A data entity recognition and relationship extraction method based on a trusted data space

Abstract

The application discloses a kind of based on trusted data space's data entity identification and relationship extraction method, belong to industrial artificial intelligence and data processing technical field, the field corpus of the present application is constructed and is carried out semantic annotation, is fused word embedding, character embedding and label embedding feature by vectorization input, introduce BiLSTM encoder and carry out context modeling to text sequence, utilize conditional random field CRF decoding output layer and carry out decoding, identify the key named entity in industrial text;Combined with dependency syntax analysis and context rule, extract the semantic relationship between entities, finally build the knowledge graph including supply-manufacturing-service semantic link.The present application realizes the structured expression of complex semantic data in trusted data space, improves the understandability, traceability and sharing value of data, has the advantages of high entity recognition accuracy, complete relationship modeling, strong visualization capability, etc., is suitable for typical scenes such as automobile manufacturing, supply chain collaboration, industrial operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of industrial artificial intelligence and data processing technology, specifically relating to a method for data entity recognition and relation extraction based on a trusted data space. Background Technology

[0002] With the increasing diversification and heterogeneity of data in industrial manufacturing scenarios, traditional management models centered on structured data are no longer sufficient to meet the complex needs of trusted data flow, semantic understanding, and relationship modeling. Especially in trusted data spaces, the parsing of large amounts of unstructured text, messages, and business description files urgently requires entity recognition and semantic relationship extraction capabilities to achieve intelligent data processing for cross-domain collaboration. Therefore, natural language processing methods for entity recognition and relationship extraction have gained widespread attention in the industrial field in recent years. Existing technologies have seen scholars and enterprises explore aspects such as optimizing entity relationship extraction model structures, attention enhancement mechanisms, and industrial text applications, resulting in a series of representative technical solutions.

[0003] Chinese patent CN112395407A proposes a method for identifying enterprise entity relationships based on a two-stage extraction strategy. It achieves effectiveness in reducing negative sample interference through subject-object entity decoupling modeling. However, this method relies on pre-set templates, has limited adaptability, and struggles to cover the complex semantic variations implicit in industrial data. Chinese patent CN114610819A improves the extraction efficiency of structured information by constructing triples based on pre-annotated attribute relationships between individuals. However, this patent is heavily reliant on domain vocabulary and structure, making it difficult to maintain stable performance in complex semantic scenarios within industrial corpora. Chinese patent CN115600595A introduces BiGRU and a multi-branch attention mechanism for entity relationship extraction, enhancing contextual modeling capabilities. However, it is only applicable to general text corpora and does not consider the complexity of data source structures or the lack of named entity standards in trusted data spaces. Chinese patent CN112200317A combines dependency analysis and entity disambiguation mechanisms to establish a relatively complete entity relationship extraction process. However, this scheme is highly dependent on syntax and performs poorly in recognizing non-standard industrial expressions (such as operation logs and command statements). Chinese patent CN113822434A proposes a model selection strategy for knowledge distillation, which can be used to assist in the optimization of relation extraction models, but it does not have relation extraction capabilities itself and needs to rely on external models.

[0004] In summary, existing technologies mostly focus on entity relationship modeling in general semantic text, lacking systematic adaptation to the structural characteristics and semantic ambiguity of multi-source heterogeneous industrial data in the trusted data space. They still have shortcomings in semantic understanding, generalization ability, and accuracy of relationship extraction. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method for data entity identification and relation extraction based on a trusted data space. The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for data entity identification and relation extraction based on a trusted data space, comprising: Collect multi-source heterogeneous text data from trusted data space to construct an industrial corpus; annotate the keywords of the training data in the industrial corpus to obtain text sequences; The text sequence is vectorized into an input representation to obtain a sequence embedding vector; The sequence embedding vector is input into the BiLSTM encoder in the pre-built BiLSTM-CRF model to perform context sequence modeling and obtain the hidden state information corresponding to each word in the text data. The hidden state information is input into the Conditional Random Field (CRF) layer in the pre-constructed BiLSTM-CRF model to construct the state transition matrix. The Viterbi algorithm is used for decoding, and the optimal entity label sequence is output as the identified entity result. Based on the identified entities, a strategy combining dependency parsing and part-of-speech tagging and distance rules is used to extract semantic relationships between entities. The entity results and the semantic relationships between entities are input into the graph database in the form of triples to construct an industrial semantic knowledge graph for visualization, semantic retrieval and traceability analysis.

[0006] In one embodiment of the present invention, the multi-source heterogeneous text data includes: Process logs, production plans, quality inspection records, supply contracts, and maintenance documents.

[0007] In one embodiment of the present invention, when annotating keywords in training corpora in an industrial corpus, the entity categories annotated include: Suppliers, components, transactions, and services.

[0008] In one embodiment of the present invention, the sequence embedding vector includes: Word embedding vectors, character embedding vectors, and tag embedding vectors; The label embedding vector is jointly trained with a BiLSTM encoder through a dynamic semantic learning mechanism to improve the boundary recognition and semantic discrimination capabilities of named entity recognition.

[0009] In one embodiment of the present invention, the BiLSTM encoder is a multi-layer bidirectional long short-term memory network used to capture word contextual dependency information to improve robustness to semantic deformation in industrial statements.

[0010] In one embodiment of the present invention, semantic relations include: Supply relationships, manufacturing relationships, delivery relationships, and service response relationships; The structure of the triplet includes: First entity, relation type, second entity.

[0011] In one embodiment of the present invention, the graph database is Nebula Graph or Neo4j, used to store entity nodes and relation edges, supporting visualization and graph query operations.

[0012] In one embodiment of the present invention, the BiLSTM-CRF model includes: Input layer, encoding layer, decoding layer, and output layer; among which, The input layer concatenates the sequence embedding vectors to form a unified input matrix, which is then fed into the encoding layer. The encoding layer consists of two LSTM links, one forward and one backward, which simultaneously captures contextual information in the text, performs context sequence modeling, and obtains the hidden state information corresponding to each word in the text data. The decoding layer constructs a state transition matrix, uses the Viterbi algorithm to decode the hidden state information, and outputs the optimal entity label sequence as the identified entity result through the output layer.

[0013] In a second aspect, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the data entity identification and relation extraction method based on trusted data space provided by the present invention.

[0014] Thirdly, the present invention provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of a data entity identification and relation extraction method based on a trusted data space provided by the present invention.

[0015] The beneficial effects of this invention are: The solution provided by this invention constructs an industrial corpus based on multi-source heterogeneous text data in a trusted data space and introduces a tag embedding mechanism, which improves the semantic perception capability of the BiLSTM-CRF model, significantly outperforming traditional CRF and unidirectional LSTM models. By automatically extracting implicit supply, transaction, and manufacturing relationships between entities, a semantic graph is constructed, solving the problem of difficult structured modeling of semantic relationships in industrial data, facilitating subsequent visualization, traceability, and analysis. Furthermore, this invention supports linkage with the trusted data space architecture, performing semantic extraction processing without data leaving the domain, and combining access control, graph visualization, and other mechanisms to ensure data security and controllability, applicable to various typical scenarios such as automobile manufacturing, supply chain collaboration, and industrial operation and maintenance. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the steps of a data entity identification and relation extraction method based on a trusted data space provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the BiLSTM-CRF model in a data entity recognition and relation extraction method based on a trusted data space provided in an embodiment of the present invention; Figure 3 This is a visual representation of an industrial semantic knowledge graph constructed in a trusted data space using a data entity identification and relation extraction method based on a trusted data space, as provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0018] Below, we will first describe a data entity identification and relation extraction method based on a trusted data space, as provided in this embodiment of the invention. Figure 1 As shown, semantic structuring and semantic knowledge graph construction applicable to multi-source heterogeneous industrial data can include: S1. Collect multi-source heterogeneous text data from the trusted data space to construct an industrial corpus; annotate the keywords of the training corpus in the industrial corpus to obtain text sequences.

[0019] Specifically, embodiments of the present invention collect multi-source heterogeneous text data from a trusted data space in typical manufacturing scenarios. This multi-source heterogeneous text data may include: process logs, production plans, quality inspection records, supply contracts, and maintenance documents (e.g., production process records, purchase orders, quality inspection reports, supply agreements). After word segmentation, stop word removal, and specification reduction, the multi-source heterogeneous text data is used to construct an industrial corpus.

[0020] For different types of text, the BIO (Begin-Inside-Outside) or BMES (Begin-Middle-End-Single) annotation system is used to annotate the keywords in the training corpus. The entity categories to be annotated can include: supplier, component, action, and service.

[0021] For example, sample annotations are shown in Table 1: Table 1 Labeling Table

[0022] S2, vectorizes the text sequence into an input representation to obtain the sequence embedding vector.

[0023] Sequence embedding vectors may include: Word embedding vector, character embedding vector, and tag embedding vector.

[0024] For each word in the text sequence, construct the corresponding word embedding vector, character embedding vector, and label embedding vector, and concatenate them to form the final input vector of the model: ; in, Indicates the first i The final input vector corresponding to each word Indicates the first i The word embedding vectors corresponding to each word Indicates the first i The character embedding vectors corresponding to each word Indicates the first i The label embedding vector corresponding to each word.

[0025] BiLSTM-CRF model, such as Figure 2 As shown, it may include: The system consists of an input layer, an encoding layer, a decoding layer, and an output layer. The input layer is responsible for converting the original text sequence into a computable vector representation, the encoding layer is responsible for extracting contextual semantic features, the decoding layer is responsible for modeling the transition constraints between labels, and the output layer provides the final entity annotation results.

[0026] The input layer concatenates the sequence embedding vectors to form a unified input matrix, which is then fed into the encoding layer; The encoding layer consists of two LSTM links, one forward and one backward, which simultaneously captures contextual information in the text, performs context sequence modeling, and obtains the hidden state information corresponding to each word in the text data. The decoding layer constructs a state transition matrix, uses the Viterbi algorithm to decode the hidden state information, and outputs the optimal entity label sequence as the identified entity result through the output layer.

[0027] This invention embodiment constructs multiple embeddings at the input layer, which may include: Word embedding vectors: Word2Vec word vectors trained on corpus; Character embedding vectors: used to identify out-of-vocabulary words or abbreviations; Label embedding vector: dynamically represents the semantic features of the label and assists in boundary discrimination.

[0028] The embedding vectors are concatenated at the input layer to form a unified input matrix, which is then fed into the encoding layer of the BiLSTM-CRF model.

[0029] S3, the label embedding vector is jointly trained with the BiLSTM encoder through a dynamic semantic learning mechanism to improve the boundary recognition and semantic discrimination capabilities of named entity recognition.

[0030] In step S3, the sequence embedding vector obtained in step S2 is input into the BiLSTM encoder for context sequence modeling. The sequence embedding vector is composed of word embedding vector, character embedding vector, and label embedding vector. The word embedding vector is used to represent word-level semantic information, the character embedding vector is used to represent the character composition features inside the word and enhance the recognition ability of out-of-vocabulary words and abbreviations, and the label embedding vector is used to provide label semantic auxiliary information. The above three types of vectors are input into the BiLSTM encoder along with the unified input sequence, and the hidden state information corresponding to each word is obtained through forward and backward bidirectional modeling.

[0031] The input sequence vector is sequentially fed into the BiLSTM encoder to generate a context representation for each word, and the output is: ; in, This indicates a positive LSTM output. This indicates the inverse LSTM output.

[0032] The aforementioned sequence embedding vectors are sequentially input into a BiLSTM encoder, which consists of two LSTM links (forward and backward) to simultaneously capture contextual information in the text. The output at each position is the deep semantic representation of the word at that position within the overall sentence context. This step ensures effective learning of the implicit relationships between upstream and downstream words in industrial corpora. The BiLSTM encoder is a multi-layer bidirectional long short-term memory network used to capture word context-dependent information, thereby improving robustness to semantic deformations in industrial sentences.

[0033] S4. The hidden state information is input into the Conditional Random Field (CRF) layer in the pre-built BiLSTM-CRF model to construct the state transition matrix. The Viterbi algorithm is used for decoding, and the optimal entity label sequence is output as the identified entity result.

[0034] The encoded hidden state information is used as a hidden state sequence and fed into the Conditional Random Field (CRF) decoding output layer to construct a label state transition matrix, optimizing the overall prediction path. The Viterbi algorithm is then used to decode the hidden state sequence, outputting the optimal entity label sequence as the identified entity result, thus achieving automatic named entity recognition. "P-COMPONENT" indicates that a "component-type entity" has been identified.

[0035] Understandably, embodiments of the present invention utilize the output layer of a Conditional Random Field (CRF) decoder to learn the state transition patterns between entities and construct a transition matrix. The overall sequence scoring function is: ; in, This represents the input text sequence. Represents the input sequence A corresponding label sequence. Indicates label From the transfer The transition score, which comes from the state transition matrix learned by the CRF. , Indicates the first i The word was tagged as The probability score, This indicates the length of the input sequence.

[0036] In this embodiment of the invention, the optimal label path is finally obtained through the Viterbi algorithm to achieve named entity recognition.

[0037] S5, based on the identified entity results, uses dependency parsing combined with a strategy based on part-of-speech and distance rules to extract semantic relationships between entities.

[0038] Building upon named entity recognition, dependency parsing combined with part-of-speech and context window rules is used to extract semantic relationships between entities. These semantic relationships can include: Supply relationships, manufacturing relationships, delivery relationships, and service response relationships.

[0039] For example: <Supplier A, supply_to, component_13>; <Component_13,installed_in,Product_E> .

[0040] The structure of a triplet can include: First entity, relation type, second entity.

[0041] For example: "Supplier A", "Provides", "Part X".

[0042] S6 inputs entity results and semantic relationships between entities into a graph database in the form of triples to construct an industrial semantic knowledge graph for visualization, semantic retrieval, and traceability analysis.

[0043] The extracted entity results and the semantic relationships between entities are constructed into triples and imported into a graph database (such as Neo4j).

[0044] The graph database is Nebula Graph or Neo4j, used to store entity nodes and relationship edges, supporting visualization and graph query operations.

[0045] The model training provided in this embodiment of the invention uses the PyTorch framework, the optimizer is Adam, and the total loss function is... CRF loss function Loss compared to tag embedding The weighted combination of the three elements. In this embodiment of the invention, the labeled training text input model is completed by constructing a word embedding vector, a character embedding vector, and a label embedding vector for each word, and then concatenating them to form a unified input vector sequence. Subsequently, the unified input vector sequence is input to a BiLSTM encoder to obtain the context hidden state representation corresponding to each position. Then, the hidden state representation is input to a CRF decoding layer to calculate the conditional probability distribution corresponding to the target label sequence, and based on this, the CRF loss function is obtained. Simultaneously, a contrastive learning constraint is introduced into the label embedding vectors. Specifically, the embedding representations corresponding to labels of the same class are used as positive sample pairs, and the embedding representations corresponding to labels of different classes are used as negative sample pairs. By measuring the representation distance between different label embeddings, a label embedding contrastive loss function is constructed. According to the preset weighting coefficient We obtain the total loss function by weighted summation of the two: ; in, These are hyperparameters used to balance the two loss terms, preferably... You can take 0.1.

[0046] A visual representation of an industrial semantic knowledge graph, such as... Figure 3As shown, entity nodes in the graph are colored by category, and different relationships are connected by directed edges, supporting semantic querying and tracing. This graph uses nodes to represent entities and edges to represent relationships, supporting semantic reasoning and data navigation operations in various industrial scenarios, including structured queries, semantic association, and path tracing. Entities such as "Supplier G", "Component_13", and "Model E" are distinguished by color; semantic connections between entities, such as "supply_to" and "assembled_in", are indicated by edges in the graph, intuitively expressing semantic paths such as multi-level supply, component ownership, and fault propagation.

[0047] Figure 3 Nodes of different colors represent different categories of industrial entities, which are connected by various types of directed edges, forming a complex semantic relationship network. Through this graph, users can clearly understand the supply chain, component traceability paths, and key interaction nodes, providing important support for data sharing decisions and anomaly tracing.

[0048] Experimental results show that the data entity recognition and relation extraction method provided in this embodiment of the invention can achieve an entity recognition accuracy of over 94% on real and reliable data space data, and can support the construction of an industrial knowledge graph containing multi-level supply and assembly semantics.

[0049] Compared with existing technologies, the data entity recognition and relation extraction method provided in this invention has the following advantages: The data entity recognition and relation extraction method constructs a corpus based on the actual industrial data context in the trusted data space and introduces a label embedding mechanism, which improves the semantic perception capability of the BiLSTM-CRF model and is significantly better than the traditional CRF and unidirectional LSTM models.

[0050] The data entity identification and relationship extraction method automatically extracts the implicit supply, transaction, and manufacturing relationships between entities to construct a semantic graph, which solves the problem of difficult structured modeling of semantic relationships in industrial data, and facilitates subsequent visualization, traceability and analysis.

[0051] The data entity recognition and relation extraction method supports linkage with the trusted data space architecture, performs semantic extraction processing without leaving the domain, and combines access control, graph visualization and other mechanisms to ensure data security and controllability. It is applicable to a variety of typical scenarios such as automobile manufacturing, supply chain collaboration and industrial operation and maintenance.

[0052] The above is a description of the data entity identification and relation extraction method provided in the embodiments of the present invention from the perspective of data mining. It should be particularly noted that: The steps of identifying entities and extracting relationships from the trusted data space in this embodiment of the invention, namely steps S5-S6 above, can also be used as an independent module for semantic association analysis and knowledge graph construction of other structured or semi-structured industrial data, thereby improving the capabilities of enterprises in scenarios such as data understanding, quality traceability and intelligent decision-making.

[0053] Therefore, in addition to the data entity recognition and relation extraction methods, this embodiment of the invention also provides a named entity recognition method based on a trusted data space and an entity relation extraction method based on a trusted data space. The above two methods have been fully described in the overall process of this embodiment of the invention, and will not be repeated here.

[0054] Secondly, embodiments of the present invention also provide an electronic device, such as... Figure 4 As shown, it includes a processor 001, a communication interface 002, a memory 003, and a communication bus 004, wherein the processor 001, the communication interface 002, and the memory 003 communicate with each other through the communication bus 004. The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of any of the data entity identification and relation extraction methods based on trusted data space provided in the first aspect of the present invention.

[0055] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0056] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0057] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0058] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0059] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0060] Thirdly, corresponding to the data entity identification and relation extraction method based on trusted data space provided in the first aspect, this embodiment of the invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the data entity identification and relation extraction methods based on trusted data space provided in the first aspect of this invention.

[0061] For the electronic device / storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant details can be found in the description of the method embodiments.

[0062] It should be noted that the electronic device and storage medium in the embodiments of the present invention are respectively electronic devices and storage media that apply the above-described data entity identification and relation extraction method based on trusted data space. Therefore, all embodiments of the above method are applicable to the electronic device and storage medium, and can achieve the same or similar beneficial effects.

[0063] It should be noted that, in the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for data entity identification and relation extraction based on a trusted data space, characterized in that, include: Collect multi-source heterogeneous text data from trusted data spaces to construct an industrial corpus; Text sequences are obtained by annotating keywords in the training corpus of the industrial corpus; The text sequence is vectorized into an input representation to obtain a sequence embedding vector; The sequence embedding vector is input into the BiLSTM encoder in the pre-built BiLSTM-CRF model to perform context sequence modeling and obtain the hidden state information corresponding to each word in the text data. The hidden state information is input into the conditional random field CRF decoding output layer of the pre-built BiLSTM-CRF model to construct the state transition matrix. The Viterbi algorithm is used for decoding, and the optimal entity label sequence is output as the identified entity result. Based on the identified entities, a strategy combining dependency parsing and part-of-speech tagging and distance rules is used to extract semantic relationships between entities. The entity results and the semantic relationships between entities are input into the graph database in the form of triples to construct an industrial semantic knowledge graph for visualization, semantic retrieval and traceability analysis.

2. The method for data entity identification and relation extraction based on a trusted data space according to claim 1, characterized in that, The multi-source heterogeneous text data includes: Process logs, production plans, quality inspection records, supply contracts, and maintenance documents.

3. The method for data entity identification and relation extraction based on a trusted data space according to claim 1, characterized in that, When annotating keywords in training corpora in an industrial corpus, the entity categories annotated include: Suppliers, components, transactions, and services.

4. The method for data entity identification and relation extraction based on a trusted data space according to claim 1, characterized in that, The sequence embedding vector includes: Word embedding vectors, character embedding vectors, and tag embedding vectors; The label embedding vector is jointly trained with a BiLSTM encoder through a dynamic semantic learning mechanism to improve the boundary recognition and semantic discrimination capabilities of named entity recognition.

5. The method for data entity identification and relation extraction based on a trusted data space according to claim 1, characterized in that, The BiLSTM encoder is a multi-layer bidirectional long short-term memory network used to capture word contextual dependency information to improve robustness to semantic deformation in industrial statements.

6. The method for data entity identification and relation extraction based on a trusted data space according to claim 1, characterized in that, The semantic relationships include: Supply relationships, manufacturing relationships, delivery relationships, and service response relationships; The structure of the triplet includes: First entity, relation type, second entity.

7. The method for data entity identification and relation extraction based on a trusted data space according to claim 1, characterized in that, The graph database is Nebula Graph or Neo4j, used to store entity nodes and relationship edges, and supports visualization and graph query operations.

8. The method for data entity identification and relation extraction based on a trusted data space according to claim 1, characterized in that, The BiLSTM-CRF model includes: Input layer, encoding layer, decoding layer, and output layer; among which, The input layer concatenates the sequence embedding vectors to form a unified input matrix, which is then fed into the encoding layer. The encoding layer consists of two LSTM links, one forward and one backward, which simultaneously captures contextual information in the text, performs context sequence modeling, and obtains the hidden state information corresponding to each word in the text data. The decoding layer constructs a state transition matrix, uses the Viterbi algorithm to decode the hidden state information, and outputs the optimal entity label sequence as the identified entity result through the output layer.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the data entity identification and relation extraction method based on a trusted data space as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the data entity identification and relation extraction method based on a trusted data space as described in any one of claims 1-8.

Images