Semi-supervised information extraction method, system, equipment and medium

Abstract

The invention discloses a semi-supervised information extraction method, system and device and a medium. The method comprises the steps of obtaining a to-be-extracted document and an ontology class; segmenting the document to be extracted to form a plurality of sentences; calculating the semantic similarity between each entity in each sentence and the ontology class, and taking the entity with the highest semantic similarity as an entity candidate of the sentence; the entity candidates of all sentences are input into the trained relation extraction model, the entity relation and attributes in the document to be extracted are obtained, the trained relation extraction model is formed by training a data set obtained based on a remote supervision strategy, the method, the system, the equipment and the medium can extract the entity relation in the engineering management document, and the extraction efficiency is improved. And the method has the characteristics of low cost, high accuracy and high characterization capability.

Description

Technical Field

[0001] The present invention belongs to the technical field of information extraction, and relates to a semi-supervised information extraction method, system, equipment and medium. Background Art

[0002] All engineering (mainly including construction and manufacturing) management relies on the collection and analysis of project information. A large amount of project information is hidden in unstructured text files such as plans and manuals. This information is very valuable for management activities and can be used for quality inspection and compliance inspection.

[0003] At present, there are still the following problems when extracting information in the field of engineering management: Data labeling is difficult and expensive: Existing entity relationship extraction models usually rely on supervised learning methods and require a large amount of labeled data to train the model. However, obtaining high-quality labeled data in a specific field is an expensive and time-consuming task. Labeling data requires a lot of manpower and material resources, which is not economical for enterprises.

[0004] Traditional remote supervision introduces a lot of noise: Although the introduction of remote supervision can alleviate the need for large-scale manual intervention, it still relies on an external large-scale knowledge base. Manually building a knowledge base in a specific field will also consume a lot of manpower and material resources, especially since there is no way to build a knowledge base that can cover enough semantically similar entities. Without a complete and specialized knowledge base, the generated labels are inaccurate or wrong, introducing noise. At the same time, its existence will not only directly affect the training and performance of the model, but also affect the accuracy of subsequent entity relationship extraction tasks. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a semi-supervised information extraction method, system, device and medium, which can extract entity relationships in engineering management documents and have the characteristics of low cost and high accuracy.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: In one aspect of the present invention, the semi-supervised information extraction method of the present invention comprises: Get the documents and ontology classes to be extracted for construction project management; Segmenting the document to be extracted to form a plurality of sentences; Calculate the semantic similarity between each entity and the ontology class in each sentence, and take the entity with the highest semantic similarity as the entity candidate of the sentence; The entity candidates of each sentence are input into the trained relationship extraction model to obtain the entity relationships and attributes in the document to be extracted, wherein the trained relationship extraction model is trained based on the data set annotated by the remote supervision strategy.

[0007] The semi-supervised information extraction method of the present invention is further improved in that: Furthermore, the process of calculating the semantic similarity between each entity and the ontology class in each sentence and taking the entity with the highest semantic similarity as the entity candidate of the sentence is as follows: For each sentence, the semantic similarity between the entity in the sentence and the ontology class is calculated through double loop and cosine similarity, and then the entity with the highest semantic similarity is determined through the maximum pooling operation, and the determined entity is used as the entity candidate of the sentence.

[0008] Furthermore, before the entity candidates of each sentence are input into the trained relation extraction model, the following steps are also included: Obtain documentation for construction project management; Segmenting the construction project management document to form a number of sentence samples; Performing data enhancement on the sentence samples, and constructing a sentence pool using the sentence samples obtained after the data enhancement; Use sentence samples in the sentence pool to construct training sets, validation sets, and test sets; Build a relationship extraction model; The relationship extraction model is trained, verified and tested using the training set, validation set and test set to obtain a trained relationship extraction model.

[0009] Furthermore, the process of constructing a training set, a validation set and a test set using sentence samples in the sentence pool is as follows: A remote supervision strategy is used to annotate sentence samples in the sentence pool to form the dataset CONSD 1; The Ont4RE (ontology for relation extraction) method is used to annotate the sentence samples in the sentence pool to form the dataset CONSD 2; Stratified sampling is performed on the datasets CONSD 1 and CONSD 2 respectively, and the annotated sentence samples obtained by stratified sampling are used to construct the training set, validation set and test set.

[0010] Furthermore, the loss function in the relationship extraction model training process is:

[0011] Where NB is the number of bags, For the prediction package For the The probability of a relationship type, Extract all trainable parameters in the model for the relation.

[0012] Furthermore, the relation extraction model includes a connected Transformer-based encoder and a relation classifier.

[0013] Furthermore, the process of performing data enhancement on the sentence sample is: ChatGPT (Chat Generative Pre-trained Transformer) is used to perform data enhancement on the sentence samples.

[0014] In a second aspect of the present invention, the semi-supervised information extraction system of the present invention comprises: The first acquisition module is used for acquiring documents and ontology classes to be extracted for construction project management; A first segmentation module, used for segmenting the document to be extracted to form a plurality of sentences; A calculation module is used to calculate the semantic similarity between each entity and the ontology class in each sentence, and take the entity with the highest semantic similarity as the entity candidate of the sentence; The extraction module is used to input the entity candidates of each sentence into the trained relationship extraction model to obtain the entity relationships and attributes in the document to be extracted, wherein the trained relationship extraction model is trained based on the data set annotated by the remote supervision strategy.

[0015] In a third aspect of the present invention, the computer device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the semi-supervised information extraction method when executing the computer program.

[0016] In a fourth aspect of the present invention, the computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the steps of the semi-supervised information extraction method are implemented.

[0017] The present invention has the following beneficial effects: The semi-supervised information extraction method, system, device and medium of the present invention input entity candidates of each sentence into the trained relation extraction model during specific operation, and obtain entity relations and attributes in the document to be extracted. Through the relation extraction model, the semantic and context information in the text data can be better captured to improve the accuracy and representation ability of extraction. At the same time, the trained relation extraction model is trained based on the data set obtained by the remote supervision strategy, which alleviates the pressure and cost of manually constructing the data set, and is highly practical. At the same time, the ontology, information extraction based on remote supervision and relation extraction model are combined, thereby improving data quality and enhancing model performance. According to experiments, the present invention improves AUPRC (Area Under the Precision-Recall Curve) by 6.1% and F1-score by 11.7% compared with traditional remote supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a training flow chart of the relationship extraction model in the present invention; Figure 3 The flowchart of the construction of the training set, the validation set and the test set in the present invention; Figure 4 is a system structure diagram of the present invention; Figure 5 This is a structural diagram of the first building block in the present invention. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only an embodiment of a part of the present invention, not all embodiments, and is not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concepts disclosed in the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0020] The accompanying drawings show schematic diagrams of structures according to embodiments disclosed in the present invention. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0021] Embodiment 1 refer to Figure 1 , the semi-supervised information extraction method of the present invention comprises: 1) Get the documents and ontology classes to be extracted; It should be noted that Stanford has developed a standard ontology establishment process [Noy, NF, & McGuinness, DL (2001). Ontology development 101: A guide to creating your first ontology.], which is as follows: 11) Determine the scope and objectives of the ontology, and construct a knowledge model about construction and manufacturing engineering management. The ontology covers various entities in the field of construction and manufacturing engineering management and the attribute relationships between entities. The various entities include people, machines, materials and environment. The goal is to achieve automatic and accurate annotation of the corpus by establishing a domain ontology as an external knowledge base for remote supervision of entity relationship extraction; 12) List important terms and determine the four core subject categories of construction and manufacturing engineering management. The four core subject categories are people, machines, materials and environment. On this basis, determine the attribute categories of multiple subject categories, such as the relationship between fine categories (such as safety officers) and coarse-grained categories (construction and manufacturing personnel), where the relationship between coarse-grained categories is defined as has-parent relationship, 13) Define the class and class hierarchy. The four core subject categories are organized into a tree-like classification system, which contains four levels. For example, the large category "material" includes more detailed subcategories, such as "steel materials", "concrete materials" and "adhesive materials"; and the category of "steel materials" can be further subdivided into subcategories such as "rebar" and "steel plates". Secondly, determine multiple attribute categories of the subject categories, which can be divided into geometric and non-geometric categories, forming another branch in the classification system. For example, the geometric class contains attributes such as length, width and area, and the non-geometric class contains attributes such as speed, weight and quantity. 14) Define the properties and instances of the class. Each class has its own specific properties. For example, for the material class, its properties include time, pressure, and quantity. 15) Define the relationship between classes. For two classes, there may be multiple relationships between the two classes. The format of the definition is "has-property". When two classes have a temperature property, the relationship is defined as "has-temp". 16) Create an instance. After defining the classes, relationships, and properties, combine the classes and properties with relationships, as well as the relationships between classes and properties, to form a triple of <class, relationship, property>. For example, a machine has a weight property, and its triple is represented as <machine, has-weight, weight>.

[0022] 2) Segmenting the document to be extracted to form a number of sentences; 3) Calculate the semantic similarity between each entity and the ontology class in each sentence, and take the entity with the highest semantic similarity as the entity candidate of the sentence; 4) Inputting the entity candidates of each sentence into the trained relationship extraction model to obtain the entity relationship in the document to be extracted, wherein the trained relationship extraction model is trained based on the data set obtained by the remote supervision strategy.

[0023] refer to Figure 2 and Figure 3 , the process of step 4) is: 41) Obtain documents for construction project management; 42) Segmenting the construction project management document to form a plurality of sentence samples; 43) performing data enhancement on the sentence samples, and constructing a sentence pool using the sentence samples obtained after the data enhancement; Specifically, documents on construction and manufacturing engineering management are collected from online public databases and partner company file systems, and the documents are screened and manually reviewed to obtain work plans, manuals, and standard documents. The contents of the work plans, manuals, and standard documents are cut into sentences, and the average length and variance of each sentence are calculated. Then, Sentence-Transformer is used to convert the sentences into semantically embedded vectors, and ChatGPT is used to enhance the semantically embedded vectors to increase the number of sentences.

[0024] In addition, the ontology of construction and manufacturing engineering management is represented by triples. Based on the established ontology structure, the triples include a head, a relationship, and a tail. For example, the Rebar class has the Weight class as an attribute, and the triple is represented as Rebar has-weight Weight. Therefore, each triple includes a subject class as a head, an entity-attribute relationship as a relationship, and an attribute class as a tail, that is, <class, relationship, attribute>.

[0025] It should also be noted that the present invention uses ChatGPT for data enhancement. Specifically, after collecting the original CEM documents of the sentences, it is found that the results of remote supervision annotation have a long tail effect, which is a common problem in actual data. To solve this problem, the text generation capability of ChatGPT is used to enhance the sentence pool. Specifically, the existing sentences and entity pairs are used to fill the prompts of ChatGPT, and it is required to generate synonymous sentences with consistent semantics but different expressions. The whole process is supervised by experts to ensure that the generated sentences are related to construction and manufacturing engineering management and are not repeated. Through the above enhancement method, the number of sentences can be expanded.

[0026] 44) The sentence samples in the sentence pool are labeled using the remote supervision strategy to form the dataset CONSD 1; the sentence samples in the sentence pool are annotated using the Ont4RE method to form the dataset CONSD 2; It should be noted that after establishing the ontology and data preprocessing, the sentence is divided into a series of tokens , and then converted into a vector through the Sentence-Transformer model , and then use the double loop and cosine similarity formula, namely:

[0027] The semantic similarity between the entity and the ontology class is calculated, and then the most similar class is determined through the maximum pooling operation. The entities corresponding to the most similar class are taken as entity candidates. Based on the identified entity candidates, the sentences are automatically annotated through heuristic rules. That is, when there is a relationship between the classes of two candidate entities in the ontology knowledge base, the sentence is annotated using this relationship. Finally, the annotated data is filtered and verified, and a limited number of samples are extracted from the sentence pool. The entity-attribute relationship of the samples is manually annotated, and the annotation is regarded as a reference benchmark. Subsequently, the Ont4RE method and the remote supervision strategy are used to annotate the same limited number of samples, respectively. At the same time, the long-tail effect is considered to balance the distribution of different relationship types in the selected samples to ensure the fairness of the evaluation.

[0028] 45) Perform stratified sampling on the datasets CONSD 1 and CONSD 2 respectively, and use the annotated sentence samples obtained by stratified sampling to construct the training set, validation set, and test set in a ratio of 7:2:1; 46) Build a relationship extraction model; It should be noted that the relationship extraction model includes a Transformer-based encoder and a relationship classifier.

[0029] 47) Using the training set, validation set and test set to train, validate and test the relationship extraction model to obtain a trained relationship extraction model.

[0030] Specifically, when two candidate entities are known, the extraction of entity attributes can be regarded as a relation classification task, and the sentence is segmented into tokens of fixed length each. , passed into the Transformer encoder, embedded into its three components, forming semantic embedding vectors , sentence boundary embedding vector and position embedding vector , the Transformer-based encoder consists of The encoder is composed of multiple self-attention layers, which can fully capture the context and understand the key components of deep semantics. For each attention head in each layer, the input By respectively combining with three other trainable matrices , and Multiply and convert to three matrices Q, K and V, where:

[0031] After integrating the matrices Q, K, and V, a self-attention operation is performed to generate the attention value of the entire sequence. for:

[0032] in, For the matrix The transposed matrix of Represents the dimension of the vector; Since multiple attention heads are used, the aggregate value is obtained by concatenating the results from different heads and applying a linear transformation to them :

[0033] in, is a learnable constant vector, , H is the number of heads, is a trainable matrix. In documents on construction and manufacturing engineering management, a specific entity can appear in multiple sentences, but these sentences may contain different relationships between entities. A bag-based approach is used to classify the relationships. Sentences with the same entity pair are output through full connection. , put them in the same package, for each package , using the self-attention mechanism, in the sentence center, attention for:

[0034] Assign a weight to each sentence, where Indicates The sum of the cosine similarities between the sentence and other sentences, the representation of the bag is the sum of the sentence representations weighted by the attention value. Specifically, for:

[0035] in, is the semantic vector of the i-th sentence; Get package representation After that, calculate With relational embedding The similarity between ,in, is the semantic vector of the relationship, Indicates package, express With relational embedding The similarity between is a bias vector, ∈ , The largest number of relationship types is selected as the prediction result, and then the softmax function is used to calculate the prediction package For the The probability of a relationship type :

[0036] in, For the The score of a relation after attention, For the The score of a relation after attention, Extract all trainable parameters in the relationship model and define cross entropy to calculate the loss, where the loss function is:

[0037] in, Indicates the number of bags, For the prediction package For the The probability of a relationship type, Extract all trainable parameters in the model for the relation.

[0038] Embodiment 2 refer to Figure 4 The semi-supervised information extraction system of the present invention comprises: The first acquisition module is used to acquire the document to be extracted and the ontology class; A first segmentation module, used for segmenting the document to be extracted to form a plurality of sentences; A calculation module is used to calculate the semantic similarity between each entity and the ontology class in each sentence, and take the entity with the highest semantic similarity as the entity candidate of the sentence; The extraction module is used to input the entity candidates of each sentence into the trained relationship extraction model to obtain the entity relations and attributes in the document to be extracted, wherein the trained relationship extraction model is trained based on the data set obtained by the remote supervision strategy.

[0039] As an implementation mode of the present invention, this embodiment also includes: The second acquisition module is used to acquire documents of construction project management; A second segmentation module is used to segment the construction project management document to form a plurality of sentence samples; An enhancement module, used to perform data enhancement on the sentence samples, and construct a sentence pool using the sentence samples obtained after data enhancement; The first construction module is used to construct a training set, a validation set and a test set using sentence samples in the sentence pool; The second building module is used to build a relationship extraction model; The training module is used to train, verify and test the relationship extraction model using the training set, verification set and test set to obtain a trained relationship extraction model.

[0040] As an embodiment of the present invention, reference Figure 5 , the first building block includes: The first annotation module is used to annotate the sentence samples in the sentence pool using a remote supervision strategy to form the dataset CONSD 1; The second annotation module is used to annotate the sentence samples in the sentence pool using the Ont4RE method to form the dataset CONSD 2; The sampling module is used to perform stratified sampling on the datasets CONSD 1 and CONSD 2 respectively, and use the annotated sentence samples obtained by stratified sampling to construct the training set, validation set and test set.

[0041] As an implementation mode of the present invention, the process of calculating the semantic similarity between each entity and the ontology class in each sentence and taking the entity with the highest semantic similarity as the entity candidate of the sentence is as follows: For each sentence, the semantic similarity between the entity in the sentence and the ontology class is calculated through double loop and cosine similarity, and then the entity with the highest semantic similarity is determined through the maximum pooling operation, and the determined entity is used as the entity candidate of the sentence.

[0042] As an implementation mode of the present invention, the loss function in the relationship extraction model training process is:

[0043] Where NB represents the number of bags, For the prediction package For the The probability of a relationship type, Extract all trainable parameters in the model for the relation.

[0044] As an implementation mode of the present invention, the relationship extraction model includes a connected Transformer-based encoder and a relationship classifier.

[0045] As an implementation manner of the present invention, the process of performing data enhancement on the sentence sample is: ChatGPT is used to perform data enhancement on the sentence samples.

[0046] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present application may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0047] Embodiment 3 A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the semi-supervised information extraction method when executing the computer program, for example, comprising: obtaining a document on construction project management; segmenting the document on construction project management to form a plurality of sentence samples; performing data enhancement on the sentence samples, and constructing a sentence pool using the sentence samples obtained after data enhancement; constructing a training set, a verification set and a test set using the sentence samples in the sentence pool; constructing a relationship extraction model; training, verifying and testing the relationship extraction model using the training set, the verification set and the test set to obtain a trained relationship extraction model; obtaining a document to be extracted and an ontology class; segmenting the document to be extracted to form a plurality of sentences; calculating the semantic similarity between each entity and the ontology class in each sentence, and taking the entity with the highest semantic similarity as the entity candidate of the sentence; inputting the entity candidate of each sentence into the trained relationship extraction model to obtain the entity relationship and attributes in the document to be extracted, wherein the trained relationship extraction model is trained based on a data set obtained by a remote supervision strategy. The memory may include a memory, such as a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk memory, etc. The processor, network interface, and memory are interconnected through an internal bus, and the internal bus may be an industrial standard architecture bus, a peripheral component interconnection standard bus, an extended industrial standard architecture bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory is used to store programs. Specifically, the program may include a program code, and the program code includes computer operation instructions. The memory may include a memory and a non-volatile memory, and provide instructions and data to the processor.

[0048] Embodiment 4 A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of implementing the semi-supervised information extraction method include: obtaining a document of construction project management; segmenting the document of construction project management to form a plurality of sentence samples; performing data enhancement on the sentence samples, and constructing a sentence pool using the sentence samples obtained after data enhancement; constructing a training set, a verification set, and a test set using the sentence samples in the sentence pool; constructing a relationship extraction model; training, verifying, and testing the relationship extraction model using the training set, the verification set, and the test set to obtain a trained relationship extraction model; obtaining a document to be extracted and an ontology class; segmenting the document to be extracted to form a plurality of sentences; calculating the semantic similarity between each entity and the ontology class in each sentence, and taking the entity with the highest semantic similarity as the entity candidate of the sentence; inputting the entity candidate of each sentence into the trained relationship extraction model to obtain the entity relationship and attributes in the document to be extracted, wherein the trained relationship extraction model is trained based on a data set obtained by a remote supervision strategy. Specifically, the computer-readable storage medium includes, but is not limited to, for example, a volatile memory and / or a non-volatile memory. The volatile memory may include a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include a read-only memory (ROM), a hard disk, a flash memory, an optical disk, a magnetic disk, etc.

[0049] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0050] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0051] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A semi-supervised information extraction method, characterized in that: include: Get the documents and ontology classes to be extracted for construction project management; Segmenting the document to be extracted to form a plurality of sentences; Calculate the semantic similarity between each entity and the ontology class in each sentence, and take the entity with the highest semantic similarity as the entity candidate of the sentence; The entity candidates of each sentence are input into the trained relationship extraction model to obtain the entity relationships and attributes in the document to be extracted, wherein the trained relationship extraction model is trained based on the data set annotated by the remote supervision strategy.

2. The semi-supervised information extraction method according to claim 1, characterized in that: The process of calculating the semantic similarity between each entity and the ontology class in each sentence and taking the entity with the highest semantic similarity as the entity candidate of the sentence is as follows: For each sentence, the semantic similarity between the entity in the sentence and the ontology class is calculated through double loop and cosine similarity, and then the entity with the highest semantic similarity is determined through the maximum pooling operation, and the determined entity is used as the entity candidate of the sentence.

3. The semi-supervised information extraction method according to claim 1, characterized in that: Before the entity candidates of each sentence are input into the trained relation extraction model, the following are also included: Obtain documentation for construction project management; Segmenting the construction project management document to form a number of sentence samples; Performing data enhancement on the sentence samples, and constructing a sentence pool using the sentence samples obtained after the data enhancement; Use sentence samples in the sentence pool to construct training sets, validation sets, and test sets; Build a relationship extraction model; The relationship extraction model is trained, verified and tested using the training set, validation set and test set to obtain a trained relationship extraction model.

4. The semi-supervised information extraction method according to claim 3, characterized in that: The process of constructing the training set, the validation set and the test set using the sentence samples in the sentence pool is as follows: A remote supervision strategy is used to annotate sentence samples in the sentence pool to form the dataset CONSD 1; The Ont4RE method is used to annotate the sentence samples in the sentence pool to form the dataset CONSD 2; Stratified sampling is performed on the datasets CONSD 1 and CONSD 2 respectively, and the annotated sentence samples obtained by stratified sampling are used to construct the training set, validation set and test set.

5. The semi-supervised information extraction method according to claim 3, characterized in that: The loss function in the relationship extraction model training process is: Where NB is the number of bags, For the prediction package For the The probability of a relationship type, Extract all trainable parameters in the model for the relation.

6. The semi-supervised information extraction method according to claim 3, characterized in that: The relation extraction model includes a connected Transformer-based encoder and a relation classifier.

7. The semi-supervised information extraction method according to claim 3, characterized in that: The process of performing data enhancement on the sentence sample is as follows: ChatGPT is used to perform data enhancement on the sentence samples.

8. A semi-supervised information extraction system, characterized in that: include: The first acquisition module is used to acquire the documents and ontology classes to be extracted for construction engineering management; A first segmentation module, used for segmenting the document to be extracted to form a plurality of sentences; A calculation module is used to calculate the semantic similarity between each entity and the ontology class in each sentence, and take the entity with the highest semantic similarity as the entity candidate of the sentence; The extraction module is used to input the entity candidates of each sentence into the trained relationship extraction model to obtain the entity relationships and attributes in the document to be extracted, wherein the trained relationship extraction model is trained based on the data set annotated by the remote supervision strategy.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the semi-supervised information extraction method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the semi-supervised information extraction method according to any one of claims 1 to 7 are implemented.

Images