Search

CN-121979976-A - Neurosurgery nursing record key event extraction method and system based on natural language processing

CN121979976ACN 121979976 ACN121979976 ACN 121979976ACN-121979976-A

Abstract

The invention provides a neurosurgery nursing record key event extraction method and system based on natural language processing, and relates to the technical field of artificial intelligence and natural language processing, wherein the method comprises the steps of identifying candidate professional terms by utilizing a term library in the neurosurgery field based on a word vector sequence; extracting core semantic vectors based on candidate technical terms, associating corresponding entity type embedded vectors with event type embedded vectors, and carrying out multi-mode fusion and nonlinear transformation on the core semantic vectors, the entity type embedded vectors and the event type embedded vectors to obtain a semantic enhanced term sequence. The invention automatically converts unstructured neurosurgery nursing records into structured key event reports capable of directly supporting clinical decisions, and solves the core problems of fuzzy semantics, fragmented information and poor practicability in the original text.

Inventors

  • Xue Daisi

Assignees

  • 首都医科大学宣武医院

Dates

Publication Date
20260505
Application Date
20251223

Claims (10)

  1. 1. A method for extracting key events of neurosurgery nursing records based on natural language processing, which is characterized in that the method comprises the following steps: Step 100, identifying candidate professional terms by utilizing a term library in the neurosurgery field based on a word vector sequence, extracting core semantic vectors of the candidate professional terms, associating corresponding entity type embedded vectors with event type embedded vectors, and carrying out multi-mode fusion and nonlinear transformation on the core semantic vectors, the entity type embedded vectors and the event type embedded vectors to obtain a term sequence with enhanced semantics; Step 200, constructing an iso-graph fusing syntax and semantic information based on a semantic enhanced term sequence, carrying out importance evaluation and structure division on the heterogeneous graph to obtain topological features and measurement indexes, obtaining attention weight coefficients based on the topological features and the measurement indexes, respectively extracting structural features and key semantic features based on the iso-graph and the attention weight coefficients, dynamically optimizing the fusion weights of the structural features and the key semantic features, and carrying out weighted fusion to obtain optimized key event features; step 300, carrying out normalized mapping based on the optimized key event characteristics to obtain a standardized event, identifying a core event sentence in the standardized event, and matching and completing the missing argument based on semantic similarity; step 400, carrying out fine-granularity event classification based on the complete event argument list to obtain event type labels, identifying and extracting structural association information of each event based on the event type labels, integrating the event type labels with the structural association information to obtain structural key event data; step 500, generating and outputting a key event report based on the structured key event data.
  2. 2. The natural language processing based neurosurgical care record key event extraction method of claim 1, further comprising, prior to step 100: preprocessing the neurosurgery nursing record text to generate a standardized text; And carrying out context semantic coding on the standardized text to obtain a word vector sequence containing the context dynamic semantics.
  3. 3. The natural language processing based neurosurgical care record key event extraction method as claimed in claim 2, wherein step 100 comprises: traversing the word vector sequence, and sequentially extracting context word vectors of each word as the semantics of each word; comparing the semantics of each word with the semantic vectors of the standard terms, abbreviations and synonyms predefined in the neurosurgery field term library one by one, and generating a similarity matching value set for each word by calculating the semantic similarity among the vectors; Screening out words corresponding to all matching values exceeding a preset similarity threshold from the similarity matching value set, and marking the words as candidate professional terms; and simultaneously, inquiring and acquiring an entity type embedded vector and an event type embedded vector corresponding to the candidate professional terms according to a predefined mapping relation in a neurosurgery field term library.
  4. 4. The natural language processing based neurosurgical care record key event extraction method as claimed in claim 3, wherein step 100 further comprises: For each candidate technical term, splicing the core semantic vector, the entity type embedded vector and the event type embedded vector to form a multi-mode fusion vector; inputting each multi-mode fusion vector into a fully-connected neural network layer to execute linear transformation so as to obtain corresponding linear characteristics; inputting the linear characteristic into a nonlinear activation function to process to obtain a corresponding activation characteristic; based on the activation feature, converting each multi-modal fusion vector into a corresponding semantic enhancement vector; and arranging the semantic enhancement vectors of all candidate technical terms according to the sequence of the semantic enhancement vectors in the text to form a semantic enhanced term sequence.
  5. 5. The natural language processing based neurosurgical care record key event extraction method as claimed in claim 4, wherein step 200 comprises: Analyzing the syntactic structure among the terms in the sequence based on the semantically enhanced term sequence to obtain syntactic relation edges among the terms; each term in the semantically enhanced term sequence is used as a graph node, and the syntactic relation side and the semantic relation side are used as sides for connecting the nodes, so that an abnormal composition for fusing syntactic and semantic information is constructed; carrying out importance evaluation on all nodes in the heterogram, calculating the centrality value and the semantic contribution value of each node, and generating a measurement index; based on node importance distribution reflected by the measurement index and the strength of the dependency relationship among the nodes, sub-dividing the heterogeneous graph to obtain a plurality of internal closely-related sub-areas; For each subarea, calculating the distribution density of the nodes and the connectivity of the edges of the subareas respectively, and extracting the distribution density and the connectivity of the edges as topological features; Based on the topological features and the measurement indexes, comprehensive calculation and mapping are performed through a trainable parameterized function, so that attention weight coefficients for dynamically adjusting feature fusion proportion are obtained.
  6. 6. The natural language processing based neurosurgical care record key event extraction method as claimed in claim 5, wherein step 200 further comprises: Inputting the heterogeneous graph into a graph neural network based on the heterogeneous graph and the attention weight coefficient, and extracting structural features containing graph topology information; The semantic enhancement vector corresponding to each node in the heterogram is used as a node feature set, the node feature set and the attention weight coefficient are input into an attention network together, and key semantic features of focus key semantics are extracted; based on the attention weight coefficient, dynamically modulating the contribution weight of the structural feature and the key semantic feature in the fusion process, and determining the final fusion weight; and according to the final fusion weight, performing weighted fusion on the structural feature and the key semantic feature to obtain the optimized key event feature.
  7. 7. The method of natural language processing based neurosurgical care record key event extraction of claim 6, wherein step 300 comprises: the method comprises the steps of optimizing key event characteristics, decoding corresponding event text expression, matching the event text expression with a neurosurgery standard medical term library, identifying nonstandard expression and spoken expression in the event text expression, and converting the nonstandard expression and the spoken expression into corresponding standard medical terms according to a predefined mapping rule in the neurosurgery standard medical term library to obtain a standardized event; Detecting and processing the standardized event through a key sentence; counting the number of standardized events in each sentence as a first index, calculating the ratio of key argument to total word number in each sentence as a second index, and evaluating the completeness of semantic structure of each sentence as a third index; the first index, the second index and the third index are weighted and summed according to preset weights to obtain a decision value of each sentence; Extracting event arguments which are contained in the adjacent sentences and are missing in the current core event sentence from the adjacent sentences with the semantic similarity higher than a preset threshold value; and integrating all the standardized events, the core event sentences and the missing event arguments to obtain a complete event argument list.
  8. 8. The neurosurgical care record key event extraction method based on natural language processing as claimed in claim 7, wherein step 400 comprises: Based on the complete event argument list, analyzing and classifying argument constitution and semantic features of each event by an event classifier, and distributing an event type label with fine granularity for each event; identifying and extracting quantitative indexes, occurrence time and symptom description information associated with each event by an information extraction unit aiming at each event type label and the corresponding event argument thereof; And aligning and combining the event type label distributed by each event with the quantization index, the occurrence time and the symptom description information of the corresponding event to obtain the structured key event data.
  9. 9. The method for extracting critical events from a neurosurgical care record based on natural language processing as claimed in claim 8, wherein the step 500 comprises: classifying and sequencing the data according to preset report dimension and format requirements based on the structured key event data, and organizing the data into an ordered report data set; Filling a predefined key event report template according to the report data set to obtain tabular event data; And converting the tabulated event data into a data format conforming to the interface specification of the clinical information system, and outputting the data as a key event report.
  10. 10. A neurosurgical care record key event extraction system based on natural language processing, the system implementing the method of any one of claims 1 to 9, comprising: The semantic enhancement module is used for identifying candidate technical terms by utilizing a neurosurgery field term library based on the word vector sequence, extracting core semantic vectors of the candidate technical terms based on the candidate technical terms, and associating corresponding entity type embedded vectors and event type embedded vectors; The system comprises an optimization module, a topological feature and a measurement index, wherein the optimization module is used for constructing an iso-composition of fusion syntax and semantic information based on a semantic enhanced term sequence, carrying out importance evaluation and structure division on the heterogeneous graph to obtain a topological feature and the measurement index, obtaining an attention weight coefficient based on the topological feature and the measurement index, respectively extracting a structural feature and a key semantic feature based on the iso-composition and the attention weight coefficient, dynamically optimizing the fusion weight of the structural feature and the key semantic feature, and carrying out weighted fusion to obtain an optimized key event feature; the normalization and completion module is used for carrying out normalization mapping based on the optimized key event characteristics to obtain a standardized event, identifying a core event sentence in the standardized event, and matching and completing the missing argument based on semantic similarity; The classification and extraction module is used for classifying the fine-granularity event based on the complete event argument list to obtain event type labels, identifying and extracting structural association information of each event based on the event type labels, integrating the event type labels with the structural association information to obtain structural key event data; And the output module is used for generating and outputting a key event report based on the structured key event data.

Description

Neurosurgery nursing record key event extraction method and system based on natural language processing Technical Field The invention relates to the technical field of artificial intelligence and natural language processing, in particular to a neurosurgery nursing record key event extraction method and system based on natural language processing. Background Neurosurgical Intensive Care Units (ICU) generate a large number of unstructured care books daily, which contain key information to assess the condition and guide the treatment. Currently, clinicians rely on manual screening from massive text to generalize key events, which are inefficient and may have difficulty meeting the timeliness requirements of emergency treatment in the event of sudden disease changes. While Natural Language Processing (NLP) technology offers the possibility of automated extraction, the generic model has the following drawbacks at the data processing level when dealing with neurosurgical specialized text: Firstly, the data understanding layer has low recognition rate on a large number of field terms and abbreviations such as GCS score, cerebral hernia and the like, and cannot establish semantic association of context, such as association of blood pressure drop and headache relief, possibly resulting in information fragmentation, secondly, the data standardization layer has poor adaptability on various expression forms (such as ICP25cmH 2 O, intracranial pressure is higher) and spoken language description in nursing records, possibly resulting in inconsistent extraction results, and finally, the data output and application layer output by the existing method has coarse event classification granularity, is not bound with specific indexes and timestamp depth, is mostly in offline batch processing, cannot provide real-time and structured decision support, for example, in suspected cases of cerebral hernia, the prior art can not automatically associate gradual change of slightly large pupils to loose and fixed with fluctuation trend of blood pressure, and cannot generate a coherent illness state evolution evidence chain rapidly. Disclosure of Invention The technical problem to be solved by the invention is to provide a neurosurgery nursing record key event extraction method and system based on natural language processing, which automatically converts unstructured neurosurgery nursing records into structured key event reports capable of directly supporting clinical decisions, and solves the core problems of fuzzy semantics, information fragmentation and poor practicability in an original text. In order to solve the technical problems, the technical scheme of the invention is as follows: The method comprises the steps of identifying candidate professional terms by utilizing a neurosurgery field term library based on a word vector sequence, extracting core semantic vectors of the candidate professional terms, associating corresponding entity type embedded vectors with event type embedded vectors, carrying out multi-mode fusion and nonlinear transformation on the core semantic vectors, the entity type embedded vectors and the event type embedded vectors to obtain a semantic enhanced term sequence; The method comprises the steps of establishing a fusion syntax and a different composition of semantic information based on a semantic enhanced term sequence, carrying out importance evaluation and structure division on the different composition to obtain topological features and measurement indexes, obtaining attention weight coefficients based on the topological features and the measurement indexes, respectively extracting structural features and key semantic features based on the different composition and the attention weight coefficients, dynamically optimizing the fusion weights of the structural features and the key semantic features, and carrying out weighted fusion to obtain optimized key event features; The method comprises the steps of optimizing key event characteristics, carrying out normalization mapping based on the optimized key event characteristics to obtain a standardized event, identifying a core event sentence in the standardized event, and matching and complementing missing arguments based on semantic similarity; Based on the complete event argument list, carrying out fine-granularity event classification to obtain event type labels, identifying and extracting structural association information of each event based on the event type labels, integrating the event type labels with the structural association information to obtain structural key event data; And generating and outputting a key event report based on the structured key event data. In a second aspect, a natural language processing based neurosurgical care record key event extraction system comprises: The semantic enhancement module is used for identifying candidate technical terms by utilizing a neurosurgery field term library based on the word vector sequence, extra