An intelligent annotation method and system for semantic structural units in medical texts

Abstract

The present invention provides a method and system for intelligent annotation of semantic structural units in medical texts, and relates to the field of data processing technology. The method includes S1: obtaining a pre-trained large language model, collecting medical text data containing annotation information and processing it into an adaptive format, and constructing an automatic recognition model for semantic structural units through instruction fine-tuning. S2: Integrating the automatic recognition model with the text annotation tool to form a human-computer interactive annotation tool, automatically annotating medical text data, displaying the annotation information on a web page, and organizing experts to manually proofread. S3: Collecting proofread data, and when the new data reaches a preset quantity threshold or the model evaluation index meets the preset conditions, retraining the model to generate a new recognition model, and returning to execute S2. The present invention has the advantages of data-driven, self-evolution and human-computer collaboration, reduces maintenance costs with the help of a large language model, enhances model performance through a self-learning mechanism, and uses a human-computer collaborative strategy to solve large model algorithm problems and improve annotation efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an intelligent annotation method and system for semantic structural units in medical texts. Background Art

[0002] In the medical field, medical texts contain a vast amount of professional knowledge, which is crucial for disease diagnosis, treatment plan formulation, medical research, etc. However, there are many problems with medical texts, among which the low degree of structuring and standardization is more prominent. Medical knowledge is usually expressed in natural language, containing a large number of professional terms, complex sentence structures, and professional clinical knowledge expressions, which makes it difficult for computers to directly understand and use this information at the big data level. Therefore, accurate and efficient representation and extraction of medical knowledge in medical texts has become the key to solving this problem.

[0003] In previous studies, Chinese patent CN112270965A disclosed a method for semantically structuring phenotypic information in medical texts, and proposed a new concept of medical knowledge representation, "semantic structural unit", which aims to more accurately represent the knowledge in medical texts through the organic combination of core words and attribute words in medical ontology. At the same time, a method for extracting semantic structural units based on dictionaries, knowledge bases, and rule bases is also proposed. This method first identifies phenotypic concepts in medical texts based on dictionaries, then uses knowledge bases and rule bases in combination with context to identify the associated attributes of phenotypic concepts, and finally uses the BRAT text annotation tool to interactively proofread the recognition results, thereby completing the extraction and annotation of semantic structural units.

[0004] However, with the continuous development of information technology, this traditional method has gradually exposed many drawbacks. On the one hand, it requires the maintenance of a huge medical dictionary, knowledge base and rule base. The knowledge in the medical field is updated rapidly, the types of diseases are diverse and new symptoms continue to emerge, which requires the continuous updating of dictionaries, knowledge bases and rule bases. From the actual processing of real-world corpora, its adaptability and flexibility for certain specialist or disease-specific corpora are poor, and it is difficult to meet the complex and changeable medical text processing needs. For example, in the face of medical texts on rare diseases, the extraction results are inaccurate because the relevant knowledge may be missing in the existing dictionaries, knowledge bases and rule bases. On the other hand, the expert manpower cost required to maintain these resources is extremely high and time-consuming. Experts need to devote a lot of energy to sort out, update and improve the knowledge therein, which not only increases the research cost, but also limits the improvement of processing efficiency.

[0005] In recent years, advanced information technologies and concepts such as large language models and human-machine collaborative strategies have developed rapidly. Large language models have powerful natural language understanding and generation capabilities. By training on massive text data, they can learn the statistical laws and patterns of language, and demonstrate amazing text semantic understanding and generation capabilities. Human-machine collaborative strategies fully combine the expertise of human experts and the efficient processing capabilities of machines. Exploring the application of these new technologies and concepts to the extraction of semantic structural units has important scientific significance and research value in promoting the development of medical text processing technology. Summary of the invention

[0006] To this end, an embodiment of the present invention provides a method and system for intelligently annotating semantic structural units in medical texts, which is used to solve the problems in the prior art that semantic structural unit identification methods require the maintenance of huge resources, have high labor costs, and have poor adaptability and flexibility to specific corpora.

[0007] In order to solve the above problems, an embodiment of the present invention provides an intelligent annotation method for semantic structure units in medical texts, the method comprising: S1: obtaining a pre-trained large language model, collecting medical text data containing semantic structure unit annotation information, processing the medical text data into a format suitable for the large language model, using the processed data to fine-tune the pre-trained large language model, and constructing a semantic structure unit automatic recognition model based on the large language model; S2: Integrate the semantic structure unit automatic recognition model with the text annotation tool to obtain a human-computer interaction annotation tool, automatically annotate the medical text data through the human-computer interaction annotation tool, display the automatically annotated semantic structure unit information on the web page, and organize experts to manually proofread the annotation results based on the human-computer interaction strategy; S3: Collect manually proofread annotated data. Whenever the newly collected annotated data reaches a preset quantity threshold, or the performance of the model on a specific evaluation indicator meets the preset conditions, the semantic structure unit automatic recognition model is retrained using the newly collected data to generate a new semantic structure unit automatic recognition model, and execute S2.

[0008] Preferably, the pre-trained large language model is ChatGLM-6B, and an efficient parameter fine-tuning method based on P-Tuning v2 is used to perform instruction fine-tuning on the pre-trained large language model.

[0009] Preferably, the text annotation tool is a BRAT universal text annotator, and customization of the BRAT universal text annotator includes configuring the semantic structure unit automatic recognition model in a specified format, defining entities, attributes and relationships of the semantic structure unit automatic recognition model, and combining an intelligent annotation interface to support automatic annotation, web page display and interactive proofreading of semantic structure units.

[0010] Preferably, the preset quantity threshold is n extraction instances of semantic structure units, and the specific evaluation index includes that the accuracy of the model on the validation set decreases by more than 5% or the recall rate is less than 80%.

[0011] Preferably, when processing the medical text data into a format suitable for a large language model, the medical text and its associated semantic structure unit annotation data are organized into a command data form of "Instructions: You are a medical information extraction expert. Please extract the semantic structure units contained in the input medical text and output them in JSON format; input: [medical text content], output: [corresponding semantic structure unit JSON format content]".

[0012] The embodiment of the present invention further provides a system for intelligently annotating semantic structural units in medical texts, which is used to implement the intelligently annotating method for semantic structural units in medical texts described above, and specifically includes: A model building module is used to obtain a pre-trained large language model, collect medical text data containing semantic structure unit annotation information, process the medical text data into a format suitable for the large language model, use the processed data to fine-tune the instructions of the pre-trained large language model, and build a semantic structure unit automatic recognition model based on the large language model; An annotation and proofreading module, which is used to integrate the semantic structure unit automatic recognition model with the text annotation tool to obtain a human-computer interaction annotation tool, automatically annotate the medical text data through the human-computer interaction annotation tool, display the automatically annotated semantic structure unit information on the web page, and organize experts to manually proofread the annotation results based on the human-computer interaction strategy; The model iteration module is used to collect manually proofread annotated data. Whenever the newly collected annotated data reaches a preset quantity threshold, or the performance of the model on a specific evaluation indicator meets the preset conditions, the newly collected data is used to retrain the semantic structure unit automatic recognition model, generate a new semantic structure unit automatic recognition model, and execute the annotation and proofreading module.

[0013] An embodiment of the present invention also provides an electronic device, which includes a processor, a memory and a bus system, wherein the processor and the memory are connected via the bus system, the memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the above-mentioned intelligent annotation method for semantic structure units in medical texts.

[0014] An embodiment of the present invention further provides a computer storage medium storing a computer software product, wherein the computer software product includes a number of instructions for enabling a computer device to execute the above-mentioned method for intelligently marking semantic structure units in medical texts.

[0015] It can be seen from the above technical solutions that the present invention has the following beneficial effects: (1) The present invention uses the powerful natural language understanding and generation capabilities of a large language model to identify and extract semantic structural units in medical texts, which greatly reduces maintenance costs compared to methods based on dictionaries, knowledge bases, and rule bases.

[0016] (2) The self-learning and evolution mechanism constructed by the present invention can drive the continuous iteration of the semantic structure unit recognition model. This feature, combined with the data-driven advantage, further reduces the model maintenance cost, and as the labeled data continues to increase, the model performance continues to improve.

[0017] (3) Make full use of the human-computer collaborative strategy and effectively solve the hallucination problem that may occur when the large model algorithm recognizes semantic structural units through visual interaction of experts. This feature works synergistically with the data-driven and self-evolutionary features to improve the performance of the automatic recognition model of semantic structural units based on large language models while continuously reducing the workload of manual interactive proofreading, significantly improving the annotation efficiency of semantic structural units in medical texts. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the implementation cases of the present invention or the technical solutions in the prior art, the following is a brief description of the drawings required for use in the embodiments. By referring to the drawings, the features and advantages of the present invention will be more clearly understood. The drawings are schematic and should not be understood as limiting the present invention in any way. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 A flowchart of a method for intelligently marking semantic structural units in a medical text provided in an embodiment; Figure 2 A schematic diagram of constructing an automatic recognition model of semantic structure units based on a large language model in an embodiment; Figure 3 A schematic diagram of human-machine collaborative labeling in an embodiment; Figure 4 A schematic diagram of a model iteration based on a self-learning mechanism in an embodiment; Figure 5 A schematic diagram of fine-tuning a large language model to obtain an automatic recognition model of semantic structure units based on the large language model in an embodiment; Figure 6 A schematic diagram of automatically annotating semantic structural units in medical texts in an embodiment; Figure 7 A block diagram of an intelligent annotation system for semantic structural units in medical texts provided in an embodiment. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1

[0020] In order to solve the problems in the prior art of semantic structure unit recognition methods, such as the need to maintain huge resources, high labor costs, poor adaptability and flexibility to specific corpora. Figure 1 As shown, an embodiment of the present invention proposes an intelligent annotation method for semantic structure units in medical texts, the method comprising: S1: Obtain a pre-trained large language model, collect medical text data containing semantic structure unit annotation information, process the medical text data into a format suitable for the large language model, use the processed data to fine-tune the pre-trained large language model, and build a semantic structure unit automatic recognition model based on the large language model; S2: Integrate the semantic structure unit automatic recognition model with the text annotation tool to obtain a human-computer interaction annotation tool, automatically annotate medical text data through the human-computer interaction annotation tool, display the automatically annotated semantic structure unit information on the web page, and organize experts to manually proofread the annotation results based on the human-computer interaction strategy; S3: Collect manually proofread annotated data. Whenever the newly collected annotated data reaches a preset quantity threshold, or the performance of the model on a specific evaluation indicator meets the preset conditions, the newly collected data is used to retrain the semantic structure unit automatic recognition model, generate a new semantic structure unit automatic recognition model, and execute S2.

[0021] It can be seen from the above technical solution that the present invention proposes an intelligent annotation method for semantic structural units in medical texts. First, a pre-trained large language model is obtained, the medical text annotation data is sorted and the model is fine-tuned to construct an automatic recognition model, which avoids the maintenance of a huge dictionary, knowledge base and rule base, and reduces labor costs. Then, the automatic recognition model is integrated with the text annotation tool to realize automatic annotation, web page display annotation information, and manual proofreading by experts. The inaccuracy problem that may occur in large models in specific tasks is solved through human-computer interaction strategies. Finally, when the newly collected annotation data reaches the preset quantity threshold or the model meets the preset conditions on specific evaluation indicators, the new data is used to retrain the model and generate a new recognition model. This process continuously improves the model performance, enhances the adaptability and flexibility to specific corpus, and also reduces the workload of subsequent manual proofreading, improving annotation efficiency.

[0022] In step S1, a pre-trained large language model is obtained, medical text data containing semantic structure unit annotation information is collected, the medical text data is processed into a format suitable for the large language model, the pre-trained large language model is fine-tuned using the processed data, and an automatic recognition model of semantic structure units based on the large language model is constructed.

[0023] Specifically, the Large Language Model (LLM) is based on deep learning technology. With a large number of parameters and complex network structures, it learns language statistical laws and patterns under the training of massive text data and has powerful natural language understanding and generation capabilities. However, because it is based on general corpus training, it has problems such as inaccuracy, repetition, and hallucination when processing specific tasks.

[0024] The present invention uses ChatGLM-6B as the pre-trained large language model. ChatGLM-6B was developed by Tsinghua University, based on the General Language Model (GLM) architecture, and has 6.2 billion parameters. It has been trained in both Chinese and English with about 1T identifiers, and combined with supervised fine-tuning, feedback self-help, human feedback reinforcement learning and other technologies, it performs well in Chinese question-answering and dialogue.

[0025] In order to make ChatGLM-6B suitable for the task of extracting semantic structural units in medical texts, the present invention collects about 8,000 medical text fragments containing semantic structural unit annotation information, organizes them into a format that is understandable to ChatGLM-6B, and uses an efficient parameter fine-tuning method based on P-Tuning v2 to adjust the neural network architecture parameters of ChatGLM-6B, thereby obtaining a semantic structural unit automatic recognition model based on a large language model, such as Figure 2 shown.

[0026] In step S2, the semantic structure unit automatic recognition model is integrated with the text annotation tool to obtain a human-computer interaction annotation tool, and the medical text data is automatically annotated by the human-computer interaction annotation tool. The automatically annotated semantic structure unit information is displayed on the web page, and experts are organized to manually proofread the annotation results based on the human-computer interaction strategy.

[0027] Specifically, Figure 3 As shown, the present invention combines the semantic structure unit automatic recognition model obtained above with the BRAT universal text annotator to obtain a human-computer interactive annotation tool, which has the ability to automatically extract and display semantic structure units on a web page. The BRAT universal text annotator is a widely used universal text annotator with the characteristics of flexible configuration and simple operation.

[0028] In order to adapt it to the intelligent annotation task of semantic structure units, the following in-depth customization was carried out: First, the semantic structure unit model was configured according to the format specified by the BRAT general text annotator, and the entities, attributes and relationships of the model were clarified, so that the BRAT general text annotator could support the annotation of the semantic structure unit model; second, the intelligent annotation interface provided by the BRAT general text annotator was used to combine it with the semantic structure unit intelligent recognition model to realize the intelligent annotation function of semantic structure units. After this transformation, the BRAT general text annotator has the ability to automatically annotate, display on web pages, and interactively proofread the semantic structure units in medical texts, and finally obtain an intelligent human-computer interactive annotation tool suitable for semantic structure units.

[0029] In step S3, the manually proofread annotated data is collected. Whenever the newly collected annotated data reaches a preset quantity threshold, or the performance of the model on a specific evaluation indicator meets the preset conditions, the semantic structure unit automatic recognition model is retrained using the newly collected data to generate a new semantic structure unit automatic recognition model, and step S2 is executed.

[0030] Specifically, as experts use the above method to carry out labeling work, the labeling system background will continue to accumulate new labeling data. The present invention has designed a self-learning mechanism that can automatically collect and organize the labeling data proofread by experts. When the labeling data accumulates more than 10,000 extraction instances of semantic structure units, or the performance of the model on specific evaluation indicators meets the preset conditions, where the specific evaluation indicators include the accuracy of the model on the validation set dropping by more than 5% or the recall rate being lower than 80%, the automatic learning mechanism is automatically triggered, and these new data are used to conduct a new round of training on the constructed semantic structure unit automatic recognition model, generate a new semantic structure unit automatic recognition model, and execute step S2 to obtain a semantic structure unit automatic recognition model with better performance, thereby reducing the workload of expert labeling, such as Figure 4 shown.

[0031] Case: Based on the large amount of medical text data containing semantic structural unit annotation information accumulated in the early stage, it is necessary to convert it into a form suitable for large language model instruction fine-tuning before it is used to fine-tune the large language model. Taking "the patient suddenly developed acupuncture-like abdominal pain, especially in the right lower abdomen" as an example, the corresponding semantic structural unit is "{core word: abdominal pain, existence: existence, pain nature: acupuncture-like, onset site: right lower abdomen}", and the instruction data format is: "Instruction: You are a medical information extraction expert. Please extract the semantic structural units contained in the input medical text according to the input medical text and output it in JSON format. Input: The patient suddenly developed acupuncture-like abdominal pain, especially in the right lower abdomen; Output: {core word: abdominal pain, existence: existence, pain nature: acupuncture-like, onset site: right lower abdomen}".

[0032] After all medical texts and their associated semantic structure unit annotation data are processed in this way, a training set for fine-tuning the large language model instructions is formed. The instruction library and the ChatGLM-6B base model are loaded using Python's transformers library, fine-tuned and saved. The fine-tuned large model can be used for automatic recognition of semantic structure units. For example, if the input is "the patient's headache lasted for 3 days and was relieved after rest", the fine-tuned large language model will output the semantic structure unit of "{core word: headache, existence: existence, duration: 3 days}", such as Figure 5 shown.

[0033] The fine-tuned large model is embedded in the BRAT universal text annotator, so that the annotator supports automatic annotation, visual display, human-computer interactive proofreading, and saving of semantic structure units. For example, in the actual annotation process, the semantic structure unit automatic recognition model based on the large model is combined with the BRAT universal text annotator to obtain a human-computer interactive annotation tool, which can automatically annotate the semantic structure units in medical texts and visualize the identified semantic structure units on the web page, including core words and related attribute words. Experts can interactively proofread the machine annotations on this basis to complete the semantic structure unit annotation of human-computer collaboration, such as Figure 6 shown.

[0034] In the process of continuous annotation of medical texts, new annotation instances will be continuously generated. For example, a certain annotation operation can generate 8 new semantic structure unit annotation examples, which will be recorded in the background database. When a certain number (such as more than 10,000 semantic structure unit extraction instances) is accumulated, the re-instruction fine-tuning mechanism of the ChatGLM-6B large model is automatically triggered. In theory, the more training data the ChatGLM-6B large model receives, the better the recognition performance of the semantic structure unit, thereby realizing the continuous optimization of the large language model by manually annotated corpus, making the annotation of semantic structure units more accurate and efficient. Embodiment 2

[0035] like Figure 7 As shown, the present invention provides an intelligent annotation system for semantic structure units in medical texts, which is used to implement the intelligent annotation method for semantic structure units in medical texts of the first embodiment, and specifically includes: A model building module is used to obtain a pre-trained large language model, collect medical text data containing semantic structure unit annotation information, process the medical text data into a format suitable for the large language model, use the processed data to fine-tune the instructions of the pre-trained large language model, and build a semantic structure unit automatic recognition model based on the large language model; The annotation and proofreading module is used to integrate the semantic structure unit automatic recognition model with the text annotation tool to obtain a human-computer interactive annotation tool, automatically annotate the medical text data through the human-computer interactive annotation tool, display the automatically annotated semantic structure unit information on the web page, and organize experts to manually proofread the annotation results based on the human-computer interactive strategy; The model iteration module is used to collect manually proofread annotated data. Whenever the newly collected annotated data reaches a preset threshold, or the performance of the model on a specific evaluation indicator meets the preset conditions, the newly collected data is used to retrain the semantic structure unit automatic recognition model, generate a new semantic structure unit automatic recognition model, and execute the annotation and proofreading modules.

[0036] The intelligent annotation system for semantic structure units in medical texts of the present embodiment is used to implement the aforementioned intelligent annotation method for semantic structure units in medical texts. Therefore, the specific implementation methods of the intelligent annotation system for semantic structure units in medical texts can be seen in the embodiment part of the intelligent annotation method for semantic structure units in medical texts in the previous text. For example, the model building module, the annotation and proofreading module, and the model iteration module are respectively used to implement steps S1, S2, and S3 in the aforementioned intelligent annotation method for semantic structure units in medical texts. Therefore, its specific implementation methods can refer to the descriptions of the corresponding embodiments of each part. In order to avoid redundancy, they will not be repeated here. Embodiment 3

[0037] An embodiment of the present invention provides an electronic device, which includes a processor, a memory and a bus system. The processor and the memory are connected through the bus system. The memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the above-mentioned intelligent annotation method of semantic structure units in medical texts. Embodiment 4

[0038] An embodiment of the present invention provides a computer storage medium storing a computer software product. The computer software product includes a number of instructions for enabling a computer device to execute the above-mentioned intelligent annotation method for semantic structure units in medical texts.

[0039] 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.

[0040] 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.

[0041] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0042] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.

Claims

1. An intelligent annotation method for semantic structural units in medical texts, characterized in that: include: S1: obtaining a pre-trained large language model, collecting medical text data containing semantic structure unit annotation information, processing the medical text data into a format suitable for the large language model, using the processed data to fine-tune the pre-trained large language model, and constructing a semantic structure unit automatic recognition model based on the large language model; S2: Integrate the semantic structure unit automatic recognition model with the text annotation tool to obtain a human-computer interaction annotation tool, automatically annotate the medical text data through the human-computer interaction annotation tool, display the automatically annotated semantic structure unit information on the web page, and organize experts to manually proofread the annotation results based on the human-computer interaction strategy; S3: Collect manually proofread annotated data. Whenever the newly collected annotated data reaches a preset quantity threshold, or the performance of the model on a specific evaluation indicator meets the preset conditions, the semantic structure unit automatic recognition model is retrained using the newly collected data to generate a new semantic structure unit automatic recognition model, and execute S2.

2. The intelligent annotation method for semantic structural units in medical text according to claim 1, characterized in that: The pre-trained large language model is ChatGLM-6B, and the pre-trained large language model is fine-tuned by using an efficient parameter fine-tuning method based on P-Tuning v2.

3. The intelligent annotation method for semantic structural units in medical text according to claim 1, characterized in that: The text annotation tool is a BRAT universal text annotator. Customization of the BRAT universal text annotator includes configuring the semantic structure unit automatic recognition model in a specified format, defining the entities, attributes and relationships of the semantic structure unit automatic recognition model, and combining an intelligent annotation interface to support automatic annotation, web page display and interactive proofreading of semantic structure units.

4. The intelligent annotation method for semantic structural units in medical text according to claim 1, characterized in that: The preset quantity threshold is n extraction instances of semantic structure units, and the specific evaluation indicators include that the accuracy of the model on the validation set drops by more than 5% or the recall rate is less than 80%.

5. The intelligent annotation method for semantic structural units in medical text according to claim 1, characterized in that: When the medical text data is processed into a format suitable for a large language model, the medical text and its associated semantic structure unit annotation data are organized into "Instructions: You are a medical information extraction expert. Please extract the semantic structure units contained in the input medical text and output them in JSON format; Input: [medical text content], output: [corresponding semantic structure unit JSON format content]” in the form of imperative data.

6. An intelligent annotation system for semantic structural units in medical texts, characterized in that: The system is used to implement the intelligent annotation method of semantic structure units in medical texts according to any one of claims 1 to 5, specifically comprising: A model building module is used to obtain a pre-trained large language model, collect medical text data containing semantic structure unit annotation information, process the medical text data into a format suitable for the large language model, use the processed data to fine-tune the instructions of the pre-trained large language model, and build a semantic structure unit automatic recognition model based on the large language model; An annotation and proofreading module, which is used to integrate the semantic structure unit automatic recognition model with the text annotation tool to obtain a human-computer interaction annotation tool, automatically annotate the medical text data through the human-computer interaction annotation tool, display the automatically annotated semantic structure unit information on the web page, and organize experts to manually proofread the annotation results based on the human-computer interaction strategy; The model iteration module is used to collect manually proofread annotated data. Whenever the newly collected annotated data reaches a preset quantity threshold, or the performance of the model on a specific evaluation indicator meets the preset conditions, the newly collected data is used to retrain the semantic structure unit automatic recognition model, generate a new semantic structure unit automatic recognition model, and execute the annotation and proofreading module.

7. An electronic device, characterized in that: The electronic device includes a processor, a memory and a bus system, the processor and the memory are connected through the bus system, the memory is used to store instructions, and the processor is used to execute the instructions stored in the memory to implement the intelligent annotation method of semantic structure units in medical texts as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that The computer storage medium stores a computer software product, which includes a number of instructions for enabling a computer device to execute the intelligent annotation method for semantic structure units in medical texts as described in any one of claims 1 to 5.

Images