Orthopedic nursing intelligent data processing system based on cloud platform

Through an intelligent data processing system based on the cloud platform, the NLP model is used to clean, segment and structure the orthopedic nursing documents, solving the problem of unstructured text analysis, and achieving efficient, accurate processing and standardized application of orthopedic nursing data.

CN120542423AInactive Publication Date: 2025-08-26SHANGHAI PUDONG HOSPITAL
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202510624785.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-15
Publication Date
2025-08-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing orthopedic nursing data processing system is difficult to effectively analyze unstructured text, which makes it difficult to capture and standardize key clinical information in a timely manner, affecting the efficient use of nursing documents.

Method used

An intelligent data processing system based on cloud platform is adopted, including nursing document storage, reading, sentence classification, structured processing and standardized modules, and a NLP model is used to identify named entities, extract relationships and detect negative words, and structured information is generated and stored in the database.

Benefits of technology

It improves the efficiency and accuracy of nursing data processing, realizes real-time capture and standardized application of key information, and supports data integration and clinical decision-making across departments and time periods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120542423A_ABST
    Figure CN120542423A_ABST
Patent Text Reader

Abstract

The invention provides an orthopedic nursing intelligent data processing system based on a cloud platform, and relates to the field of intelligent data processing.The orthopedic nursing intelligent data processing system comprises the steps that firstly, a newly-added nursing document is transmitted to a temporary storage area of a cloud end and read, and then format cleaning and text segmentation are conducted on the nursing document to generate sequence distribution described by sentence granularity; then, the sequence is distributed and input into an NLP model deployed at a cloud end to complete named entity recognition, relation extraction, negative word detection and attribute extraction to obtain structured information of the nursing document, and then the structured information is subjected to standardization and duplicate removal processing to generate standard structured information of the nursing document; and the information is stored in a structured database of the cloud. In this way, the efficiency and accuracy of nursing data processing can be improved, and the deep value of data can be released.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of intelligent data processing, and more specifically, to an orthopedic care intelligent data processing system based on a cloud platform. Background Art

[0002] With the rapid development of orthopedic diagnosis and treatment technologies, the field of orthopedic nursing is facing the dual challenges of mining data value and improving clinical efficiency. Orthopedic nursing documents serve as a continuous record of changes in patients' vital signs, nursing procedures, and disease progression. The dynamic information they contain, such as wound healing progress, pain assessment, and limb mobility, is of great reference value for clinical decision-making.

[0003] However, existing orthopedic nursing data processing systems have long been constrained by the difficulty of parsing unstructured text. Traditional manual processing models require medical staff to repeatedly review free-text information scattered throughout medical records and handover reports, integrating key clinical indicators through manual comparison and memory recall. This model not only makes it difficult to capture time-sensitive information in nursing documents (such as the temporal changes in inflammatory signs), but also hinders the automated extraction and standardized application of key clinical information due to the prevalence of negative expressions (such as "no exudate"), ambiguous descriptions (such as "no significant improvement in activity limitation"), and terminology differences caused by the personalized recording habits of medical staff (such as the confusion between "redness and swelling" and "localized congestion").

[0004] Therefore, an optimized intelligent data processing solution for orthopedic care is desired. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an orthopedic care intelligent data processing system based on a cloud platform.

[0006] According to one aspect of the present application, a cloud-based orthopedic care intelligent data processing system is provided, which includes:

[0007] Nursing document storage module, used to transfer newly added nursing documents to the temporary storage area in the cloud;

[0008] A nursing document reading module, configured to read the newly added nursing document from the temporary storage area;

[0009] A nursing document sentence segmentation module is used to perform format cleaning and text segmentation on the newly added nursing documents to obtain a sequence distribution of granular descriptions of nursing document sentences;

[0010] A structured processing module, configured to input the sequence distribution of the granular description of the nursing document into the NLP model deployed on the cloud to perform named entity recognition, relationship extraction, negation word detection, and attribute extraction to obtain structured information of the nursing document;

[0011] A standardization processing module, configured to perform standardization and deduplication processing on the nursing document structured information to obtain standard nursing document structured information;

[0012] The data storage module is used to store the standard structured information of the nursing document in the structured database on the cloud.

[0013] Compared with the existing technology, the cloud-based orthopedic nursing intelligent data processing system provided by this application first transfers the newly added nursing documents to the temporary storage area in the cloud and reads them, then performs format cleaning and text segmentation on the nursing documents to generate a sequence distribution of sentence-granular descriptions, and then inputs the sequence distribution into the NLP model deployed in the cloud to complete named entity recognition, relationship extraction, negation word detection and attribute extraction to obtain nursing document structured information, and then standardizes and deduplicates the structured information to generate standard structured information of nursing documents, and stores it in a structured database in the cloud. In this way, the efficiency and accuracy of nursing data processing can be improved, and it is conducive to releasing the deep value of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0015] Figure 1 This is a system block diagram of an orthopedic care intelligent data processing system based on a cloud platform according to an embodiment of the present application.

[0016] Figure 2 This is a block diagram of a structured processing module in a cloud platform-based orthopedic care intelligent data processing system according to an embodiment of the present application.

[0017] Figure 3 This is a block diagram of a named entity relationship extraction unit in a cloud platform-based orthopedic care intelligent data processing system according to an embodiment of the present application.

[0018] Figure 4 This is a block diagram of the first-second named entity fine-grained modeling subunit in the cloud platform-based orthopedic care intelligent data processing system according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0020] With the rapid development of orthopedic diagnosis and treatment technologies, orthopedic nursing faces the dual challenges of mining data value and improving clinical efficiency. Nursing documents record dynamic information such as patients' wound healing, pain assessment, and limb mobility, and are of great reference value for clinical decision-making. However, existing data processing methods are limited by the difficulty of parsing unstructured text and rely on manual review and comparison, making it difficult to capture time-sensitive information such as inflammatory signs in a timely manner. At the same time, problems such as negative expressions, ambiguous descriptions, and terminology differences further increase the difficulty of information extraction and standardized application, restricting the efficient use of nursing documents.

[0021] Based on this, this application proposes an orthopedic care intelligent data processing system based on a cloud platform. Figure 1 FIG is a system block diagram of an orthopedic care intelligent data processing system based on a cloud platform according to an embodiment of the present application. Figure 1 As shown, in the cloud platform-based orthopedic nursing intelligent data processing system 100, it includes: a nursing document storage module 110, which is used to transmit the newly added nursing documents to a temporary storage area in the cloud; a nursing document reading module 120, which is used to read the newly added nursing documents from the temporary storage area; a nursing document sentence segmentation module 130, which is used to perform format cleaning and text segmentation on the newly added nursing documents to obtain a sequence distribution of sentence granularity descriptions of the nursing documents; a structured processing module 140, which is used to input the sequence distribution of sentence granularity descriptions of the nursing documents into the NLP model deployed on the cloud to perform named entity recognition, relationship extraction, negation word detection and attribute extraction to obtain nursing document structured information; a standardization processing module 150, which is used to standardize and deduplicate the nursing document structured information to obtain standard structured information of the nursing documents; and a data storage module 160, which is used to store the standard structured information of the nursing documents in a structured database on the cloud.

[0022] Specifically, to address the semantic parsing barriers caused by the unstructured nature of nursing records, the system uses text segmentation technology to convert free text into sentence-sized sequences, building a computable semantic unit foundation for subsequent NLP models. The multi-task NLP model deployed in the cloud can accurately identify negative sign descriptions such as "no exudate" by simultaneously performing negation word detection and attribute extraction. It also uses relationship extraction technology to dynamically associate information containing time dimensions with specific clinical indicators, addressing the flaw of traditional manual processing where temporal information is easily missed. To address terminology differences and diverse recording styles, the system uses a standardization module to perform term mapping on extracted synonymous expressions such as "redness and swelling" and "local congestion." Combined with a deduplication algorithm, it eliminates duplicate records across shifts, allowing key indicators such as incision healing progress and limb mobility, scattered across multiple text segments, to be integrated into temporally coherent structured data. The cloud platform architecture not only supports the real-time processing of massive amounts of nursing documents, but also achieves global indexing of data across departments and time periods through a structured database, enabling researchers to quickly extract nursing parameter sets for specific stages of the disease process, thereby breaking through the constraints of information fragmentation on clinical decision support and scientific research analysis under the traditional model.

[0023] In an embodiment of the present application, the nursing document storage module 110 is used to transfer newly added nursing documents to a temporary storage area in the cloud. It should be understood that a large number of nursing documents will be generated in orthopedic nursing, and transferring them to a temporary storage area in the cloud can achieve centralized storage of data. This makes it easier for the system to manage all newly added nursing documents in a unified manner, avoiding the data being scattered in different places, making it difficult to effectively integrate and process them. In other words, storing newly added nursing documents in a temporary storage area makes it convenient for the system to read data from this area for subsequent processing. The setting of the temporary storage area allows data to be processed in a specific location, and the system can read and process these data in sequence according to a predetermined process, thereby improving the efficiency and accuracy of data processing.

[0024] In an embodiment of the present application, the nursing document reading module 120 is used to read the newly added nursing documents from the temporary storage area. It should be understood that reading the newly added nursing documents from the temporary storage area can provide original input data for the subsequent nursing document data processing process, ensuring that the newly added unstructured nursing documents can be accurately acquired by the system and included in the processing scope. Specifically, the temporary storage area serves as a buffer area for the cloud to receive newly added nursing documents. Its role is to temporarily store the unprocessed original data to avoid data transmission interruption or processing conflict problems that may be caused by real-time processing. By reading data from the temporary storage area, the system can obtain the newly added documents to be processed in a reliable manner, forming a complete data flow chain from data reception, temporary storage to processing, providing the prerequisite for subsequent operations such as clear formatting, text segmentation, natural language processing, etc., ensuring the continuity and stability of the entire data processing process, and ensuring that all newly added nursing document data can be captured in a timely manner and enter the standardized processing process to avoid data omissions or processing delays.

[0025] In an embodiment of the present application, the nursing document sentence segmentation module 130 is used to perform format cleaning and text segmentation on the newly added nursing documents to obtain a sequence distribution of sentence-granular descriptions of the nursing documents. Accordingly, since the original nursing documents may have various format inconsistencies, such as different recording methods and typesetting habits of different medical staff, they may contain non-standard format elements such as special characters, spaces, and line breaks. These inconsistencies will affect the accuracy and efficiency of subsequent data processing. Through format cleaning, the data can be unified into a standard format, which is convenient for computers to process and analyze. Natural language processing (NLP) models usually need to process text data in a specific format and granularity. Nursing documents are complex natural language texts. Directly using them as input to the NLP model makes it difficult for the model to understand and process the semantic information therein. Segmenting the text into a sequence distribution of sentence-granular descriptions can provide the NLP model with clearer and easier-to-process input units, which meets the NLP model's processing requirements for text data. In this way, each sentence can be regarded as an independent semantic unit, and the system can perform operations such as named entity recognition, relationship extraction, negation detection, and attribute extraction on these units, thereby more effectively extracting key information from the nursing documents.

[0026] In the embodiment of the present application, the structured processing module 140 is used to input the sequence distribution of the granular description of the nursing document into the NLP model deployed on the cloud to perform named entity recognition, relationship extraction, negation word detection and attribute extraction to obtain the structured information of the nursing document. Specifically, Figure 2 FIG is a block diagram of a structured processing module in an orthopedic care intelligent data processing system based on a cloud platform according to an embodiment of the present application. Figure 2As shown, the structured processing module 140 includes: a named entity recognition unit 141, which is used to perform named entity recognition on the sequence distribution of the sentence granularity description of the nursing document based on the NLP model to obtain a set of named entities; a named entity relationship extraction unit 142, which is used to perform relationship extraction on the set of named entities to obtain a set of named entity relationship types; a negation detection unit 143, which is used to perform negation detection on the set of named entities to obtain a set of named entity negation detection results; an attribute extraction unit 144, which is used to perform attribute extraction on the set of named entities to obtain a set of named entity attribute detection results; an information integration unit 145, which is used to integrate the set of named entities, the set of named entity relationship types, the set of named entity negation detection results and the set of named entity attribute detection results to obtain the nursing document structured information.

[0027] In an embodiment of the present application, the named entity recognition unit 141 is used to perform named entity recognition on the sequence distribution of the granular description of the nursing document sentences based on the NLP model to obtain a set of named entities. It should be understood that nursing documents contain a large amount of information about the patient's condition, nursing operations, medical equipment, etc., which exists in the form of natural language and is expressed in a flexible way. Named entities are words or phrases with specific meanings in the text, such as patient names, disease names, body parts, nursing measures, etc. Through named entity recognition, these complex text contents can be preliminarily classified and extracted, and the key entity information can be separated out to further analyze the relationship between them, determine whether there is a negation, and extract related attributes.

[0028] In particular, in a specific embodiment of the present application, the implementation process of performing named entity recognition on the sequence distribution of the granular description of the nursing document sentence based on the NLP model to obtain a set of named entities includes: first, data annotation and model training need to be carried out, which is the basic link to ensure recognition accuracy. After a large amount of orthopedic nursing document texts are collected, detailed manual annotation is performed according to pre-defined entity categories. These categories cover symptoms / signs (such as pain, redness, swelling, fever, etc.), anatomical parts (such as knee joints, hips, incisions, etc.), medical procedures (such as dressing changes, physical therapy, etc.), vital signs / measurements (such as body temperature 38.5℃, pain score VAS7, etc.), drugs (such as morphine, cefazolin, etc.), and time expressions (such as this afternoon, the third day after surgery, etc.). For example, in the text "The patient developed redness and swelling of the knee joint on the third day after surgery, and the dressing was changed, with a pain score of VAS7", "the third day" needs to be marked as a time expression, "knee joint" as an anatomical part, "redness and swelling" as a symptom / sign, "dressing change" as a medical procedure, and "pain score VAS7" as a vital sign / measurement. After completing a large number of such annotations, the data is used to train an NLP model (such as a BERT-based pre-trained model for the medical field) so that the model can fully learn the linguistic features, expressions, and contextual associations of different entity types in orthopedic nursing texts, laying the foundation for subsequent recognition. Next, the key step of inputting the model for recognition is entered. Each nursing document sentence granularity description in the sequence distribution of nursing document sentence granularity descriptions is input into the trained NLP model in sequence. The model performs deep feature extraction on each word or phrase in the sentence, and uses a multi-layer neural network to calculate and determine whether it belongs to a certain type of named entity. Taking the sentence "On the third day after surgery, cefazolin was given for anti-inflammatory treatment" as an example, the model will recognize "the third day" as a time expression and "cefazolin" as a drug. After the model performs this operation on each sentence, it outputs the entities and their categories recognized in each sentence, and then aggregates and integrates the recognition results of all sentences to form a complete set of named entities. This collection includes various entities extracted from the granular text of nursing documents: symptom / sign entities reflect the patient's physical condition, anatomical site entities identify the location of the problem, medical procedure entities record the nursing actions performed, vital sign / measurement entities provide quantitative health indicators, medication entities indicate medication usage, and time expression entities identify event time nodes. This entity data provides a solid foundation for subsequent higher-level processing such as relationship extraction and semantic analysis, gradually transforming the unstructured information in nursing documents into structured, valuable clinical data, providing key support for data processing and clinical decision support in the field of orthopedic nursing.

[0029] In an embodiment of the present application, the named entity relationship extraction unit 142 is used to perform relationship extraction on the set of named entities to obtain a set of named entity relationship types. Accordingly, nursing documents not only contain various named entities, but also have rich semantic relationships between entities. For example, there is a "manifested as" relationship between diseases and symptoms, and a "targeted" relationship between nursing measures and diseases. Only by understanding these relationships can we truly understand the complete semantics expressed by nursing documents and provide more in-depth information for subsequent analysis and application. By extracting the relationships between named entities, we can deeply explore the semantic information in nursing documents and reveal the intrinsic connections between entities. This helps medical staff to have a more comprehensive and in-depth understanding of the patient's disease progression, treatment process, and the effectiveness of nursing measures, etc., and provide stronger support for clinical decision-making. However, when processing orthopedic nursing documents, traditional relationship extraction technology is often subject to the implicit expression of entity relationships in free text and complex semantic coupling. For example, when two independent descriptions appear in the nursing record: "Incision exudate decreased on the third day after surgery" and "The patient complained of worsening pain at night," manual review can infer the potential correlation between worsening pain and incision condition. However, the existing algorithm lacks the ability to jointly parse the time adverbial "third day after surgery" and the symptom verb "decreased," resulting in the inability to automatically establish a three-dimensional relationship network between "incision exudate-time decay" and "pain level-diurnal fluctuation."

[0030] Based on this, the technical concept of this application is to construct multi-dimensional semantic associations to break through the semantic barriers of orthopedic nursing entity relationship analysis. After obtaining the set of named entities, the system first performs context-aware semantic embedding encoding on the named entities in the nursing documents to capture the semantic features of the entities. In view of the time sensitivity unique to orthopedic nursing (such as "the third day after surgery") and the dynamic correlation characteristics of symptoms (such as "reduced exudate" and "increased pain"), the system adopts a fine-grained interactive method to project the semantic vectors of the two entities into the orthopedic feature space for principal component correlation analysis, and intelligently obtains the relationship between the named entities by decoupling the explicit / implicit semantic association patterns between entities. This technical solution effectively overcomes the defects of traditional methods in insufficient parsing of complex semantic coupling in free text, and can identify dynamic associations such as "incision exudate-time decay" and potential patterns of "pain level-diurnal fluctuations", providing technical support for the establishment of a three-dimensional relationship network across nursing records, and significantly improving the accuracy of causal inference of clinical events and evaluation of the effectiveness of nursing measures.

[0031] Specifically, Figure 3 FIG is a block diagram of a named entity relationship extraction unit in an orthopedic care intelligent data processing system based on a cloud platform according to an embodiment of the present application. Figure 3As shown, the named entity relationship extraction unit 142 includes: a first and a second named entity extraction subunit 1421, which is used to extract the first named entity and the second named entity from the set of named entities; a named entity semantic embedding subunit 1422, which is used to perform semantic embedding encoding on the first named entity and the second named entity to obtain a first named entity semantic embedding encoding vector and a second named entity semantic embedding encoding vector; a first-second named entity fine-grained modeling subunit 1423, which is used to perform performance-oriented named entity semantic principal component association analysis on the first named entity semantic embedding encoding vector and the second named entity semantic embedding encoding vector to obtain a semantic fine-grained relationship modeling matrix between the first and second named entities; a relationship type label determination subunit 1424, which is used to determine the relationship type label between the first named entity and the second named entity based on the semantic fine-grained relationship modeling matrix between the first and second named entities.

[0032] In an embodiment of the present application, the first and second named entity extraction subunit 1421 is used to extract the first named entity and the second named entity from the set of named entities. It should be understood that in natural language processing, semantic relationships are usually established between two or more entities. For the entity relationship analysis in nursing documents, starting from two named entities is more in line with the logic of semantic association processing. For example, when judging the relationship between "disease" and "symptoms", these two entities are analyzed as the first and second named entities respectively, which makes it easier to find the semantic relationship such as "manifested as" between them. After performing the relationship analysis on each pair of first and second named entities, the obtained relationship information can be integrated to gradually construct a relationship network for the entire named entity set. This relationship network can comprehensively display the complex relationships between various entities in nursing documents, and provide richer and more accurate data support for clinical decision-making and scientific research analysis.

[0033] In an embodiment of the present application, the named entity semantic embedding subunit 1422 is used to perform semantic embedding encoding on the first named entity and the second named entity to obtain a first named entity semantic embedding encoding vector and a second named entity semantic embedding encoding vector. Specifically, in an embodiment of the present application, the named entity semantic embedding subunit is used to: perform BioBERT-based semantic embedding encoding on the first named entity and the second named entity to obtain the first named entity semantic embedding encoding vector and the second named entity semantic embedding encoding vector.

[0034] Accordingly, considering that in the semantic parsing scenario of orthopedic nursing documents, the accurate identification of the relationship between entities is highly dependent on the understanding of the deep semantics of clinical terms. Since orthopedic nursing records are full of professional terms (such as "joint stiffness" and "callus formation") and temporal descriptions of symptoms (such as "incisional exudate on the third day after surgery"), it is difficult for traditional general semantic encoding models to effectively capture the exclusive semantic features of these medical entities. For example, "redness and swelling" and "local congestion" differ significantly in surface vocabulary, but both point to the core indications of inflammatory response; and "improvement in limited activity" and "increase in ROM" are expressed in different forms, but both reflect the recovery process of limb function. If a general text encoding method is used, the semantic similarity of such medical concepts is easily ignored, resulting in the inability to establish correct clinical logical associations during subsequent relationship extraction. Therefore, this application performs BioBERT-based semantic embedding encoding on the first named entity and the second named entity to obtain the first named entity semantic embedding encoding vector and the second named entity semantic embedding encoding vector. Notably, BioBERT's pre-training on a corpus of biomedical literature has enabled its internal parameters to acquire a deep semantic memory for specialized terms such as "wound healing" and "pain score." When processing entities with a time dimension, such as "exudate decreased on the third day after surgery," the model automatically activates semantic cognition of time decay patterns, establishing a potential association between the temporal adverbial "third day" and the symptom changes of "exudate decreased." When encoding the entity "pain worsens at night," the model not only captures "pain" as the core symptom attribute but also, through context-aware mechanisms, identifies the dynamic impact of the time modifier "nighttime" on symptom severity. In this way, BioBERT's semantic embedding encoding enables the extraction of richer and more detailed semantic features from the text of nursing documents. These features help the model better understand the underlying relationships between named entities, improving the accuracy of relationship extraction.

[0035] In this embodiment of the present application, the first-second named entity fine-grained modeling subunit 1423 is used to perform performance-oriented named entity semantic principal component association analysis on the first named entity semantic embedding encoding vector and the second named entity semantic embedding encoding vector to obtain a semantic fine-grained relationship modeling matrix between the first and second named entities. Specifically, Figure 4 FIG. 1 is a block diagram of the first-second named entity fine-grained modeling subunit in the orthopedic care intelligent data processing system based on the cloud platform according to an embodiment of the present application. Figure 4As shown, the first-second named entity fine-grained modeling subunit 1423 includes: a first-second named entity principal component association analysis secondary subunit 1423-1, which is used to perform principal component fine-grained association analysis on the first named entity semantic embedding coding vector and the second named entity semantic embedding coding vector to obtain a set of first-second named entity semantic embedding principal component fine-grained association coding vectors; a performance factor topological association matrix generation secondary subunit 1423-2, which is used to calculate the performance factor between any two first-second named entity semantic embedding principal component fine-grained association coding vectors in the set of the first-second named entity semantic embedding principal component fine-grained association coding vectors to obtain a first-second named entity semantic embedding performance factor topological association matrix; a first-second named entity semantic convolutional coding secondary subunit 1423-3, which is used to perform graph convolution coding on the set of the first-second named entity semantic embedding principal component fine-grained association coding vectors and the first-second named entity semantic embedding performance factor topological association matrix to obtain a semantic fine-grained relationship modeling matrix between the first-second named entities.

[0036] It should be understood that the relationships between named entities in nursing documents are often complex and implicit, and it is difficult to accurately grasp them only through surface text information. Traditional methods are limited by the lack of parsing capabilities of feature collinearity and nonlinear semantic associations, resulting in significant bottlenecks in the modeling of complex relationships between entities. Since the original semantic vectors of entity features in nursing records often have multicollinearity (for example, "postoperative days" and "symptom changes" are strongly coupled in the time dimension), this feature redundancy not only weakens the discriminability of relationship classification, but also obscures the potential interaction patterns between entities (such as the dynamic association between time adverbials and symptom changes). At the same time, the implicit relationships unique to orthopedic nursing (such as the indirect association between "reduced incision exudate" and "increased pain") are often manifested as nonlinear, cross-segment semantic coupling, and conventional linear modeling methods are difficult to capture such high-order interaction features, resulting in the system being unable to restore the three-dimensional relationship network formed during manual review. Based on this, the present application performs a performance-oriented named entity semantic principal component association analysis on the first named entity semantic embedding coding vector and the second named entity semantic embedding coding vector to obtain a semantic fine-grained relationship modeling matrix between the first and second named entities.

[0037] Specifically, in the embodiment of the present application, the first-second named entity principal component association analysis secondary subunit 1423-1 is used to perform feature principal component analysis on the first named entity semantic embedding coding vector and the second named entity semantic embedding coding vector to obtain a set of first named entity semantic embedding principal component coding vectors and a set of second named entity semantic embedding principal component coding vectors. This process can be expressed as follows:

[0038]

[0039] Where V1 is the first named entity semantic embedding encoding vector, PCA(·) is the feature principal component analysis, C1 is the first named entity semantic embedding sample covariance matrix calculated by V1, X is the set of the first named entity semantic embedding principal component encoding vector, x1, x2, x i and x m are the first, second, i-th and m-th first named entity semantic embedding principal component encoding vectors in the set of the first named entity semantic embedding principal component encoding vectors, Λ1 is the first named entity semantic embedding diagonal matrix, λ 11 ,λ 1m x1 and x m The corresponding eigenvalues, T is the transposition operation, V2 is the second named entity semantic embedding encoding vector, C2 is the second named entity semantic embedding sample covariance matrix calculated by V2, Y is the set of the second named entity semantic embedding principal component encoding vectors, y1, y2, y i and y m are the first, second, i-th and m-th second named entity semantic embedding principal component encoding vectors in the set of second named entity semantic embedding principal component encoding vectors, Λ2 is the second named entity semantic embedding diagonal matrix, λ 21 ,λ 2m y1 and y m The corresponding eigenvalues;

[0040] The principal component kernel fine-grained association coding is performed on each corresponding set of the first named entity semantic embedding principal component coding vector and the second named entity semantic embedding principal component coding vector in the set of the first named entity semantic embedding principal component coding vector and the set of the second named entity semantic embedding principal component coding vector to obtain a set of the first-second named entity semantic embedding principal component fine-grained association coding vector. The process can be expressed by the formula:

[0041]

[0042] Among them, ‖·‖ is the Euclidean norm of the calculated vector, α and β are x i and y i The corresponding weighting coefficient, v i It is the i-th first-second named entity semantic embedding principal component fine-grained association encoding vector in the set of first-second named entity semantic embedding principal component fine-grained association encoding vectors.

[0043] It should be understandable that due to the multicollinearity of the original named entity semantic embedding encoding vector (such as the strong coupling of "postoperative days" and "symptom changes" in the time dimension), semantic redundancy and interaction pattern ambiguity will occur. Traditional methods have difficulty in parsing nonlinear semantic coupling due to feature collinearity, while the feature principal component analysis operation can map high-dimensional features to a low-dimensional, linearly independent principal component space through orthogonal transformation, eliminating the strong coupling between redundant features. In other words, the feature principal component analysis operation decouples redundant features while reducing the dimensionality, retaining the inherent independent semantic structure of the data (such as the decoupling of time series and symptom changes), providing a low-redundancy, high-discriminative principal component feature basis for subsequent relationship modeling, and avoiding the interference of collinearity on association pattern mining.

[0044] Accordingly, considering that implicit relationships in orthopedic nursing (such as "reduction of incision exudate" and "increased pain") are often manifested as nonlinear semantic coupling across sentence segments, the linear association between principal components is not sufficient to capture such high-order interactions. Based on this, the present application maps low-dimensional principal component vectors to high-dimensional feature space through the principal component kernel fine-grained association encoding operation to learn complex dynamic patterns between features. That is, this step uses the nonlinear expression ability of deep neural networks to convert the decoupled principal component pairs (such as time principal component and symptom principal component) into high-dimensional association coding vectors, so that the original linearly inseparable potential relationship (such as the indirect impact of time series changes on symptoms) is made explicit in the high-dimensional space. For example, by encoding the dynamic association between "reduction of exudate" and "increased pain", the principal component space and the graph structure space are bridged to provide a computable intermediate representation for topological associations.

[0045] Specifically, in the embodiment of the present application, the performance factor topological association matrix generation secondary subunit 1423-2 is used to calculate the performance factor between any two first-second named entity semantic embedding principal component fine-grained association coding vectors in the set of the first-second named entity semantic embedding principal component fine-grained association coding vectors to obtain the first-second named entity semantic embedding performance factor topological association matrix. This process can be expressed as follows:

[0046]

[0047] Among them, v j is the jth first-second named entity semantic embedding principal component fine-grained association encoding vector in the set of first-second named entity semantic embedding principal component fine-grained association encoding vectors, v i, and v j,k v i and v j The kth eigenvalue in v, n is i and v j The number of eigenvalues, s(v1,v1), s(v i ,vj ), s(v m ,v1),s(v1,v m ) and s(v m ,v m ) are the performance operators of each position in the topological association matrix of the first-second named entity semantic embedding performance factor, M A is the first-second named entity semantic embedding performance factor topological correlation matrix.

[0048] It should be understandable that the strength of implicit relationships needs to be dynamically quantified in combination with task objectives. For example, the association between "incision exudate" and "pain" may present different weights due to individual differences in patients. As a customized measurement function, the performance factor can extract task-related interaction strength from the first-second named entity semantic embedding principal component fine-grained association encoding vector, and construct an adjacency matrix to characterize the connection pattern between nodes. That is, the performance factor converts the semantic association of the first-second named entity semantic embedding principal component fine-grained association encoding vector (such as the synergy between the symptom principal component and the treatment principal component) into a numerical topological weight, forming a first-second named entity semantic embedding performance factor topological association matrix that reflects the potential interaction strength between entities. For example, by calculating the performance factor of the encoding vectors of "reduced exudate" and "increased pain", the confidence of their indirect association can be quantified, providing a structured relationship prior for graph convolution and avoiding the limitations of manual rule design.

[0049] Specifically, in the embodiment of the present application, the first-second named entity semantic convolutional coding secondary subunit 1423-3 is used to: perform graph convolutional coding on the set of principal component fine-grained association coding vectors of the first-second named entity semantic embeddings and the first-second named entity semantic embedding performance factor topological association matrix to obtain the first-second named entity semantic fine-grained relationship modeling matrix. This process can be expressed as follows:

[0050]

[0051] Among them, GCN is graph convolutional coding, m is the number of vectors in the set of first-second named entity semantic embedding principal component fine-grained association encoding vectors, M c It is the semantic fine-grained relationship modeling matrix between the first and second named entities.

[0052] It should be understandable that although the topological association matrix of the semantic embedding performance factor of the first-second named entity defines the strength of association, high-order interactions across nodes (such as the chain relationship of "reduced exudation-intensified pain-improved mobility") need to rely on the propagation of graph structure information. Through the graph convolutional network encoding operation, node features (principal component association coding) and topological association information can be aggregated, and local neighborhood relationships (such as the directly associated "redness-inflammation") and global high-order patterns (such as the "time-symptom-treatment" association across multi-hop nodes) are fused in message passing. The final output is the semantic fine-grained relationship modeling matrix between the first and second named entities. Through multi-level feature abstraction, it can restore the fine-grained semantic network between entities (such as the dynamic graph of symptom evolution driven by the time dimension), and achieve three-dimensional relationship reasoning capabilities at the level of manual review.

[0053] Preferably, in another example of the present application, the first-second named entity semantic convolutional coding secondary sub-unit 1423-3 is used to: based on the first-second named entity semantic embedding performance factor topological association matrix, perform cross-section function Gaussian optimization under the space-dominant topological field on each first-second named entity semantic embedding principal component fine-grained association coding vector in the set of the first-second named entity semantic embedding principal component fine-grained association coding vectors to obtain a set of first-second named entity semantic embedding principal component fine-grained association optimized coding vectors; perform graph convolution coding on the set of the first-second named entity semantic embedding principal component fine-grained association optimized coding vectors and the first-second named entity semantic embedding performance factor topological association matrix to obtain the first-second named entity semantic fine-grained relationship modeling matrix.

[0054] In particular, in the calculation process of the graph convolutional neural network model, the spatially dominant and spatially distributed forms will be used as the master and slave corresponding to the topological style space (i.e., v i →M A ). Given the first-second named entity semantic embedding principal component fine-grained association encoding vector v i ∈R 1×n Topological correlation matrix M with the first-second named entity semantic embedding performance factor A ∈R m×m There is a dimensional discontinuity between them, so it is necessary to construct the first-second named entity semantic embedding cross-section function matrix M Bi ∈R n×m As a space standard transition specification field. In addition, to correct the first-second named entity semantic embedding cross-section function matrix M Bi To account for the deviations caused by the disordered distribution caused by the random potential field, the following operations need to be performed:

[0055] Each first-second named entity semantic embedding is a principal component fine-grained association encoding vector v iInitialize the first-second named entity semantic embedding cross-section function matrix M with the corresponding Bi Perform matrix multiplication to generate each v i The corresponding first-second named entity semantic embedding cross-section compact vector v Bi , and then each v Bi Perform two-dimensional splicing to obtain the first-second named entity semantic embedding cross-section compact matrix M d :

[0056]

[0057] M d =(v B1 T ,v B2 T ,…,v Bi T ,…,v Bm T )

[0058] In this way, the first-second named entity semantic embedding cross-section compact matrix M is calculated d Topological correlation matrix M with the first-second named entity semantic embedding performance factor A Gaussian correlation coefficient, to iteratively update the first-second named entity semantic embedding cross-section function matrix M Bi , and drives it close to zero:

[0059]

[0060] Among them, σ 2 is the first-second named entity semantic embedding cross-section compact matrix M d The topological correlation matrix M with the first-second named entity semantic embedding performance factor A The variance of the set of all matrix values ​​;

[0061] By updating the first-second named entity semantic embedding cross-section function matrix M Bi 'Reconstruct the first-second named entity semantic embedding principal component fine-grained association encoding vector:

[0062]

[0063] Among them, v′ i It is v i The optimized first-second named entity semantic embedding principal component fine-grained association optimized encoding vector;

[0064] This method can effectively suppress the distribution disorder defects induced by random potential fields, thereby improving the topological spatial consistency and computational accuracy of graph convolution.

[0065] Subsequently, the set of optimized encoding vectors for the principal component fine-grained associations of the first and second named entity semantic embeddings and the topological association matrix of the first and second named entity semantic embedding performance factors are subjected to graph convolutional encoding to obtain a modeling matrix for the semantic fine-grained relationship between the first and second named entities. It should be understood that the optimized encoding vectors for the principal component fine-grained associations of the first and second named entity semantic embeddings, obtained after optimization, eliminate the distribution disorder induced by the random potential field in the original feature space through normalization and dimensionality reduction operations using the cross-section function matrix, making the encoding vector representation in the topological space more consistent and stable. Under this premise, the graph convolutional network uses the optimized encoding vectors for the principal component fine-grained associations of the first and second named entity semantic embeddings as node features and the topological association matrix of the first and second named entity semantic embedding performance factors as adjacency relationships, effectively aggregating semantic information from multi-order neighboring nodes during information transfer. Through multi-layer convolution operations, the model is able to capture not only direct associations (such as the explicit connection between a single pair of symptoms) but also implicit high-order associations across sentence segments (such as the causal chain where the number of days after surgery indirectly affects pain intensity through changes in exudate). In addition, the improvement of topological space consistency reduces the interference of redundant noise on the adjacency matrix, allowing the message passing mechanism of graph convolution to focus more on the essential association patterns between entities, thereby enhancing the model's ability to model complex medical logic (such as the dynamic balance between symptom chain reactions and treatment effects).

[0066] In an embodiment of the present application, the relationship type label determination subunit 1424 is used to determine the relationship type label between the first named entity and the second named entity based on the semantic fine-grained relationship modeling matrix between the first and second named entities. Specifically, in an embodiment of the present application, the relationship type label determination subunit is used to: use a classifier-based relationship generator to process the semantic fine-grained relationship modeling matrix between the first and second named entities to obtain the relationship type label between the first named entity and the second named entity. It should be understood that although the semantic fine-grained relationship modeling matrix between the first and second named entities contains rich entity semantic relationship information, it is only a numerical matrix and cannot directly reflect the relationship type that is easy for people to understand. The classifier can learn the mapping rules between different semantic relationship patterns and relationship types, and can effectively interpret and transform the numerical information in the matrix, map it to pre-defined relationship type categories, so that the relationship between entities is presented in the form of intuitive labels. Simply put, the classifier uses a fully connected layer to reorganize the features of the semantic fine-grained relationship modeling matrix between the first and second named entities, extracting discriminative patterns that are strongly correlated with the relationship type. The output layer uses a softmax function to calculate the probability distribution of each preset relationship type, and generates the final relationship label through maximum probability judgment. It is particularly important to note that the classifier is trained. During the training process, the classifier learns the nonlinear mapping function from the relationship modeling matrix to the label based on the labeled entity relationship samples, enabling it to identify the implicit clinical semantic associations between entities in the nursing scenario.

[0067] In summary, the named entity relationship extraction unit has been clearly explained. After obtaining a set of named entities, it first performs context-aware semantic embedding encoding on the named entities in the nursing documents to capture the semantic features of the entities. Then, targeting the time sensitivity and dynamic symptom association characteristics unique to orthopedic nursing, it uses a fine-grained interactive approach to project the semantic vectors of the two entities into an orthogonal feature space for principal component correlation analysis. By decoupling the explicit / implicit semantic association patterns between entities, the relationship between the named entities is intelligently derived. This effectively overcomes the shortcomings of traditional methods in parsing complex semantic coupling in free text, and can identify dynamic associations such as "incision exudate-time decay" and potential patterns such as "pain level-diurnal fluctuation." This provides technical support for establishing a three-dimensional relationship network across nursing records, significantly improving the accuracy of causal inference of clinical events and evaluation of the effectiveness of nursing measures.

[0068] In an embodiment of the present application, the negation word detection unit 143 is used to perform negation word detection on the set of named entities to obtain a set of named entity negation word detection results. Accordingly, the descriptions in nursing documents often contain negative expressions, such as "no exudate" and "no pain aggravation". The existence of these negative words greatly affects the semantic understanding of the text. If negation word detection is not performed, simply analyzing the named entities may lead to a misunderstanding of the actual situation represented by the entity, resulting in information interpretation deviation. For example, if the negative word "no" is ignored and "no exudate" is understood as "exudate", it will seriously mislead the patient's condition judgment. By performing negation word detection on the set of named entities, the positive or negative status of the description related to each named entity can be accurately judged, thereby improving the accuracy of the extracted information.

[0069] In particular, in a specific embodiment of the present application, the implementation process of performing negation word detection on the set of named entities to obtain a set of named entity negation word detection results includes: first, after completing the named entity recognition and obtaining the named entity set through the NLP model, each named entity and its text context are extracted. For example, for the text "The wound has no redness and swelling, suspected infection, and the patient complains of pain", the named entities such as "redness and swelling", "infection", and "pain" and their corresponding contexts "no redness and swelling", "suspected infection", and "complaint of pain" are extracted respectively. Then, these extracted text contexts are input into the trained NLP model. The model has been pre-trained for the recognition of negation words (such as "no", "not", "non", etc.), uncertain expressions (such as "suspected", "possible", "probably", etc.) and affirmative expressions (such as "complaint", "existence", etc.), and can analyze whether such expressions exist in the text and determine their association with the named entities. For example, when analyzing "no redness or swelling," the model identifies the negation word "none" and determines that it applies to the symptom / sign entity "redness or swelling." When analyzing "suspected infection," the model identifies the uncertain expression "suspected" and associates it with the symptom / sign entity "infection." When analyzing "chief complaint of pain," the model identifies the affirmative expression "chief complaint" and associates it with the symptom / sign entity "pain." The NLP model then determines the state of each named entity based on the type of expression detected. If a negation word is detected, the named entity's state is "negative" (e.g., "redness or swelling" is negated); if an uncertain expression is detected, the state is "uncertain" (e.g., "infection" is uncertain); if a positive expression or no special expression is detected (default affirmation), the state is "positive" (e.g., "pain" is affirmed). Finally, each named entity and its corresponding state (positive, negative, uncertain) are summarized to form a set of named entity negation detection results. This set contains information about the negation or affirmation of each named entity in the original text, for example, "redness or swelling - negative," "infection - uncertain," "pain - positive," etc.

[0070] In an embodiment of the present application, the attribute extraction unit 144 is used to extract attributes from the set of named entities to obtain a set of named entity attribute detection results. It should be understood that in nursing documents, named entities usually have multiple attributes, such as physiological attributes such as "body temperature" and "blood pressure" of patients, treatment-related attributes such as "surgery name" and "drug name", and pain-related attributes such as "nature of pain" and "degree". Understanding these attributes is crucial to fully and accurately understanding the object or concept represented by each named entity. For example, it is not enough to know the named entity "patient". It is also necessary to understand the patient's age, gender, condition and other attributes in order to better understand the patient's specific situation. After obtaining the set of named entity attribute detection results, detailed and accurate information about the named entity can be provided to medical staff, managers, scientific researchers, etc.

[0071] In particular, in a specific embodiment of the present application, the implementation process of performing attribute extraction on a set of named entities to obtain a set of named entity attribute detection results includes the following: first, after completing named entity recognition and obtaining a set of named entities (such as symptoms / signs, anatomical parts, etc.) through an NLP model, for each named entity, the descriptive content related to it is located in the nursing document text. For example, for the named entity "pain", the text is searched for expressions such as "the patient complained of sharp pain, with a pain score of VAS7"; for "wound exudate", the text is searched for expressions such as "the wound exudate is light yellow and the amount is relatively large". Next, the text fragment containing the named entities and their descriptions is input into a trained NLP model. This model has been pre-trained on the attribute features of different types of named entities (such as symptoms / signs, medical procedures, etc.) and can identify attribute-related keywords and semantic relationships in the text. For example, for the entity "pain", the model can identify "sharp pain" as a quality attribute and "VAS7" as a degree attribute; for "wound exudate", it can identify "light yellow" as a color attribute and "large" as a quantity attribute. Then, the NLP model extracts attribute information from the text through semantic analysis and pattern matching. Based on the language habits and logical relationships in the medical field, the model will determine which words or phrases belong to the attributes of a specific entity. For example, "sharp pain" and "dull pain" are often used to describe the nature of pain, "mild", "moderate", "heavy", and "score" are used to describe the degree of pain, and "color" and "amount" are used to describe the attributes of wound exudate. Finally, each named entity is associated with its corresponding attribute and summarized to form a set of named entity attribute detection results. This set contains the specific attribute data of each named entity, such as: "Pain - Nature: Sharp Pain, Degree: VAS7" and "Wound Exudate - Color: Light Yellow, Amount: More". These attribute information further enriches the connotation of the named entity, provides more detailed and comprehensive support for the in-depth analysis and application of subsequent nursing data, and helps medical staff to more accurately grasp the patient's condition and care details.

[0072] In an embodiment of the present application, the information integration unit 145 is used to integrate the set of named entities, the set of named entity relationship types, the set of named entity negation word detection results and the set of named entity attribute detection results to obtain the nursing document structured information. Accordingly, since the set of separate named entities, the set of named entity relationship types, the set of named entity negation word detection results or the set of named entity attribute detection results can only reflect one aspect of the nursing document information, it is impossible to fully display the semantics and knowledge contained in the document. Only by integrating them can the content of the nursing document be fully understood from multiple dimensions. For example, if you only know the named entity "fracture" and do not understand its relationship with other entities such as "surgery", whether there is a negation situation (such as "no fracture displacement") and related attributes (such as fracture location, degree, etc.), you cannot accurately grasp the patient's condition and nursing points. Therefore, by integrating the above four sets, a complete nursing document structured information model can be constructed. This model presents various information in nursing documents in a structured manner, including entities, relationships between entities, negations, and attributes, making the information clearer, more organized, and easier to understand and analyze. For example, within this structured information, one can intuitively see the patient's disease diagnosis, related symptoms, nursing measures taken, and the relationships between these elements, providing medical staff with comprehensive patient information.

[0073] In this embodiment of the present application, the standardization processing module 150 is used to standardize and deduplicate the structured information in the nursing documents to obtain standardized structured information. It should be understood that within the structured information in nursing documents, due to differences in recording habits, professional backgrounds, and terminology used by different medical staff, the same concept or information may be expressed in multiple different ways. For example, for the condition "wound redness and swelling," one medical staff may record it as "localized skin redness and swelling," while another may express it as "redness and swelling around the wound." This diversity makes it difficult to uniformly process and analyze information, thus necessitating standardization. Furthermore, during the recording and processing of nursing documents, duplicate records may occur. For example, medical staff on different shifts may repeatedly record certain information about the same patient during shift handovers, or there may be overlap in some information between different nursing record documents. This duplicate data not only takes up storage space but also interferes with subsequent data analysis and utilization, affecting processing efficiency, necessitating deduplication.

[0074] In particular, in a specific embodiment of the present application, the implementation process of standardizing and deduplicating the structured information of nursing documents to obtain standard structured information for nursing documents includes the following: Standardization is an important step in unifying data representation and eliminating ambiguity. First, term mapping is performed to compare the terms in the structured information with an authoritative standard terminology system (such as SNOMED CT or an internally defined standard dictionary). In orthopedic nursing documents, different medical staff may have different expressions for the same clinical concept, such as "redness and swelling" and "localized congestion." Using the terminology system to confirm that these refer to the same pathological condition, standardization is achieved to avoid misunderstandings caused by differences in expression. Second, unit unification is performed, standardizing the units of numerical information. For example, a temperature record of "38.5°C" and "38.5 degrees Celsius" are uniformly converted to "°C"; a pain score of "VAS 7 points" is standardized to "Pain intensity (VAS score) 7" to ensure consistent units and format for easier statistical analysis. After standardization is completed, deduplication processing begins to eliminate data redundancy and integrate related information. Deduplication within the same record requires checking whether duplicate content is extracted from the structured information of a single nursing document. For example, if a text entry contains "VAS pain score 7" twice, the system uses keyword matching or data comparison to retain only one valid record, avoiding meaningless duplication and improving data storage efficiency and readability. Information across different records is merged. For information on the same event or status in different nursing documents for the same patient (such as medical records and shift handover reports), the system links them using the patient ID, record timestamp, and entity content. For example, if one record mentions "small amount of wound exudate" and another mentions "light yellow wound exudate," the system merges them into "small amount of wound exudate, light yellow color" to form a complete description. If different records have conflicting descriptions of the same entity, such as one stating "no redness or swelling" and another stating "redness or swelling," the most recent record is prioritized or clinical logic (such as whether there are follow-up measures) is used to select the most accurate information to ensure data authenticity and reliability. After standardization and deduplication, format conversion is required to convert the processed structured information into a format suitable for storage and querying. Depending on the data characteristics and usage requirements, this can be converted into relational database tables, NoSQL documents, or graph database nodes / edges. For example, data is organized into a table format containing fields such as patient ID, entity name, attributes, standardized values, and recording time, in preparation for storage in a structured cloud database. This conversion ensures that the stored data can be efficiently retrieved and accessed, meeting various needs such as clinical decision-making and scientific research analysis.

[0075] In an embodiment of the present application, the data storage module 160 is used to store the standard structured information of the nursing documents in the structured database on the cloud. That is, the amount of standard structured information obtained after the nursing documents are processed is huge, and needs to be preserved for a long time for subsequent reference, analysis and utilization. The structured database on the cloud has a strong storage capacity, which can meet the storage needs of large amounts of data, and can effectively organize and manage the data, making it convenient for medical staff and researchers to quickly locate and obtain the required information. For example, as time goes by, the nursing document data accumulated by the hospital will increase, and the cloud database can easily cope with the data growth without the problem of insufficient storage capacity.

[0076] In summary, the cloud-based orthopedic nursing intelligent data processing system 100 according to the embodiment of the present application is described. It first transfers the newly added nursing documents to the temporary storage area in the cloud and reads them. Then, it performs format cleaning and text segmentation on the nursing documents to generate a sequence distribution of sentence-granular descriptions. The sequence distribution is then input into the NLP model deployed in the cloud to complete named entity recognition, relationship extraction, negation word detection, and attribute extraction to obtain nursing document structured information. The structured information is then standardized and deduplicated to generate standard structured information of the nursing documents, and stored in a structured database in the cloud. In this way, the efficiency and accuracy of nursing data processing can be improved, and the deep value of data can be released.

Claims

1. The cloud platform-based orthopedic nursing intelligent data processing system is characterized by: include: Nursing document storage module, used to transfer newly added nursing documents to the temporary storage area in the cloud; A nursing document reading module, configured to read the newly added nursing document from the temporary storage area; A nursing document sentence segmentation module is used to perform format cleaning and text segmentation on the newly added nursing documents to obtain a sequence distribution of granular descriptions of nursing document sentences; A structured processing module, configured to input the sequence distribution of the granular description of the nursing document into the NLP model deployed on the cloud to perform named entity recognition, relationship extraction, negation word detection, and attribute extraction to obtain structured information of the nursing document; A standardization processing module, configured to perform standardization and deduplication processing on the nursing document structured information to obtain standard nursing document structured information; The data storage module is used to store the standard structured information of the nursing document in the structured database on the cloud.

2. The cloud platform-based orthopedic nursing intelligent data processing system according to claim 1 is characterized in that: The structured processing module includes: A named entity recognition unit, configured to perform named entity recognition on the sequence distribution of the granular description of the nursing document sentences based on the NLP model to obtain a set of named entities; a named entity relationship extraction unit, configured to perform relationship extraction on the set of named entities to obtain a set of named entity relationship types; a negation detection unit, configured to perform negation detection on the set of named entities to obtain a set of named entity negation detection results; an attribute extraction unit, configured to extract attributes from the set of named entities to obtain a set of named entity attribute detection results; An information integration unit is used to integrate the set of named entities, the set of named entity relationship types, the set of named entity negation word detection results, and the set of named entity attribute detection results to obtain the nursing document structured information.

3. The cloud platform-based orthopedic nursing intelligent data processing system according to claim 2 is characterized in that: The named entity relationship extraction unit includes: A first and second named entity extraction subunit, configured to extract a first named entity and a second named entity from the set of named entities; a named entity semantic embedding subunit, configured to perform semantic embedding coding on the first named entity and the second named entity to obtain a first named entity semantic embedding coding vector and a second named entity semantic embedding coding vector; A first-second named entity fine-grained modeling subunit is configured to perform a performance-oriented named entity semantic principal component association analysis on the first named entity semantic embedding coding vector and the second named entity semantic embedding coding vector to obtain a semantic fine-grained relationship modeling matrix between the first and second named entities; The relationship type label determination subunit is used to determine the relationship type label between the first named entity and the second named entity based on the semantic fine-grained relationship modeling matrix between the first and second named entities.

4. The cloud platform-based orthopedic nursing intelligent data processing system according to claim 3 is characterized in that: The named entity semantic embedding subunit is used to: perform BioBERT-based semantic embedding encoding on the first named entity and the second named entity to obtain the first named entity semantic embedding encoding vector and the second named entity semantic embedding encoding vector.

5. The cloud platform-based orthopedic nursing intelligent data processing system according to claim 3 is characterized in that: The first-second named entity fine-grained modeling sub-unit includes: A second-level subunit for principal component association analysis between the first and second named entities, configured to perform principal component fine-grained association analysis on the first named entity semantic embedding coding vector and the second named entity semantic embedding coding vector to obtain a set of first-second named entity semantic embedding principal component fine-grained association coding vectors; A performance factor topological association matrix generation secondary subunit is used to calculate the performance factor between any two first-second named entity semantic embedding principal component fine-grained association coding vectors in the set of the first-second named entity semantic embedding principal component fine-grained association coding vectors to obtain a first-second named entity semantic embedding performance factor topological association matrix; The second-level sub-unit of semantic convolutional coding between the first and second named entities is used to perform graph convolutional coding on the set of principal component fine-grained association coding vectors of the semantic embedding of the first and second named entities and the topological association matrix of the semantic embedding performance factor of the first and second named entities to obtain the semantic fine-grained relationship modeling matrix between the first and second named entities.

6. The cloud platform-based orthopedic nursing intelligent data processing system according to claim 5 is characterized in that: The second-level sub-unit of principal component association analysis between the first and second named entities is used to: Performing feature principal component analysis on the first named entity semantic embedding coding vector and the second named entity semantic embedding coding vector to obtain a set of first named entity semantic embedding principal component coding vectors and a set of second named entity semantic embedding principal component coding vectors; The set of the first named entity semantic embedding principal component coding vectors and the set of the second named entity semantic embedding principal component coding vectors are respectively subjected to principal component kernel fine-grained association coding for each corresponding group of the first named entity semantic embedding principal component coding vectors and the second named entity semantic embedding principal component coding vectors to obtain the set of the first-second named entity semantic embedding principal component fine-grained association coding vectors.

7. The cloud platform-based orthopedic nursing intelligent data processing system according to claim 5 is characterized in that: The first-second named entity semantic convolutional encoding secondary sub-unit is used to: Based on the first-second named entity semantic embedding performance factor topological association matrix, performing cross-section function Gaussian optimization under a space-dominant topological field on each first-second named entity semantic embedding principal component fine-grained association encoding vector in the set of the first-second named entity semantic embedding principal component fine-grained association encoding vectors to obtain a set of first-second named entity semantic embedding principal component fine-grained association optimized encoding vectors; The set of the first-second named entity semantic embedding principal component fine-grained association optimization encoding vectors and the first-second named entity semantic embedding performance factor topological association matrix are graph convolutionally encoded to obtain a semantic fine-grained relationship modeling matrix between the first-second named entities.

8. The cloud platform-based orthopedic nursing intelligent data processing system according to claim 7 is characterized in that: The relationship type label determination subunit is used to: use a classifier-based relationship generator to process the semantic fine-grained relationship modeling matrix between the first and second named entities to obtain a relationship type label between the first named entity and the second named entity.

Citation Information

Cited By

  • Nursing teaching task-oriented nursing field text annotation corpus construction method

    CN122020186A