A method for evaluating human clinical information based on natural language processing algorithm

Abstract

The present invention discloses a method for evaluating human clinical information based on a natural language processing algorithm, including dependency syntactic analysis, analyzing input text sentences to obtain the syntactic structure of the sentences, from grammatical analysis to sequence labeling, through the "bidirectional long short-term memory conditional random field" model Bi-LSTM-CRF, the tuple extraction task is converted into a sequence labeling task, the application field of the method for evaluating human clinical information based on a natural language processing algorithm, and a knowledge graph is established, which is summarized into five steps, namely, knowledge representation, extraction, fusion, reasoning, and quality assessment. The present invention realizes dependency syntactic analysis, and uses the output result of syntactic analysis as input to obtain more indication information to realize the process from grammatical analysis to sequence labeling. On the one hand, it is the demand of language understanding itself, and syntactic analysis is an important part of language understanding. On the other hand, it also provides support for other natural language processing tasks, increasing the application field of natural language processing algorithms.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for evaluating human clinical information based on a natural language processing algorithm. Background Art

[0002] With the rapid development of computer and Internet technology, people's daily lives are increasingly inseparable from information and data. This is especially true in the medical industry, where medical big data plays an increasingly important role in people's health lives. Data is the lifeline of the healthcare industry. From blood pressure readings and surgical records to insurance claims, immunization history, patient demographics, and payment receipts, every action of every member in the healthcare ecosystem relies on an endless stream of information. At present, the commonly used components for processing data streams are: kafka, flink, storm, spark streaming, etc., but the existing technology lacks congestion control solutions for data, especially congestion control solutions for medical big data. Therefore, when the amount of data is too large, it may cause a data flow node to crash, resulting in data congestion, and the processing efficiency of medical data is low. Therefore, we make improvements to this and propose a human clinical information evaluation method based on natural language processing algorithms. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0004] The present invention provides a human clinical information evaluation method based on a natural language processing algorithm, comprising:

[0005] Dependency syntactic analysis: analyzing the input text sentences to obtain the syntactic structure of the sentences;

[0006] From syntax analysis to sequence labeling, the tuple extraction task is transformed into a sequence labeling task through the "Bidirectional Long Short-Term Memory Conditional Random Field" model Bi-LSTM-CRF;

[0007] Application areas of human clinical information evaluation methods using natural language processing algorithms;

[0008] Building a knowledge graph can be summarized into five steps, namely knowledge representation, extraction, fusion, reasoning, and quality assessment. For the medical field, the first three steps are the core steps.

[0009] As a preferred technical solution of the present invention, the dependency syntactic analysis is, on the one hand, a requirement of language understanding itself, and syntactic analysis is an important part of language understanding; on the other hand, it also provides support for other natural language processing tasks.

[0010] As a preferred technical solution of the present invention, the Bi-LSTM-CRF is a bidirectional LSTM plus a CRF layer. In this network model, it is important to understand the concept of CRF. LSTM here actually serves as a feature function for CRF.

[0011] As a preferred technical solution of the present invention,

[0012] The knowledge representation includes improving the ability to integrate data and perform downstream knowledge reasoning tasks;

[0013] The knowledge extraction includes entity, relationship and attribute extraction;

[0014] The knowledge fusion includes data integration, disambiguation and processing of knowledge from different sources under the same framework specification.

[0015] As a preferred technical solution of the present invention, the knowledge fusion also includes a medical knowledge graph construction framework and the crawling and application of medical big data. The steps of crawling and applying the medical big data are as follows:

[0016] Use the Internet to search for reliable and authoritative medical data sources, and use multi-threading technology to crawl all data in sections according to the department to which the disease belongs to, to ensure that all disease contents are covered. During the data crawling process, analyze the data structure, including department, symptoms, cause, complications, treatment, prevention, etc.;

[0017] Segment the crawled data for subsequent algorithm design, and finally store all the data in a structured manner in the local MySQL database;

[0018] Use the Elastic Search database to quickly store, search, and analyze massive amounts of data. Extract some of the data stored in MySQL and store it in the Elastic Search database, then develop the corresponding interface to return data in JSON format. The main functions of the ES database are: first, provide real-time search results when the user enters a symptom word in the input box; second, search for all related symptoms when clicking on a part;

[0019] Using Neo4j graph database to store some structured data facilitates algorithm design, and combines it with algorithms such as word segmentation, retrieval, exclusion, and statistics to improve the accuracy of diagnosis.

[0020] As a preferred technical solution of the present invention, the application fields of the desensitization system include clinical decision support system and information retrieval system. The clinical decision support system includes

[0021] With the help of medical knowledge graphs, medical decision support systems can provide intelligent diagnosis, treatment plan recommendations and referral guidelines based on patient symptom descriptions and laboratory test data;

[0022] Analyze doctors' diagnosis and treatment plans, find out any gaps and fill them up, and reduce or even avoid misdiagnosis;

[0023] Human-computer interaction systems require NLP technology to implement, including question-answering systems, semantic analysis, and speech recognition;

[0024] The information retrieval system comprises:

[0025] Establish a medical information search engine: Compared with traditional search engines, search engines based on knowledge graphs can better understand user semantics, optimize user questions, and expand and rewrite queries based on knowledge graphs;

[0026] Establish a medical question-and-answer system: The most common function of a patient-oriented medical question-and-answer system is to provide patients with popular medical knowledge, mainly structured knowledge. The main function of the question-and-answer system is to assist education and provide accurate structured knowledge query. It can also assist scientific research tasks, such as Alibaba's clinical medicine research assistance platform.

[0027] The beneficial effects of the present invention are as follows: the human clinical information evaluation method based on natural language processing algorithm has the following advantages:

[0028] Dependency syntactic analysis uses the output of syntactic analysis as input to obtain more indicative information. From grammatical analysis to sequence labeling, on the one hand, it is the demand of language understanding itself, and syntactic analysis is an important part of language understanding. On the other hand, it also provides support for other natural language processing tasks. It expands the application field of natural language processing algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0030] In the attached picture:

[0031] Figure 1 It is a flow chart from grammatical analysis to sequence labeling of the present invention;

[0032] Figure 2 It is the business-driven knowledge graph construction framework of the present invention. DETAILED DESCRIPTION

[0033] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0034] Example: Figure 1-2As shown, the present invention provides a human clinical information evaluation method based on a natural language processing algorithm, comprising:

[0035] Dependency syntactic analysis: analyzing the input text sentences to obtain the syntactic structure of the sentences;

[0036] From syntax analysis to sequence labeling, the tuple extraction task is transformed into a sequence labeling task through the "Bidirectional Long Short-Term Memory Conditional Random Field" model Bi-LSTM-CRF;

[0037] Application areas of human clinical information evaluation methods using natural language processing algorithms;

[0038] Building a knowledge graph can be summarized into five steps, namely knowledge representation, extraction, fusion, reasoning, and quality assessment. For the medical field, the first three steps are the core steps.

[0039] Among them, dependency syntactic analysis is, on the one hand, the demand of language understanding itself. Syntactic analysis is an important part of language understanding. On the other hand, it also provides support for other natural language processing tasks.

[0040] Among them, Bi-LSTM-CRF is a bidirectional LSTM plus a CRF layer. In this network model, it is important to understand the concept of CRF. LSTM here actually serves as a feature function for CRF.

[0041] in,

[0042] Knowledge representation includes improving data integration capabilities and performing downstream knowledge reasoning tasks;

[0043] Knowledge extraction includes entity, relationship and attribute extraction;

[0044] Knowledge fusion involves data integration, disambiguation and processing of knowledge from different sources under the same framework specifications.

[0045] Among them, knowledge fusion also includes the construction framework of medical knowledge graph and the crawling and application of medical big data. The steps of crawling and applying medical big data are as follows:

[0046] Use the Internet to search for reliable and authoritative medical data sources, and use multi-threading technology to crawl all data in sections according to the department to which the disease belongs to, to ensure that all disease contents are covered. During the data crawling process, analyze the data structure, including department, symptoms, cause, complications, treatment, prevention, etc.;

[0047] Segment the crawled data for subsequent algorithm design, and finally store all the data in a structured manner in the local MySQL database;

[0048] Use the Elastic Search database to quickly store, search, and analyze massive amounts of data. Extract some of the data stored in MySQL and store it in the Elastic Search database, then develop the corresponding interface to return data in JSON format. The main functions of the ES database are: first, provide real-time search results when the user enters a symptom word in the input box; second, search for all related symptoms when clicking on a part;

[0049] Using Neo4j graph database to store some structured data facilitates algorithm design, and combines it with algorithms such as word segmentation, retrieval, exclusion, and statistics to improve the accuracy of diagnosis.

[0050] Among them, the application areas of desensitization systems include clinical decision support systems and information retrieval systems. Clinical decision support systems include

[0051] With the help of medical knowledge graphs, medical decision support systems can provide intelligent diagnosis, treatment plan recommendations and referral guidelines based on patient symptom descriptions and laboratory test data;

[0052] Analyze doctors' diagnosis and treatment plans, find out any gaps and fill them up, and reduce or even avoid misdiagnosis;

[0053] Human-computer interaction systems require NLP technology to implement, including question-answering systems, semantic analysis, and speech recognition;

[0054] The information retrieval system includes:

[0055] Establish a medical information search engine: Compared with traditional search engines, search engines based on knowledge graphs can better understand user semantics, optimize user questions, and expand and rewrite queries based on knowledge graphs;

[0056] Establish a medical question-and-answer system: The most common function of a patient-oriented medical question-and-answer system is to provide patients with popular medical knowledge, mainly structured knowledge. The main function of the question-and-answer system is to assist education and provide accurate structured knowledge query. It can also assist scientific research tasks, such as Alibaba's clinical medicine research assistance platform.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention is described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for evaluating human clinical information based on a natural language processing algorithm, characterized in that: include: Dependency syntactic analysis: analyzing the input text sentences to obtain the syntactic structure of the sentences; From syntax analysis to sequence labeling, the tuple extraction task is transformed into a sequence labeling task through the "Bidirectional Long Short-Term Memory Conditional Random Field" model Bi-LSTM-CRF; Identify the application areas of human clinical information assessment methods using natural language processing algorithms; Building a knowledge graph can be summarized into five steps, namely knowledge representation, extraction, fusion, reasoning, and quality assessment. For the medical field, the first three steps are the core steps; Knowledge representation includes improving data integration capabilities and performing downstream knowledge reasoning tasks; Knowledge extraction includes entity, relationship and attribute extraction; Knowledge fusion includes data integration, disambiguation and processing of knowledge from different sources under the same framework specification; The knowledge fusion also includes the construction of a medical knowledge graph framework and the crawling and application of medical big data. The steps of crawling and applying medical big data are as follows: Use the Internet to search for reliable and authoritative medical data sources, and use multi-threading technology to crawl all data in sections according to the department to which the disease belongs to, to ensure that all disease contents are covered. During the data crawling process, analyze the data structure, including department, symptoms, cause, complications, treatment, and prevention; Segment the crawled data for subsequent algorithm design, and finally store all the data in a structured manner in the local MySQL database; The Elastic Search database is used to quickly store, search and analyze massive amounts of data. Part of the data stored in MySQL is extracted and stored in the Elastic Search database. Then the corresponding interface is developed to return data in JSON format. The functions of the Elastic Search database are: first, it provides real-time search results when the user enters symptom words in the input box; second, it searches for all related symptoms when clicking on a part; Use Neo4j graph database to store some structured data to facilitate algorithm design, and use word segmentation, retrieval, exclusion, and statistical algorithms to improve the accuracy of diagnosis; The application fields of the desensitization system include clinical decision support system and information retrieval system. The clinical decision support system includes: With the help of medical knowledge graph, the medical decision support system provides intelligent diagnosis, treatment plan recommendations and referral guidelines based on the patient's symptom description and laboratory data; Analyze doctors' diagnosis and treatment plans, find out any gaps and fill them up, and reduce or even avoid misdiagnosis; Human-computer interaction systems require NLP technologies to be implemented, including question-answering systems, semantic analysis, and speech recognition; The information retrieval system comprises: Establish a medical information search engine: Compared with traditional search engines, search engines based on knowledge graphs can better understand user semantics, optimize user questions, and expand and rewrite queries based on knowledge graphs; Establish a medical question-and-answer system: The most common function of a patient-oriented medical question-and-answer system is to provide patients with popular medical knowledge, mainly structured knowledge. The function of the question-and-answer system is to assist education, provide accurate structured knowledge query, and also assist in scientific research tasks.

2. A human clinical information evaluation method based on natural language processing algorithm according to claim 1, characterized in that: The dependency syntactic analysis is, on the one hand, a requirement of language understanding itself, and syntactic analysis is an important part of language understanding. On the other hand, it also provides support for other natural language processing tasks.

3. A human clinical information evaluation method based on natural language processing algorithm according to claim 1, characterized in that: The Bi-LSTM-CRF is a bidirectional LSTM plus a CRF layer. In this network model, it is important to understand the concept of CRF. LSTM here actually acts as a feature function for CRF.

Images