, providing event classification and extraction services.

[0067] Correspondingly, the event extraction module can perform operations such as event feature extraction, event template matching, event classification and extraction, and event relationship modeling.

[0068] For event feature extraction, this embodiment employs multiple features to capture key characteristics of the event, including verbs, noun phrases, time phrases, and parts of speech. By extracting these features, key elements such as actions, participants, and time within the event can be identified, thereby achieving accurate event extraction.

[0069] For event target matching, this embodiment employs an event template matching method to identify and extract specific types of events. An event template is a predefined pattern or rule that describes the relationships between various elements within an event. By matching text against the event template, specific types of events and their related information can be identified.

[0070] For event classification and extraction, this embodiment introduces a recurrent neural network (RNN) model. An event classification and extraction model is constructed based on the RNN model to further improve the accuracy of event extraction. Automatic event classification and extraction are achieved by learning from a large amount of labeled data and using the event classification and extraction model.

[0071] In addition to identifying and extracting individual events, this embodiment can also model the relationships between events. By analyzing the temporal sequence, logical relationships, and semantic connections in the text, a network of relationships between events is constructed, further enriching the event information in the knowledge graph.

[0072] The knowledge graph construction module provides data model definition services, data storage services, data update and import services, and visualization and interaction services. The data model definition service is used to define entities, relations, and attributes, as well as the organizational relationships between entities, relations, and attributes, based on the graph structure. The data storage service uses the graph database as the storage engine for the knowledge graph. The data update and import service provides a data update and import interface that supports updating the knowledge graph in a full or incremental manner. The visualization and interaction service is used to browse and navigate the knowledge graph based on the data model and graph database through a graphical interface or visualization tools.

[0073] In this embodiment, for the graph-structured data model, entities are treated as nodes in the graph, relationships as edges in the graph, and attributes as attributes of nodes and edges.

[0074] Correspondingly, the knowledge graph construction module in this embodiment can provide operations such as data model design, data storage, data import and update, visualization and interaction.

[0075] The design of a knowledge graph data model includes the definition and organization of entities, relations, and attributes. This embodiment uses a graph structure as the data model, where entities are nodes, relations are edges, and attributes are the properties of nodes and edges. By defining appropriate entity types, relation types, and attribute types, a rich and flexible knowledge graph data model can be established.

[0076] Data Storage: This embodiment uses a graph database as the storage engine for the knowledge graph. A graph database is a database system specifically designed for storing and querying graph data. It can efficiently store large-scale nodes and edges and provide flexible query and navigation functions. The storage model and indexing mechanism of the graph database can effectively support the query and analysis operations of the knowledge graph.

[0077] Data Import and Updates: Knowledge graphs require regular updates and maintenance to ensure the accuracy and timeliness of their content. This embodiment provides data import and updates for a library, allowing new data sources to be integrated into the knowledge graph and incrementally updated and synchronized, thus ensuring the continuous evolution and updating of the knowledge graph.

[0078] Visualization and Interaction: To better display and utilize the content of the knowledge graph, this embodiment provides visualization and interaction functions based on the knowledge graph and graph database. Users can browse and navigate the knowledge graph through a graphical interface or visualization tools to more intuitively understand and explore the relationships between knowledge.

[0079] The knowledge representation and retrieval module provides knowledge representation services, knowledge retrieval services, similarity calculation services, and reasoning extension services. The knowledge representation service uses graph representation learning technology to vectorize entities and relationships in the knowledge graph. The knowledge retrieval service supports users in querying and filtering data through query languages ​​or API interfaces to obtain entities and relationships. The similarity calculation service matches potentially related new knowledge based on the similarity between entities or relationships. The reasoning extension service analyzes the logical relationships and semantic connections between entities and relationships to discover new entities and relationships from the knowledge graph.

[0080] In this embodiment, the knowledge retrieval service supports querying based on entity attributes, relationship types, and event times as conditions to obtain entities and relationships that meet the conditions. The reasoning extension service is used to analyze the logical relationships and semantic connections between entities and relationships through logical reasoning or graph algorithms to discover new entities and relationships from the knowledge graph.

[0081] Correspondingly, in this embodiment, the knowledge representation and retrieval module can provide operations such as knowledge representation, knowledge retrieval, similarity calculation, reasoning, and expansion.

[0082] Knowledge Representation: This embodiment employs graph representation learning technology to vectorize entities and relations in a knowledge graph. Graph representation learning is a technique that maps nodes and edges in a graph to a low-dimensional vector space. By learning the representation vectors of nodes and edges, the semantic relationships between them can be captured. By mapping entities and relations to a continuous vector space, more efficient and flexible knowledge reasoning and analysis can be performed.

[0083] Knowledge Retrieval: In this embodiment, users can perform complex queries and filtering operations using query languages ​​or API interfaces. Users can query based on entity attributes, relationship types, event times, and other conditions to retrieve entities and relationships that meet those conditions. Knowledge retrieval helps users quickly find relevant knowledge and supports the needs of knowledge reasoning and analysis.

[0084] Similarity Calculation: This embodiment also provides similarity calculation for knowledge representation and retrieval. By calculating the similarity between entities or the relevance between relationships, potential connections and similar knowledge can be discovered. Similarity calculation is based on a vector space model and graph matching algorithms, helping users discover new knowledge and connections.

[0085] Reasoning and Expansion: The knowledge representation and retrieval in this embodiment support reasoning and expansion based on knowledge graphs. By analyzing the logical relationships and semantic connections between entities and relations, reasoning operations can be performed to discover new entities and relations. Reasoning can be implemented through logical reasoning, graph algorithms, and other methods, helping users to mine more implicit knowledge and connections from the knowledge graph.

[0086] The system in this embodiment automates the processing and knowledge representation of large amounts of text data by applying natural language processing algorithms to information extraction and knowledge graph construction. This system improves the accuracy and efficiency of information extraction and constructs a consistent and scalable knowledge graph, providing effective support for knowledge representation, retrieval, and application. Furthermore, through considerations of real-time processing and scalability, this technical solution can adapt to the ever-increasing volume of data and the demands of real-time applications.

[0087] Example 2:

[0088] This invention discloses an information extraction and knowledge graph construction method, which extracts information and constructs a knowledge graph using the system disclosed in Example 1. The method includes steps such as data preprocessing, entity recognition, event extraction, knowledge graph construction, knowledge representation, and search.

[0089] Data preprocessing: The continuous text sequence is segmented into discrete words or tags by word segmentation service, the part-of-speech tagging service determines the part of speech or word class of each word by part-of-speech tagging service, and the syntactic relationship between words in the sentence is determined by syntactic analysis service to obtain the syntactic structure.

[0090] In this embodiment, for data preprocessing, word segmentation service is provided by a word segmentation model based on a hidden Markov model; part-of-speech tagging service is provided by a part-of-speech tagging model based on a maximum entropy model; and syntactic analysis service is provided by a syntactic analysis model based on a statistical constituent syntactic analyzer.

[0091] This includes specific data preprocessing operations such as word segmentation, part-of-speech tagging, and syntactic analysis.

[0092] Word segmentation is the process of dividing a continuous sequence of text into discrete words or tags. In this embodiment, a Hidden Markov Model is used as the word segmentation algorithm model.

[0093] Part-of-speech tagging (POS) is the process of determining the part of speech or word class for each word. This embodiment uses the maximum entropy model as the POS tagging algorithm. The maximum entropy model can select appropriate features based on different languages ​​and application scenarios, and allows multiple features to be combined for modeling. This flexibility enables the maximum entropy model to fully utilize contextual information, lexical information, and other linguistic features, improving the accuracy of POS tagging. In POS tagging tasks, the labeling of each word typically depends on the labels of other words in its context. The maximum entropy model can solve polysemy and ambiguity problems by considering global contextual information. It can capture the transition probabilities between parts of speech, thus more accurately inferring the labeling of each word. The maximum entropy model has good interpretability, providing the weight and contribution of each feature in the model. This makes the model's output interpretable and understandable, facilitating model debugging and improvement.

[0094] Syntactic analysis is the process of determining the grammatical relationships between words in a sentence. This embodiment uses a statistically based constituent parser as the syntactic analysis model. The constituent parser can perform structured analysis of sentences, dividing them into words and phrases, and determining the hierarchical and dependency relationships between them. The constituent parser can provide richer contextual information to improve the performance of the language model. By revealing the phrase structure and dependency relationships in the sentence, the parser can provide more accurate contextual representations for language model generation and prediction.

[0095] Entity recognition: The entity feature extraction service learns the contextual information and semantic features of entities, and learns long-term dependencies and local features in the text sequence to obtain entity features. Based on the dependencies between labels, the entity label prediction service learns the contextual features of entities to predict the entity label of each word.

[0096] In this embodiment, for entity feature extraction, the contextual information and semantic features of the entity are learned through multi-feature learning. The entity features include part-of-speech, word form, words in the context window, and bag-of-words model. The entity recognition model built based on recurrent neural network and convolutional neural network performs the following: learning long-term dependencies and local features in the text sequence to obtain the entity features; and the entity recognition model built based on conditional random field model provides entity label prediction service.

[0097] As a specific implementation of entity recognition, it includes operations such as feature extraction and entity label prediction.

[0098] In entity recognition, feature extraction is a crucial step. On the one hand, this embodiment uses a variety of features to capture the contextual information and semantic features of entities. These features include part-of-speech tags, word forms, words within the context window, bag-of-words models, etc. By comprehensively utilizing these features, the accuracy and robustness of entity recognition can be improved. On the other hand, in order to further improve the performance of entity recognition, this embodiment introduces recurrent neural networks (RNNs) and convolutional neural networks (CNNs) from deep learning models. By constructing an entity recognition model using the above models, long-term dependencies and local features in text sequences can be learned, and entity recognition can be performed through end-to-end training.

[0099] For entity label prediction, this embodiment uses the classic Conditional Random Field (CRF) model as the entity label prediction model. The CRF model can take into account the interdependencies between labels and predict the entity label of each word by learning contextual features. The CRF model is widely used in entity recognition tasks and has good performance.

[0100] Event Extraction: Key features of events are extracted through event feature extraction service. Based on predefined event templates, specific types of events are identified and extracted through event template matching service. Events are classified and extracted by learning from labeled events and based on event classification and extraction service. Relationships between events are constructed based on the time sequence, logical relationship and semantic connection in the text through event relationship modeling service.

[0101] In this embodiment, for event extraction, key features of the event are extracted based on multiple features. These key features include verbs, noun phrases, time phrases, and parts of speech. Key factors in the event are identified by learning these key features, including actions, participants, and time. The event template describes the relationships between the various elements of the event. This embodiment provides event classification and extraction services by configuring an event classification and extraction model based on a recurrent neural network.

[0102] The specific implementation of event extraction includes operations such as event feature extraction, event template matching, event classification and extraction, and event relationship modeling.

[0103] For event feature extraction, this embodiment employs multiple features to capture key characteristics of the event, including verbs, noun phrases, time phrases, and parts of speech. By extracting these features, key elements such as actions, participants, and time within the event can be identified, thereby achieving accurate event extraction.

[0104] For event target matching, this embodiment employs an event template matching method to identify and extract specific types of events. An event template is a predefined pattern or rule that describes the relationships between various elements within an event. By matching text against the event template, specific types of events and their related information can be identified.

[0105] For event classification and extraction, this embodiment introduces a recurrent neural network (RNN) model. An event classification and extraction model is constructed based on the RNN model to further improve the accuracy of event extraction. Automatic event classification and extraction are achieved by learning from a large amount of labeled data and using the event classification and extraction model.

[0106] To model the relationships between events, we construct a network of relationships between events by analyzing the temporal sequence, logical relationships, and semantic connections in the text, thereby further enriching the event information in the knowledge graph.

[0107] Knowledge graph construction: Based on graph structure, define entities, relations and attributes, as well as the organizational relationships between entities, relations and attributes, through data model definition services. Use graph database as the storage engine for knowledge graph, and provide data update import interface for graph database. The data update import interface supports updating knowledge graph in full or incremental way. Based on data model and graph database, browse and navigate knowledge graph through graphical interface or visualization tools.

[0108] In this embodiment, for a graph-structured data model, entities are treated as nodes in the graph, relationships as edges in the graph, and attributes as attributes of nodes and edges.

[0109] As a specific implementation of knowledge graph construction, it includes operations such as data model design, data storage, data import and update, visualization and interaction.

[0110] The design of a knowledge graph data model includes the definition and organization of entities, relations, and attributes. This embodiment uses a graph structure as the data model, where entities are nodes, relations are edges, and attributes are the properties of nodes and edges. By defining appropriate entity types, relation types, and attribute types, a rich and flexible knowledge graph data model can be established.

[0111] Data Storage: This embodiment uses a graph database as the storage engine for the knowledge graph. A graph database is a database system specifically designed for storing and querying graph data. It can efficiently store large-scale nodes and edges and provide flexible query and navigation functions. The storage model and indexing mechanism of the graph database can effectively support the query and analysis operations of the knowledge graph.

[0112] Data Import and Updates: Knowledge graphs require regular updates and maintenance to ensure the accuracy and timeliness of their content. This embodiment provides data import and updates for a library, allowing new data sources to be integrated into the knowledge graph and incrementally updated and synchronized, thus ensuring the continuous evolution and updating of the knowledge graph.

[0113] Visualization and Interaction: To better display and utilize the content of the knowledge graph, this embodiment provides visualization and interaction functions based on the knowledge graph and graph database. Users can browse and navigate the knowledge graph through a graphical interface or visualization tools to more intuitively understand and explore the relationships between knowledge.

[0114] Knowledge Representation and Retrieval: Based on graph representation learning technology, entities and relationships in the knowledge graph are vectorized. Through the knowledge retrieval service, users can query and filter data through query language or API interface to obtain entities and relationships. Based on the similarity between entities or relationships, new knowledge with potential associations and similarities is matched. Through the reasoning extension service, the logical relationships and semantic connections between entities and relationships are analyzed to discover new entities and relationships from the knowledge graph.

[0115] In this embodiment, for knowledge representation and retrieval, queries are performed based on entity attributes, relationship types, and event times as conditions. Entities and relationships that meet the conditions are obtained through knowledge retrieval services. Logical reasoning or graph algorithms are used to analyze the logical relationships and semantic connections between entities and relationships, and new entities and relationships are discovered from the knowledge graph.

[0116] As a specific implementation of knowledge representation and retrieval, it includes operations such as knowledge representation, knowledge retrieval, similarity calculation, reasoning, and expansion.

[0117] Knowledge Representation: This embodiment employs graph representation learning technology to vectorize entities and relations in a knowledge graph. Graph representation learning is a technique that maps nodes and edges in a graph to a low-dimensional vector space. By learning the representation vectors of nodes and edges, the semantic relationships between them can be captured. By mapping entities and relations to a continuous vector space, more efficient and flexible knowledge reasoning and analysis can be performed.

[0118] Knowledge Retrieval: In this embodiment, users can perform complex queries and filtering operations using query languages ​​or API interfaces. Users can query based on entity attributes, relationship types, event times, and other conditions to retrieve entities and relationships that meet those conditions. Knowledge retrieval helps users quickly find relevant knowledge and supports the needs of knowledge reasoning and analysis.

[0119] Similarity Calculation: This embodiment also provides similarity calculation for knowledge representation and retrieval. By calculating the similarity between entities or the relevance between relationships, potential connections and similar knowledge can be discovered. Similarity calculation is based on a vector space model and graph matching algorithms, helping users discover new knowledge and connections.

[0120] Reasoning and Expansion: The knowledge representation and retrieval in this embodiment support reasoning and expansion based on knowledge graphs. By analyzing the logical relationships and semantic connections between entities and relations, reasoning operations can be performed to discover new entities and relations. Reasoning can be implemented through logical reasoning, graph algorithms, and other methods, helping users to mine more implicit knowledge and connections from the knowledge graph.

[0121] The present invention has been shown and described in detail above with reference to the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above embodiments, those skilled in the art will know that more embodiments of the present invention can be obtained by combining the means in the different embodiments described above, and these embodiments are also within the protection scope of the present invention.

Claims

1. An information extraction and knowledge graph construction system, characterized in that, Comprise: a data preprocessing module for providing word segmentation services, part-of-speech tagging services, and syntactic analysis services, the word segmentation services for splitting continuous text sequences into discrete words or tokens, the part-of-speech tagging services for determining the part of speech or word class of each word, and the syntactic analysis services for determining the syntactic relationships between words in a sentence to obtain the syntactic structure; an entity recognition module for providing entity feature extraction services and entity label prediction services, the entity feature extraction services for learning the contextual information and semantic features of entities and for learning the long-term dependencies and local features in text sequences to obtain the features of entities, and the entity label prediction services for predicting the entity labels of each word based on the dependencies between labels and the contextual features of entities; an event extraction module for providing event feature extraction services, event template matching services, event classification and extraction services, and event relationship modeling services, the event feature extraction services for extracting the key features of events, the event template matching services for identifying and extracting specific types of events based on predefined event templates, the event classification and extraction services for classifying and extracting events by learning labeled events, and the event relationship modeling services for constructing the relationships between events based on the temporal order, logical relationships, and semantic connections in text; a knowledge graph construction module for providing data model definition services, data storage services, data update import services, and visualization interaction services, the data model definition services for defining entities, relationships, and attributes and the organizational relationships between entities, relationships, and attributes based on graph structures, the data storage services for using graph databases as the storage engines of knowledge graphs, the data update import services for providing data update import interfaces that support updating knowledge graphs in full or incremental ways, and the visualization interaction services for browsing and navigating knowledge graphs through graphical interfaces or visualization tools based on data models and graph databases; a knowledge representation and retrieval module for providing knowledge representation services, knowledge retrieval services, similarity calculation services, and reasoning extension services, the knowledge representation services for vectorizing the entities and relationships in knowledge graphs based on graph representation learning techniques, the knowledge retrieval services for supporting users to query and filter data through query languages or API interfaces to obtain entities and relationships, the similarity calculation services for matching potentially associated similar new knowledge based on the similarity between entities or the similarity between relationships, and the reasoning extension services for discovering new entities and relationships from knowledge graphs by analyzing the logical relationships and semantic connections between entities and relationships through logical reasoning or graph algorithms; The reasoning extension services are used to discover new entities and relationships from knowledge graphs by analyzing the logical relationships and semantic connections between entities and relationships through logical reasoning or graph algorithms. The event extraction module is configured with an event classification and extraction model based on a recurrent neural network, and the event classification and extraction model provides event classification and extraction services. The knowledge retrieval service supports querying based on the attributes of entities, the types of relationships, and the time of events as conditions to obtain entities and relationships that meet the conditions. 2.The information extraction and knowledge graph construction system of claim 1, wherein, The data preprocessing module is configured with a word segmentation model based on a hidden Markov model, and the word segmentation model provides word segmentation services. The data preprocessing module is configured with a part-of-speech tagging model based on a maximum entropy model, and the part-of-speech tagging model provides part-of-speech tagging services. The data preprocessing module is configured with a syntactic analysis model based on a statistical component-based syntactic analyzer, and the syntactic analysis model provides syntactic analysis services. The working mode of the syntactic analysis model is to divide a sentence into multiple words and analyze the word structure and grammatical relationships to obtain the syntactic structure. 3.The information extraction and knowledge graph construction system of claim 1, wherein, The entity feature extraction service is configured to learn the context information and semantic features of entities through multi-feature learning. Entity features include part-of-speech, word form, words within a context window, and bag-of-words. The entity feature extraction module is configured with an entity recognition model based on a recurrent neural network and a convolutional neural network. The entity recognition model learns long-term dependencies and local features in text sequences to obtain entity features. The entity recognition module is configured with an entity recognition model based on a conditional random field model. The entity recognition model provides entity label prediction services. 4.The information extraction and knowledge graph construction system of claim 1, wherein, For a graph-structured data model, entities are nodes in the graph, relationships are edges in the graph, and attributes are properties of nodes and edges.

5. An information extraction and knowledge graph construction method, characterized in that, An information extraction and knowledge graph construction system according to any one of claims 1-4 is used to extract information and construct a knowledge graph. The method includes the following steps: Data preprocessing: continuous text sequences are divided into discrete words or tokens through word segmentation services, the part-of-speech of each word is determined through part-of-speech tagging services, and the grammatical relationships between words in a sentence are determined through syntactic analysis services to obtain syntactic structures. Entity recognition: the context information and semantic features of entities are learned through entity feature extraction services, and long-term dependencies and local features in text sequences are learned to obtain entity features. The context features of entities are learned based on the dependencies between labels to predict the entity labels of each word through entity label prediction services. Event extraction: extracting key features of events through an event feature extraction service, identifying and extracting specific types of events based on predefined event templates through an event template matching service, classifying and extracting events based on event classification and extraction services by learning labeled events, building relationships between events based on temporal order, logical relationships, and semantic connections in text through an event relationship modeling service; Knowledge graph construction: defining entities, relationships, and attributes and the organizational relationships between them based on graph structure through data model definition services, using a graph database as a storage engine for the knowledge graph, providing data update import interfaces for the graph database, which support updating the knowledge graph in full or incremental mode, and browsing and navigating the knowledge graph through graphical interfaces or visualization tools based on data models and graph databases; Knowledge representation and retrieval: vectorizing representations of entities and relationships in the knowledge graph based on graph representation learning techniques, supporting user data queries and data filtering through query languages or API interfaces to obtain entities and relationships through knowledge retrieval services, and matching potentially associated similar new knowledge based on the similarity between entities or relationships; discovering new entities and relationships from the knowledge graph by analyzing the logical relationships and semantic connections between entities and relationships through logical reasoning or graph algorithms; Among them, for event extraction, key features of events are extracted based on multiple features, including verbs, noun phrases, time phrases, and parts of speech, and key factors in events are identified by learning key features of events, including actions, participants, and times; the event template describes the relationships between various elements in the event; event classification and extraction services are provided through a configured event classification and extraction model based on recurrent neural networks; Among them, for knowledge representation and retrieval, entities and relationships that meet the conditions are obtained through knowledge retrieval services according to the attributes of entities, the types of relationships, and the times of events. 6.The information extraction and knowledge graph construction method of claim 5, wherein, For data preprocessing, a segmentation service is provided through a segmentation model based on a hidden Markov model; A part-of-speech tagging service is provided through a part-of-speech tagging model based on a maximum entropy model; A syntax analysis service is provided through a syntax analysis model based on a statistical constituent syntax analyzer. 7.The information extraction and knowledge graph construction method of claim 5, wherein, For entity extraction, the context information and semantic features of entities are learned through multiple features, including parts of speech, word forms, words within a context window, and bag-of-words models; Through an entity recognition model based on recurrent neural networks and convolutional neural networks, the following is performed: learning long-term dependencies and local features in text sequences to obtain entity features; An entity label prediction service is provided through an entity recognition model based on a conditional random field model. 8.The information extraction and knowledge graph construction method of claim 5, wherein, For the data model of the graph structure, entities are nodes in the graph, relationships are edges in the graph, and attributes are attributes of nodes and edges.

Images