System for predicting the likelihood of disease occurrence and health impact in a population

Abstract

A system for predicting the likelihood of disease occurrence and health impact of a population group according to an embodiment of the present invention comprises: a data collection module that collects and stores a plurality of documents containing unstructured raw data regarding the environment and health from an external source; an input module that receives a natural language-based query containing the key term from a user; a preprocessing module that preprocesses whitespace, punctuation, and noise in the form of stop words within the text of the query; an SR execution and query reconstruction module that extracts a key document among the plurality of documents collected by the data collection module and reconstructs a first key sentence extracted from the key document into natural language, classifies the type of query according to the semantic relationship of the key term within the key document used to generate the first key sentence, and reconstructs the query; and a computation module that searches for documents of the same type as the type of the reconstructed query from the plurality of documents, extracts a second key sentence containing correlation information between a harmful substance and a specific disease from the searched documents of the same type as the reconstructed query, and predicts the likelihood of disease occurrence and health impact of the population group based on the correlation coefficient between the harmful substance and the specific disease. and may include an output module that outputs the results of the correlation evaluation between harmful substances and specific diseases.

Description

Technology Field

[0001] The present invention relates to a system for predicting the likelihood of disease occurrence and health effects in a population group, and more specifically, to a system for predicting the likelihood of disease occurrence and health effects in a population group that outputs predicted results based on an evaluation of the correlation between harmful substances and specific diseases. Background Technology

[0002] The applicant has filed patent applications for 'System and method for predicting health effects based on environmental hazardous factors (Application No.: 10-2025-0103303)', 'Interactive response system and method for evaluating correlation between hazardous substances and specific diseases (Application No.: 10-2025-0103304)', and 'System for providing integrated information on agricultural by-products and method of operation thereof (Application No.: 10-2025-0109623)' to evaluate the correlation between environmental hazardous factors and specific diseases.

[0003] Environmental diseases are reported to be more vulnerable in children, adolescents, and the elderly, and can be caused or exacerbated by various exposure routes (e.g., inhalation, ingestion, skin exposure, etc.) to everyday environmental hazards such as fine dust and chemicals.

[0004] Identifying the scientific correlation between exposure to environmental hazards and environmental diseases such as allergies and circulatory diseases is an essential element for protecting public health and establishing environmental health policies.

[0005] As such, Open Science (OS) is gaining prominence in research on environmental disease-exposure correlations, which require the identification of scientific correlations, due to the ethical and cost limitations of clinical research. In particular, there is an urgent need to establish an OS ecosystem in the field of environmental health research, where collection, synthesis, and utilization are essential as factual science.

[0006] However, due to causes such as a lack of public data for environmental health research, a lack of systematization of models for defining environmental diseases and evaluating exposure correlations, and insufficient technology for processing big data of vast environmental health research information, the OS infrastructure within the research field in Korea for identifying environmental disease-exposure correlations is very poor.

[0007] Meanwhile, for the study of the correlation between environmental diseases and exposure, a systematic review (SR) is essential, which is a technique that comprehensively reviews all related studies rather than individual study results to arrive at a final conclusion.

[0008] However, systematic reviews (SR) have the problem of being a complex and excessively time-consuming approach because they require rigorous procedures, such as literature search and extraction by multiple researchers and evaluation of literature bias (reliability), from the research question to the drawing of conclusions.

[0009] Furthermore, there are issues such as the difficulty of automated literature analysis for natural language-based queries regarding the correlation between environmental diseases and exposure, the lack of structures for structuring environmental and health data, and the fact that responses using AI models are limited to simple outputs without explanations regarding reliability or sources, resulting in insufficient explainable interpretation (eXplainable AI, XAI) and user interaction in terms of policy and medical application. Prior art literature

[0010] Republic of Korea Published Patent Application No. 10-2020-0080570 (Published July 7, 2020) Republic of Korea Registered Patent Application No. 10-2702075 (Registered August 29, 2024) Republic of Korea Application No. 10-2025-0103303 (Filed July 29, 2025) Republic of Korea Application No. 10-2025-0103304 (Filed July 29, 2025) Republic of Korea Application No. 10-2025-0109623 (Filed August 8, 2025) The problem to be solved

[0011] Accordingly, the present invention has been devised to solve the aforementioned problems, and the objective of the present invention is to provide a system for predicting the likelihood of disease occurrence and health effects in a population based on the correlation between harmful substances and specific diseases, while reducing the time and labor costs for systematic literature review (SR) research.

[0012] In addition, the objective of the present invention is to provide a system for predicting the likelihood of disease occurrence and health effects in a population, which visualizes the results of the correlation assessment between harmful substances and specific diseases so that a user can visually check them when the likelihood of disease occurrence and health effects are predicted.

[0013] Furthermore, the objective of the present invention is to provide a system for predicting the likelihood of disease occurrence and health impacts in a population that enables highly reliable interaction with a user by responding to a query requesting an evaluation of the correlation between a harmful substance entered by the user and a specific disease by presenting the source and inference path.

[0014] Furthermore, the objective of the present invention is to provide a system for predicting the likelihood of disease occurrence and health impact in a population, which adds synonyms and sub-conceptual languages ​​for each disease name, chemical name, and exposure condition included in the query according to the type of natural language-based query input by a user, and then reconstructs the query into a search expression structure for searching for literature of the same type as the query, thereby ensuring that literature of the same type as the query is searched from the natural language-based query input by the user with minimal omission.

[0015] Furthermore, the objective of the present invention is to provide a system for predicting the likelihood of disease occurrence and health impact in a population, which provides the user with the inference path used to derive the sources of literature retrieved through a reconstructed query and the prediction results of the likelihood of disease occurrence and health impact in a population as a response to a natural language-based query input by the user.

[0016] However, the technical problems to be solved by the present invention are not limited to those mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below. means of solving the problem

[0017] A system for predicting the likelihood of disease occurrence and health impact of a population according to an embodiment of the present invention for achieving the above-mentioned purpose comprises: a data collection module that collects and stores a plurality of documents containing unstructured raw data regarding the environment and health from the outside; an input module that receives a natural language-based query containing the key term from a user; a preprocessing module that preprocesses whitespace, punctuation, and noise in the form of stop words within the text of the query; an SR execution and query reconstruction module that extracts a key document among the plurality of documents collected by the data collection module and reconstructs a first key sentence extracted from the key document into natural language, classifies the type of query according to the semantic relationship of the key term within the key document used to generate the first key sentence, and reconstructs the query; and a computation module that searches for documents of the same type as the type of the reconstructed query from the plurality of documents, extracts a second key sentence containing correlation information between a harmful substance and a specific disease from the searched documents of the same type as the reconstructed query, and predicts the likelihood of disease occurrence and health impact of a population based on the correlation coefficient between the harmful substance and the specific disease. and may include an output module that outputs the results of the correlation evaluation between harmful substances and specific diseases.

[0018] In addition, the data collection module can refine the text regarding the environment and health of the raw data into natural language, structure it in the ontology normalization module into a format capable of extracting the key terms, extract the meta-information of the literature, and store it in a state where the text of the raw data and the meta-information are matched.

[0019] And the above meta-information may include information regarding the time of creation, region, and source of the above document.

[0020] In addition, a system for predicting the likelihood of disease occurrence and health effects of a population according to one embodiment of the present invention may include an ontology normalization module that extracts disease names, chemical names, and exposure conditions included in the raw data of the literature as key terms, and then connects the key terms to an ontology to structure the semantic relationships of the key terms.

[0021] And the ontology normalization module can extract the key term from the text of the raw data, connect meta information corresponding to the key term to the ID and URL of the ontology, structure the semantic relationship of the key term in the form of subject-predicate-object (triple), convert the structured result into a knowledge graph in the format of RDF (Resource Description Framework) or OWL (Web Ontology Language), and provide the knowledge graph so that semantic-based inference can be performed in the SR execution and query reconstruction module and the operation module.

[0022] In addition, the above-mentioned key terms can be linked to the ID and URL of the DOID (Disease Ontology ID), the ID and URL of the chemical substance name of the CHEBI (Chemical Entities of Biological Interest), and the ID and URL of the exposure condition of the ENVO (Environment Ontology), respectively.

[0023] And the above SR execution and query reconstruction module automatically generates literature search keywords to search for documents containing disease names, chemical names, and exposure conditions from the above preprocessed query, extracts key documents among multiple documents searched through the literature search keywords in which disease names, chemical names, and exposure conditions are listed in the title or abstract, assigns a score for the reliability of the key documents to each key document based on the possibility of distortion of the key documents, removes key documents among the key documents in which the score is less than a preset score, and can extract the first key sentence from key documents in which the score is greater than or equal to the preset score.

[0024] In addition, the first core sentence mentioned above may be a natural language-based sentence containing correlation information between harmful substances and specific diseases within the core document.

[0025] And the above SR execution and query reconstruction module can generate a reconstructed query by adding synonyms and sub-conceptual languages ​​to the preprocessed query according to the type of the disease name, chemical name, and exposure condition, respectively, when the type of query is classified according to the semantic relationship of the key term in the key document used to generate the first key sentence, and by reconstructing the query with added synonyms and sub-conceptual languages ​​into a search expression structure for searching for documents of the same type as the query with added synonyms and sub-conceptual languages.

[0026] In addition, the above computation module can extract a second core sentence containing correlation information between harmful substances and specific diseases from literature of the same type as the type of the above reconstruction query.

[0027] And the above second key sentence may be a sentence in which correlation information between the above harmful substance and a specific disease is reconstructed in natural language.

[0028] In addition, the output module can output the results of the correlation assessment between the hazardous substance and a specific disease in at least one form, such as a GIS-based dashboard and report, or numerical information that quantifies the results of the correlation assessment between the hazardous substance and a specific disease.

[0029] And the results of the correlation assessment between the above hazardous substance and a specific disease may include regional disease distribution, concentration of environmental hazardous factors, disease trends, trends in the concentration of environmental hazardous factors, and results of comparing the probability of disease occurrence between specific points in time or specific regions, calculated based on the probability of disease occurrence and health effects of the population predicted by the above computation module.

[0030] Additionally, the output module may be an interactive response module that provides the user, as an answer to a query entered into the input module, the source of the literature retrieved from the computation module and the inference path used to derive the prediction results of the probability of disease occurrence and health effects of the population according to the XAI (eXplainable AI) method.

[0031] delete Effects of the invention

[0032] The present invention can automatically perform the workflow of a systematic review (SR) to predict the likelihood of disease occurrence and health effects in a population based on correlation coefficients between harmful substances and specific diseases, thereby reducing the time and labor costs required for the systematic review (SR) study.

[0033] In addition, since the present invention can predict the likelihood of disease occurrence and health effects in a population, it can contribute to the establishment of OS infrastructure within research sites for identifying environmental disease-exposure correlations in Korea.

[0034] The present invention may provide a user with at least one of a GIS (Geographic Information System)-based dashboard and report for displaying regional disease distribution calculated based on the probability of disease occurrence and health effects of a predicted population, concentration of environmental hazardous factors, disease trends, trends in the concentration of environmental hazardous factors, and results of correlation evaluation between hazardous substances and specific diseases, which are results of comparing the probability of disease occurrence between specific points in time or specific regions.

[0035] Furthermore, the present invention enables highly reliable interaction with the user by responding to a query requesting an evaluation of the correlation between harmful substances entered by the user and specific diseases by presenting the source and inference path.

[0036] Furthermore, the present invention reconstructs the query into a search expression structure for searching for literature of the same type as the query by adding synonyms and sub-conceptual languages ​​for each of the disease name, chemical name, and exposure condition included in the query according to the type of natural language-based query entered by the user, thereby enabling the retrieval of literature of the same type from the natural language-based query entered by the user with minimal omission.

[0037] In addition, the present invention can provide the user with an inference path used to derive the source of the literature retrieved through the reconstructed query and the prediction results of the likelihood of disease occurrence and health effects of the population as a response to a natural language-based query input by the user.

[0038] Meanwhile, the effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description below. Brief explanation of the drawing

[0039] FIG. 1 is a block diagram illustrating the components of a system for predicting the likelihood of disease occurrence and health effects in a population group according to one embodiment of the present invention. FIG. 2 is a flowchart of a method for predicting the likelihood of disease occurrence and health effects in a population group according to one embodiment of the present invention. FIG. 3 is a flowchart illustrating the detailed process of raw data collection and preprocessing steps according to one embodiment of the present invention. FIG. 4 is a flowchart illustrating the detailed process of an ontology normalization step according to one embodiment of the present invention. FIG. 5 is a flowchart illustrating the detailed process of the query input and preprocessing steps according to one embodiment of the present invention. FIG. 6 is a flowchart illustrating the detailed process of the step of automatically performing a systematic review and reconstructing a query according to one embodiment of the present invention. FIG. 7 is a flowchart illustrating the detailed process of the step of calculating the correlation coefficient between a harmful substance and a specific disease according to one embodiment of the present invention. FIG. 8 is a flowchart illustrating the detailed process of the output step of the correlation evaluation result between a harmful substance and a specific disease according to one embodiment of the present invention. FIG. 9 is a drawing illustrating a GIS-based dashboard according to an embodiment of the present invention. FIG. 10 is a drawing illustrating a GIS-based report according to one embodiment of the present invention. FIG. 11 is a drawing for explaining numerical information according to one embodiment of the present invention. FIG. 12 is a drawing for explaining a reconstruction query according to an embodiment of the present invention. Specific details for implementing the invention

[0040] Hereinafter, embodiments of the present invention are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present invention. However, since the description of the present invention is merely an example for structural or functional explanation, the scope of the present invention should not be interpreted as being limited by the embodiments described in the text. That is, since the embodiments are subject to various modifications and may take various forms, the scope of the present invention should be understood to include equivalents capable of realizing the technical concept. Furthermore, the objectives or effects presented in the present invention do not imply that a specific embodiment must include all of them or only such effects; therefore, the scope of the present invention should not be understood as being limited by them.

[0041] The meaning of the terms described in this invention should be understood as follows.

[0042] Terms such as "first" and "second" are intended to distinguish one component from another, and the scope of rights shall not be limited by these terms. For example, the first component may be named the second component, and similarly, the second component may be named the first component. When a component is referred to as being "connected" to another component, it should be understood that it may be directly connected to that other component, or that there may be other components in between. Conversely, when a component is referred to as being "directly connected" to another component, it should be understood that there are no other components in between. Meanwhile, other expressions describing the relationship between components, such as "between" and "exactly between," or "adjacent to" and "directly adjacent to," shall be interpreted in the same manner.

[0043] A singular expression should be understood to include a plural expression unless the context clearly indicates otherwise, and terms such as "include" or "have" are intended to specify the existence of the set-up features, numbers, steps, actions, components, parts, or combinations thereof, and should be understood not to preclude the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0044] Unless otherwise defined, all terms used herein have the same meaning as generally understood by those skilled in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted as having meanings consistent with the context of the relevant technology and should not be interpreted as having an ideal or overly formal meaning unless explicitly defined in this invention.

[0046] System and method for predicting disease occurrence probability and health impact in a population

[0047] Hereinafter, a system (100) for predicting the likelihood of disease occurrence and health effects of a population group according to one embodiment of the present invention will be described in detail with reference to the attached drawings.

[0048] FIG. 1 is a block diagram illustrating the components of a system for predicting the likelihood of disease occurrence and health effects in a population group according to one embodiment of the present invention.

[0049] Referring to FIG. 1, a population disease occurrence probability and health impact prediction system (100) according to one embodiment of the present invention includes a data collection module (110), an ontology normalization module (120), an input module (130), a preprocessing module (140), an SR execution and query reconstruction module (150), a computation module (160), and an output module (170).

[0050] In one embodiment, the data collection module (110) can collect and store a plurality of documents containing unstructured raw data about the environment and health from the outside.

[0051] In one embodiment, the external may be a public institution, such as the Ministry of Environment or the Korea Disease Control and Prevention Agency, capable of collecting unstructured raw data regarding the environment and health.

[0052] In one embodiment, the unstructured raw data regarding the environment and health includes at least unstructured text regarding the environment and health, and may additionally include graphs, tables, etc. regarding the environment and health.

[0053] In one embodiment, the literature includes reports, documents, news, papers, etc. containing unstructured raw data on the environment and health that can be collected from public institutions.

[0054] In one embodiment, the ontology normalization module (120) can extract key terms included in the raw data of the literature collected from the data collection module (110), and then connect the key terms to an ontology to structure the semantic relationships of the key terms.

[0055] In one embodiment, the key terms include disease names, chemical names, and exposure conditions within the unstructured raw data.

[0056] In one embodiment, the disease name refers to the official diagnostic name of a health problem, such as acute leukemia, cancer, or lung cancer, which can predict or analyze incidence, prevalence, risk, etc., at the population level.

[0057] In one embodiment, the chemical name refers to a name designating a chemical factor among environmental hazardous factors that may exist in or be exposed to a population or environment, such as benzene, PM2.5 (ultrafine dust), lead, and nitrogen dioxide (NO2).

[0058] In one embodiment, the exposure conditions may be context information for defining whether an environmental hazardous agent has affected the human body under regional, age, and gender conditions.

[0059] In one embodiment, the ontology includes a DOID (Disease Ontology ID) to which a disease name is linked, a CHEBI (Chemical Entities of Biological Interest) to which a chemical name is linked, and an ENVO (Environment Ontology) to which an exposure condition is linked, and each ontology can be linked to each other.

[0060] In addition, disease names, chemical names, and exposure conditions can be interconnected through ontologies to be constructed as relational data (Linked Open Data, LOD) and utilized as an external knowledge base.

[0061] In one embodiment, the input module (130) may receive a natural language-based query (e.g., "Does exposure to PFOA affect the developmental growth of a fetus?") containing a key term for requesting an evaluation of the correlation between a harmful substance and a specific disease through an input means (not shown) such as a terminal or keyboard not shown in the drawing.

[0062] In one embodiment, the preprocessing module (140) can remove noise by preprocessing the noise within the text of the query input to the input module (130).

[0063] In this case, noise may include spaces, punctuation, and stop words within the text of the query.

[0064] In one embodiment, the SR execution and query reconstruction module (150) may be a module for automatically performing the workflow of a Systematic Review (SR) while reducing the time and labor costs for the study of a Systematic Review (SR).

[0065] Additionally, the SR execution and query reconstruction module (150) can automatically generate literature search keywords to search for documents containing disease names, chemical names, and exposure conditions from a query preprocessed through the preprocessing module (140) among multiple documents collected from the data collection module (110), and can extract key literature among multiple literatures searched through the literature search keywords in which disease names, chemical names, and exposure conditions are listed in the title or abstract, and can automatically perform the workflow of a systematic review (SR) by extracting a first key sentence from the key literature based on a score for the reliability of the key literature and then reconstructing the first key sentence into natural language.

[0066] Here, the first core sentence may be a natural language-based sentence containing correlation information between hazardous substances and specific diseases within the core literature.

[0067] And the SR execution and query reconstruction module (150) can generate a reconstruction query by classifying the type of query (question purpose) according to the semantic relationship of the key term in the key document used to generate the first key sentence reconstructed in natural language, adding synonyms and sub-conceptual languages ​​to the preprocessed query according to the classified query type, and reconstructing the query with added synonyms and sub-conceptual languages ​​into a search expression structure for searching for documents of the same type as the query with added synonyms and sub-conceptual languages.

[0068] In this case, the types of inquiries may include inquiries regarding the correlation between hazardous substances and specific diseases (impact), inquiries regarding the risk of environmental hazardous factors (prediction), and inquiries regarding whether environmental hazardous factors cause specific diseases (causation).

[0069] In addition, the synonyms and sub-conceptual languages ​​for each disease name, chemical name, and exposure condition refer to synonyms, alternatives, and concept hierarchy subclasses defined within the ontology connected through the ontology normalization module (120), and by being added to the query, it enables search expansion and semantic-based precision querying through the query.

[0070] In one embodiment, the operation module (160) can search for documents of the same type as the type of reconstruction query from a plurality of documents collected from the data collection module (110).

[0071] Additionally, the computation module (160) can extract a second core sentence containing correlation information between a harmful substance and a specific disease from literature of the same type as the type of reconstructed query retrieved, and by calculating the correlation coefficient between the harmful substance and the specific disease from the second core sentence, it can predict the likelihood of disease occurrence and health effects of a population based on the correlation coefficient between the harmful substance and the specific disease.

[0072] Here, the second key sentence may be a sentence in which correlation information between harmful substances and specific diseases is reconstructed in natural language.

[0073] And the computation module (160) can conduct an evaluation of the correlation between the harmful substance and the specific disease based on the predicted probability of disease occurrence and health effects of the population and derive the results of the correlation evaluation between the harmful substance and the specific disease.

[0074] In one embodiment, the results of the correlation assessment between hazardous substances and specific diseases include regional disease distribution calculated based on the likelihood of disease occurrence and health effects of a population, the concentration of environmental hazardous factors, disease trends, trends in the concentration of environmental hazardous factors, and results of comparing the likelihood of disease occurrence between specific points in time or specific regions.

[0075] In one embodiment, the regional disease distribution refers to the spatial distribution of the likelihood of disease occurrence or the level of risk of disease occurrence in each region (e.g., at the city, county, or district level).

[0076] In one embodiment, the concentration of an environmental hazardous factor is the predicted concentration value of the environmental hazardous factor (e.g., PM2.5, benzene, etc.) in each region at a specific time.

[0077] In one embodiment, the disease trend is the result of predicting the change in the probability of disease occurrence over time (e.g., day, month, year).

[0078] In one embodiment, the trend in the concentration of environmental hazardous factors is the result of predicting the change in the concentration of environmental hazardous factors over time.

[0079] In one embodiment, the result of comparing the probability of disease occurrence between a specific point in time or a specific region is the result of relatively comparing the probability of disease occurrence between a first region and a second region or between a first point in time or a second point in time.

[0080] In one embodiment, the output module (170) can output the results of the correlation assessment between harmful substances and specific diseases in at least one form among the GIS-based dashboard shown in FIG. 9, the GIS-based report shown in FIG. 10, and numerical information quantified as shown in FIG. 11.

[0081] In one embodiment, the GIS-based dashboard is a web-based interactive UI, and the information regarding the prediction results of the likelihood of disease occurrence and health impact of the visualized population may change depending on user operation.

[0082] In one embodiment, the GIS-based report can output information included in the prediction results of disease occurrence probability and health impact of a population all at once as a summary in a document format (e.g., PDF, DOC, etc.).

[0083] Additionally, the output module (170) may output a portion of the GIS-based dashboard and a portion of the report together and provide them to the user.

[0084] As a specific example, a GIS-based dashboard can output different information on the prediction results of the likelihood of disease occurrence and health impact of a population depending on user operation, and a GIS-based report can output summary information in the form of a report that analyzes the information on the prediction results of the likelihood of disease occurrence and health impact of a population displayed on the GIS-based dashboard.

[0085] Through this output module (170), the user can visually check the predicted results of the likelihood of disease occurrence and health impact of the population through at least one of the GIS-based dashboards and reports displayed on the screen of the output module (170).

[0086] The output module (170) may be an interactive response module that provides at least one of a GIS-based dashboard, report, and numerical information to the user as a response to a query entered into the input module (130).

[0087] As a specific example, the output module (170) can be implemented as a GPT-based chatbot capable of interacting with the user by providing at least one of a GIS-based dashboard and report and numerical information to the user as an answer to a query entered into the input module (130).

[0088] In addition, the output module (170) can provide the user with the inference path used to derive the source of the literature retrieved from the computation module (160) and the prediction results of the likelihood of disease occurrence and health effects of the population according to the XAI (eXplainable AI) method as an answer to the query entered into the input module (130), thereby enabling reliable interaction with the user.

[0089] Hereinafter, with reference to FIGS. 2 to FIGS. 12, a method (S100) for predicting the likelihood of disease occurrence and health effects of a population group according to one embodiment of the present invention, which is performed by the system (100) for predicting the likelihood of disease occurrence and health effects of a population group according to the present invention, will be described in detail.

[0090] FIG. 2 is a flowchart of a method for predicting the probability of disease occurrence and health impact of a population group according to an embodiment of the present invention; FIG. 3 is a flowchart illustrating the detailed process of the raw data collection and preprocessing step according to an embodiment of the present invention; FIG. 4 is a flowchart illustrating the detailed process of the ontology normalization step according to an embodiment of the present invention; FIG. 5 is a flowchart illustrating the detailed process of the query input and preprocessing step according to an embodiment of the present invention; FIG. 6 is a flowchart illustrating the detailed process of the systematic review automatic execution and query reconstruction step according to an embodiment of the present invention; FIG. 7 is a flowchart illustrating the detailed process of the correlation coefficient calculation step between a harmful substance and a specific disease according to an embodiment of the present invention; FIG. 8 is a flowchart illustrating the detailed process of the correlation evaluation result output step between a harmful substance and a specific disease according to an embodiment of the present invention; FIG. 9 is a diagram illustrating a GIS-based dashboard according to an embodiment of the present invention; FIG. 10 is a diagram illustrating a GIS-based report according to an embodiment of the present invention; and FIG. 11 is a diagram for explaining numerical information according to an embodiment of the present invention. Figure 12 is a drawing for explaining a reconstruction query according to an embodiment of the present invention.

[0091] Referring to FIG. 2, a method (S100) for predicting the likelihood of disease occurrence and health effects of a population according to one embodiment of the present invention may proceed in the order of raw data collection and preprocessing step (S110), ontology normalization step (S120), query input and preprocessing step (S130), automatic systematic review and query reconstruction step (S140), correlation coefficient calculation step (S150), and correlation evaluation result output step (S160).

[0092] In one embodiment, the raw data collection and preprocessing step (S110) may proceed in the order of the detailed process shown in FIG. 3.

[0093] Referring to FIG. 3, the data collection module (110) can collect unstructured raw data about the environment and health from the outside (S111).

[0094] After that, the data collection module (110) can refine the text about the environment and health of the raw data into natural language (S112).

[0095] In the above step (S112), refining the text about environment and health into natural language is intended to process the text about environment and health into a structured form so that the ontology normalization module (120) can accurately extract key terms and perform the connection between the key terms and the ontology and RDF / OWL structuring.

[0096] After the above step (S112), the data collection module (110) can structure the text of the raw data refined into natural language into a format in which key terms can be extracted by the ontology normalization module (120) (S113).

[0097] In the above step (S113), format conversion of the text of raw data refined into natural language is intended to provide a structured input that can be processed by the ontology normalization module (120), the SR execution and query reconstruction module (150), and the operation module (160), and convertible formats may include JSON, XML, RDF, OWL, TSV, etc.

[0098] After the above step (S113), the data collection module (110) can extract the meta-information of the literature and store it in a database (not shown) provided in the health impact prediction system (100) in a state where the text of the raw data and the meta-information are matched (S114).

[0099] In the above step (S114), storing the text and meta-information of the raw data together in a database (not shown) is intended to resolve semantic redundancy and ambiguity when linking (or mapping) key terms to an ontology.

[0100] In one embodiment, meta-information includes information about the time, region, and source of a document containing raw data, and can be linked to an ontology along with key terms.

[0101] Meanwhile, the ontology normalization step (S120) can proceed in the order of the detailed process shown in Fig. 4.

[0102] Referring to FIG. 4, the ontology normalization module (120) can extract key terms from the text of raw data stored in a database (not shown) (S121).

[0103] After that, the ontology normalization module (120) can connect the core term and the meta information corresponding to the core term to the ID and URL of the ontology (S122).

[0104] In the above step (S122), among the key terms, the disease name can be linked to the ID and URL of DOID (Disease Ontology ID), among the key terms, the chemical name can be linked to the ID and URL of CHEBI (Chemical Entities of Biological Interest), and among the key terms, the exposure condition can be linked to the ID and URL of ENVO (Environment Ontology).

[0105] In the above step (S122), the core terms and meta information can be interconnected through connection with the ontology ID and URL to be constructed as relational data (Linked Open Data, LOD) and can be utilized as an external knowledge base.

[0106] After the above step (S122), the ontology normalization module (120) can structure the semantic relationship of the core terms connected to the ID and URL of the ontology in the form of subject-predicate-object (triple) (S123).

[0107] In the above step (S123), structuring the semantic relationship of the core term in the form of subject-predicate-object (triple) is intended to enable inference in the SR execution and query reconstruction module (150) and the computation module (160) by expressing the semantic relationship between the disease name, chemical name, and exposure condition of the core term.

[0108] After the above step (S123), the ontology normalization module (120) can convert the structured core terms into a knowledge graph in an RDF (Resource Description Framework) or OWL (Web Ontology Language) format that can be inferred by the SR execution and query reconstruction module (150) and the operation module (160) (S124).

[0109] In the above step (S124), converting the key terms into a knowledge graph in RDF or OWL format is intended to allow the SR execution and query reconstruction module (150) and the operation module (160) to construct a relationship network that can be inferred.

[0110] In the above step (S124), converting the core term into a knowledge graph in RDF or OWL format means converting the core term into a SPARQL queryable form that fits the structure of relational data (LOD), thereby enabling the SR execution and query reconstruction module (150) to generate literature search keywords by inferring semantic relationships from the knowledge graph when a research question is input from a user.

[0111] After the above step (S124), the ontology normalization module (120) can provide a knowledge graph in RDF or OWL format to the SR execution and query reconstruction module (150) and the computation module (160) so that semantic-based inference can be performed in the SR execution and query reconstruction module (150) and the computation module (160) (S125).

[0112] In one embodiment, the query input and preprocessing step (S130) may proceed in the order of the detailed process shown in FIG. 5.

[0113] Referring to FIG. 5, the input module (130) can receive a natural language-based query from the user (S131).

[0114] After the above step (S131), the preprocessing module (140) can preprocess noise within the text of the natural language-based query to remove the noise (S132).

[0115] In the above step (S132), noise may include spaces, punctuation, and stop words within the text of the query.

[0116] In one embodiment, the step of automatically performing a systematic review and reconstructing a query (S140) may proceed in the order of the detailed process shown in FIG. 6.

[0117] Referring to FIG. 6, the SR execution and query reconstruction module (150) can automatically generate literature search keywords to search for documents containing disease names, chemical names, and exposure conditions from a query preprocessed through the preprocessing module (140) (S141).

[0118] In the above step (S141), it is desirable that the preprocessed query includes disease names, chemical names, and exposure conditions in natural language.

[0119] In the above step (S141), the SR execution and query reconstruction module (150) can infer semantic relationships between disease names, chemical names, and exposure conditions through a knowledge graph in RDF or OWL format, and generate literature search keywords to search for literature related to the query based on this.

[0120] After the above step (S141), the SR execution and query reconstruction module (150) can extract key literature among the multiple literatures collected by the data collection module (110) that are searched through literature search keywords, in which the disease name, chemical name, and exposure condition are listed in the title or abstract (S142).

[0121] In the above step (S142), the SR execution and query reconstruction module (150) can infer semantic relationships between disease names, chemical names, and exposure conditions through a knowledge graph in RDF or OWL format, and based on this, extract key literature among multiple retrieved literature in which disease names, chemical names, and exposure conditions are listed in the title or abstract.

[0122] After the above step (S142), the SR execution and query reconstruction module (150) can assign a score for the reliability of each key document based on the possibility of distortion (bias) of the key document (S143).

[0123] In the above step (S143), the SR execution and query reconstruction module (150) can determine the possibility of distortion by analyzing the text of the core literature based on techniques such as randomized, control group, and drop-out, and then calculate a score for the reliability of the core literature based on the determined possibility of distortion and assign it to the core literature.

[0124] At this time, if the score of a key document is less than a preset score (S144-NO), the SR execution and query reconstruction module (150) can remove the key document assigned a score less than the preset score (S145).

[0125] In contrast, if the score of the core document is greater than or equal to a preset value (S144-YES), the SR execution and query reconstruction module (150) can reconstruct the first core sentence into natural language after extracting the first core sentence from the core document with a score greater than or equal to a preset value (S146).

[0126] After the above step (S146), the SR execution and query reconstruction module (150) can classify the type of query (query purpose) according to the semantic relationship of the key terms in the key document used to generate the first key sentence reconstructed in natural language (S147).

[0127] In the above step (S147), the types of queries may include, as described above, a query regarding the correlation between hazardous substances and specific diseases (impact), a query regarding the risk of environmental hazardous factors (prediction), and a query regarding whether environmental hazardous factors cause specific diseases (causation).

[0128] After the above step (S147), the SR execution and query reconstruction module (150) can add synonyms and sub-concept languages ​​for each of the disease name, chemical name, and exposure condition to the preprocessed query according to the type of classified query (S148).

[0129] In the above step (S148), the synonyms and sub-conceptual languages ​​for each of the disease name, chemical name, and exposure condition refer to synonyms, alternatives, and concept hierarchy subclasses defined within the ontology connected through the ontology normalization module (120), and by being added to the query, it enables search expansion and semantic-based precision querying through the query.

[0130] After the above step (S148), the SR execution and query reconstruction module (150) can generate a reconstructed query by reconstructing the query into a search expression structure for searching for documents of the same type as the query with added synonyms and sub-conceptual languages ​​among a plurality of documents (S149).

[0131] In the above step (S149), the SR execution and query reconstruction module (150) may be equipped with an AI model that has been trained, verified, and evaluated to generate a reconstruction query.

[0132] In the above step (S149), the AI ​​model may be a fine-tuned GPT or BERT family model that classifies the type of query and reconstructs the query to generate a reconstructed query.

[0133] In one embodiment, the reconstructed query can be compared with the search results before the inclusion of synonyms and sub-conceptual language (e.g., 8 documents) as illustrated in FIG. 12, so that more documents (e.g., 1,722 documents) are retrieved by reflecting the search expansion.

[0134] Meanwhile, the correlation coefficient calculation step (S150) can proceed in the order of the detailed process shown in FIG. 7.

[0135] Referring to FIG. 7, the operation module (160) classifies the type of the reconstructed query according to the semantic relationship of the key terms included in the reconstructed query generated from the execution and query reconstruction module (150), and can search for documents of the same type as the type of the reconstructed query (S151).

[0136] After the above step (S151), the computation module (160) can extract a second key sentence containing correlation information between harmful substances and specific diseases from literature of the same type as the type of reconstructed query found (S152).

[0137] After the above step (S152), the computation module (160) can calculate the correlation coefficient between the harmful substance and the specific disease from the second key sentence, and predict the likelihood of disease occurrence and health effects of the population based on the correlation coefficient between the harmful substance and the specific disease (S153).

[0138] After the above step (S153), the computation module (160) can perform an evaluation of the correlation between the harmful substance and a specific disease based on the probability of disease occurrence and health effects of the predicted population and derive the results of the correlation evaluation between the harmful substance and the specific disease.

[0139] Meanwhile, the correlation evaluation result output step (S160) can proceed in the order of the detailed process shown in FIG. 8.

[0140] Referring to FIG. 8, the output module (170) may provide the user with the results of an evaluation of the correlation between harmful substances and specific diseases in at least one form among the GIS-based dashboards, reports, and numerical information shown in FIG. 9 to 11 as a response to a natural language-based query entered by the user into the input module (130), or provide the user with the source of the literature retrieved from the computation module (160) and the inference path used to derive the results of the prediction of the likelihood of disease occurrence and health effects of the population as a response to the query entered into the input module (130) (S161).

[0141] After the above step (S161), if the user inputs a follow-up query to evaluate the correlation between harmful substances and specific diseases through the input module (130) (S162-YES), the output module (170), which is a conversational response module implemented as a GPT-based chatbot, can interact with the user by generating a natural language-based response (S163).

[0143] Effects according to the present invention

[0144] The population disease occurrence probability and health impact prediction system (100) of the present invention can automatically perform a workflow of a systematic review (SR) to predict the disease occurrence probability and health impact of a population based on the correlation coefficient between a harmful substance and a specific disease, thereby reducing the time and labor costs for the systematic review (SR) study.

[0145] In addition, the population disease occurrence probability and health impact prediction system (100) of the present invention can predict the disease occurrence probability and health impact of a population, so it can contribute to the establishment of OS infrastructure within research sites for identifying environmental disease-exposure correlations in Korea.

[0146] And the population disease occurrence probability and health impact prediction system (100) of the present invention may provide the user with at least one of a GIS (Geographic Information System)-based dashboard and report for displaying regional disease distribution, concentration of environmental hazardous factors, disease trends, trends in concentration of environmental hazardous factors, and results of correlation evaluation between hazardous substances and specific diseases, which are results of comparing the disease occurrence probability between specific points in time or specific regions, based on the predicted disease occurrence probability and health impact of the population.

[0147] In addition, the population disease occurrence probability and health impact prediction system (100) of the present invention can have highly reliable interaction with the user by responding to a query requesting an evaluation of the correlation between a harmful substance entered by the user and a specific disease by presenting the source and inference path.

[0148] In addition, the population disease occurrence probability and health impact prediction system (100) of the present invention can search for literature with the same type as the query from the natural language-based query entered by the user by adding synonyms and sub-concept languages ​​for each disease name, chemical name, and exposure condition included in the query according to the type of natural language-based query entered by the user, and then reconstructing the query into a search expression structure for searching for literature with the same type as the query, thereby enabling literature with the same type as the query to be searched without omission as much as possible from the natural language-based query entered by the user.

[0149] In addition, the population disease occurrence probability and health impact prediction system (100) of the present invention may provide the user with the source of the literature retrieved through the reconstructed query and the inference path used to derive the prediction results of the population disease occurrence probability and health impact as a response to a natural language-based query input by the user.

[0151] As described above, the detailed description of the preferred embodiments of the present invention disclosed is provided to enable those skilled in the art to implement and practice the present invention. Although the present invention has been described with reference to preferred embodiments, those skilled in the art will understand that various modifications and changes can be made to the present invention without departing from the scope of the invention. For example, those skilled in the art may utilize each configuration described in the embodiments described above in combination with one another. Accordingly, the present invention is not intended to be limited to the embodiments shown herein, but to be given the broadest scope consistent with the principles and novel features disclosed herein.

[0152] The present invention may be embodied in other specific forms without departing from the technical spirit and essential features of the invention. Accordingly, the above detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention shall be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention. The invention is not intended to be limited to the embodiments shown herein, but to be given the broadest possible scope consistent with the principles and novel features disclosed herein. Furthermore, embodiments may be constructed by combining claims that are not explicitly related in the claims, or included as new claims through amendments made after filing. Explanation of the symbols

[0153] 100: Population Disease Probability and Health Impact Prediction System, 110: Data collection module, 120: Ontology Normalization Module, 130: Input module, 140: Preprocessing module, 150: SR Execution and Query Reconstruction Module, 160: Computation module, 170: Output module.

Claims

Claim 1 A data collection module that collects and stores multiple documents containing unstructured raw data regarding the environment and health from the outside; an input module that receives a natural language-based query containing key terms from a user; a preprocessing module that preprocesses whitespace, punctuation, and noise (stop words) within the text of the query; an SR execution and query reconstruction module that extracts key documents from the multiple documents collected by the data collection module, reconstructs a first key sentence extracted from the key documents into natural language, classifies the type of query according to the semantic relationship of key terms within the key documents used to generate the first key sentence, and reconstructs the query; a computation module that searches for documents of the same type as the type of the reconstructed query from the multiple documents, extracts a second key sentence containing correlation information between hazardous substances and specific diseases from the searched documents of the same type as the reconstructed query, and predicts the probability of disease occurrence and health impact of a population based on the correlation coefficient between hazardous substances and specific diseases; and an output module that outputs the results of the correlation evaluation between hazardous substances and specific diseases. and includes an ontology normalization module that extracts disease names, chemical names, and exposure conditions included in the raw data of the above literature as keywords, and then connects the keywords to an ontology to structure the semantic relationships of the keywords; wherein the ontology normalization module extracts the keywords from the text of the above raw data, connects meta-information corresponding to the keywords to the ID and URL of the ontology, structures the semantic relationships of the keywords in the form of subject-predicate-object (triple), converts the structured result into a knowledge graph in RDF (Resource Description Framework) or OWL (Web Ontology Language) format, and provides the knowledge graph to enable semantic-based inference in the SR execution and query reconstruction module and the computation module, and wherein the SR execution and query reconstruction moduleA literature search keyword is automatically generated from the above-mentioned preprocessed query to search for documents containing disease names, chemical names, and exposure conditions; key documents are extracted from multiple documents searched via the literature search keyword where disease names, chemical names, and exposure conditions are listed in the title or abstract; a reliability score is assigned to each key document based on the potential for distortion of the key documents; key documents among the key documents with a score below a preset score are removed, while the first key sentence is extracted from key documents with a score equal to or greater than the preset score; the text of the key documents is analyzed based on at least one of randomized, control group, and drop-out techniques to determine the potential for distortion, and a reliability score for the key documents is calculated based on the determined potential for distortion; the first key sentence is a natural language-based sentence containing correlation information between hazardous substances and specific diseases within the key documents; and the SR execution and query reconstruction module, when the query type is classified according to the semantic relationship of key terms within the key documents used to generate the first key sentence, [describes] disease names, chemical names, and exposure conditions according to the classified query type Each synonym and sub-conceptual language is added to the preprocessed query, and the query with the added synonym and sub-conceptual language is reconstructed into a search expression structure for searching for literature of the same type as the query with the added synonym and sub-conceptual language to generate a reconstructed query, and the operation module extracts a second core sentence containing correlation information between hazardous substances and specific diseases from literature of the same type as the type of the reconstructed query, and the second core sentence is a sentence in which the correlation information between hazardous substances and specific diseases is reconstructed in natural language, and the output module, according to the XAI (eXplainable AI) method,A system for predicting the likelihood of disease occurrence and health impact of a population, characterized in that it is a conversational response module that provides the user with the source of literature retrieved from the above computation module and the inference path used to derive the prediction results of the likelihood of disease occurrence and health impact of the above population as an answer to a query entered into the above input module, and is a GPT-based chatbot capable of interacting with the user; the above SR execution and query reconstruction module is equipped with an AI model that has been trained, verified, and evaluated to generate the above reconstruction query, and the AI ​​model is a GPT or BERT-family model that has been fine-tuned to classify the type of the above query and reconstruct the above query to generate the above reconstruction query; and the above computation module performs a correlation evaluation between environmental hazardous factors and specific diseases based on the predicted likelihood of disease occurrence and health impact of the population to derive the correlation evaluation results between environmental hazardous factors and specific diseases. Claim 2 A system for predicting the likelihood of disease occurrence and health impact of a population according to claim 1, wherein the data collection module refines the text regarding the environment and health of the raw data into natural language, structures it in the ontology normalization module into a format capable of extracting the key terms, extracts the meta-information of the literature, and stores it in a state where the text of the raw data and the meta-information are matched. Claim 3 A system for predicting the likelihood of disease occurrence and health effects in a population, characterized in that, in claim 2, the meta-information includes information regarding the time of creation, region, and source of the literature. Claim 4 delete Claim 5 delete Claim 6 A system for predicting the likelihood of disease occurrence and health effects in a population according to claim 1, wherein the key terms are respectively linked to the ID and URL of DOID (Disease Ontology ID), the ID and URL of chemical substance name of CHEBI (Chemical Entities of Biological Interest), and the ID and URL of exposure condition of ENVO (Environment Ontology). Claim 7 delete Claim 8 delete Claim 9 delete Claim 10 delete Claim 11 delete Claim 12 A system for predicting the likelihood of disease occurrence and health impact in a population according to claim 1, wherein the output module outputs the results of the correlation evaluation between the harmful substance and a specific disease in at least one form among a GIS-based dashboard and report, and numerical information that quantifies the results of the correlation evaluation between the harmful substance and a specific disease. Claim 13 A system for predicting the likelihood of disease occurrence and health impact of a population according to claim 12, wherein the correlation evaluation result between the hazardous substance and a specific disease includes regional disease distribution, concentration of environmental hazardous factors, disease trends, trends in the concentration of environmental hazardous factors, and results of comparing the likelihood of disease occurrence between a specific point in time or specific regions, calculated based on the likelihood of disease occurrence and health impact of the population predicted by the computation module. Claim 14 delete Claim 15 delete

Images