Information processing system

Abstract

To provide an information processing system for structuring non-structured text information so as to enable a change of content, etc., including preceding or following context, due to structuring and / or passage of time to be easily grasped.SOLUTION: In an information processing system 1, structuring means extracts a structured word applying a structuring model to non-structured text information, adds an information type to the structured word derived by classifying the type of information of the word, adds one of a plurality of labels classified by the role or functionality of words in a sentence composition, stores the structured word included in the non-structured text information in a table including a plurality of cells corresponding respectively to the plurality of labels, and stores the table in memory as one data unit. Display processing means displays the structured words for each data unit and displays a plurality of data units side by side so that same or common structured words are aligned, when a same or common structured word is included among the plurality of data units.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an information processing system for structuring unstructured text information, such as freely entered text information, and in particular to an information processing system for structuring text information relating to the same or a common event, which is freely entered multiple times over time. [Background technology]

[0002] In recent years, there have been attempts to accumulate electronic medical record information created by various medical institutions as big data, and to use it as a statistical information database for research and development in the medical field, or to use the details of treatments and examinations at other medical institutions as reference information. However, at present, the medical information that can be collected and used is originally structured and consists of items (disease name, medication, etc.) entered into electronic medical records, etc. The information is limited to information such as medication information, test information, etc., and the progression of symptoms and other information can be freely entered as text information. Unstructured information showing the condition is not in a state where it can be utilized. Disease names, medication information, test information, etc. are also used for insurance claims and are standardized, so the input contents are fixed in advance and are mainly selected from pull-down menus or from masters, and the information is structured and easy to utilize because the information collected from multiple electronic medical records is common. On the other hand, the progress of symptoms and condition are entered as free text information as the patient's chief complaint and the doctor's findings, and there are no rules for entering medical records into electronic medical records, such as differences in expression depending on the person entering them, unique abbreviations being used, and the input contents being undefined, and the information is not organized in a format that can be utilized.

[0003] In light of this situation, efforts are being made to utilize unstructured medical information. For example, Patent Document 1 discloses that, for text information freely entered in a free input field of an electronic medical record, a predetermined information type is detected and extracted, and target information corresponding to the information type is extracted using one or more natural language analysis processes, such as dependency analysis, context analysis, or machine learning using a learning model using a neural network, or a comparison dictionary, and the extracted target information is stored in a table format for each information type to form structured information.

[0004] Registry data has been attracting attention as one of the uses of medical information, especially in clinical development of medicines and medical devices. A registry is a database that collects medical and health information such as treatment details and treatment progress for specific diseases and health conditions, and there is a growing need for research and development that uses registry data as a secondary source, with the aim of reducing costs and shortening the time required for drug research and development. A known example of registry data is a patient registry, where the registration unit is a patient, and medical information such as specific diseases, disease groups, and treatments, such as what disease a patient has and what condition they are in, is collected. Patient registries are also called disease registries. Another use of medical information is the creation of medical-related documents (electronic medical records, medical certificates, referral letters, nursing care plans, instructions, etc.) in the work of medical institutions. These documents are created by doctors and medical institution staff based on information written in electronic medical records, etc. Currently, these tasks are created manually, which takes time, and is also a factor that puts pressure on medical staff's work hours.

[0005] Big data is also beginning to be used in various fields other than the medical field. Big data is data that can be used to derive knowledge that is useful for various businesses, and is a group of data from a variety of sources. Examples of big data include weather, temperature, and humidity data for each region, purchase history and entry history accumulated on online shopping sites and blog sites, multimedia data such as music and videos provided on web distribution sites, and social media data. Examples of data include social media data such as profiles and comments written by participants in social media; sensor data such as location, ride history, and temperature detected by GPS, IC cards, and RFID; and customer data such as direct mail data and membership card data managed in CRM (Customer Relationship Management) systems. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2020-086541 A Summary of the Invention [Problem to be solved by the invention]

[0007] In Patent Document 1, specific target information judged to be important is extracted and structured for each information type from freely entered text information in an electronic medical record entered at a certain consultation of a certain patient (for example, information types include "current illness history", "past medical history", "oral medication", "physical findings", and "post-hospital progress", and corresponding target information includes "symptoms" for the information type "current illness history", "disease name" as past medical history for the information type "past medical history", "drug name" for the information type "oral medication", and "disease name" as diagnosis for the information type "post-hospital progress"). For this reason, it is possible to simply search whether specific target information is included in the freely entered text information, but there is a risk that part or all of the information, such as the content and meaning of the freely entered text information and the relationship between the target information, may be lost, and there are cases where the content and progress of treatment cannot be understood. For example, even if a description is about a certain treatment method A, it is used in various contexts, such as whether the treatment method A was "performed", "performed in the past", "decided to be performed in the future", "considered to be performed", or "will be performed if certain conditions are met", so even if it is possible to recognize that the word "treatment method A" is written in the electronic medical record by structuring it, this alone is insufficient and uncertain as medical information, and difficult to use. In addition, Patent Document 1 also describes that positive and negative expressions and additional information for symptoms and disease names are stored in the remarks column in association with symptoms and disease names as target information, but the information stored in the remarks column is not assigned an information type and is not structured.

[0008] Furthermore, Patent Document 1 only discloses structuring text information at a certain consultation. In an actual electronic medical record, the contents of consultations and the history of treatment for a certain patient are accumulated in chronological order, and it is desirable to be able to grasp the contents of treatment and the course of treatment, which change over time even after structuring. For example, in diseases that require long-term treatment or follow-up, such as cancer, it is not uncommon for treatment to continue for several years, so the medical records become huge and complicated. For this reason, it was a very time-consuming task to select and organize the necessary information from the huge amount of information. In addition, as described above, since past implementations and future implementations are described, simply extracting keywords that seem important from a medical perspective does not necessarily allow the actual course of treatment to be understood, and it was difficult to correctly understand the time axis of each keyword and correctly structure it. When extracting information necessary for the registry or format that is ultimately required, it is also desirable to organize the structured information in chronological order for common or related events. However, Patent Document 1 does not give any special consideration to content that changes over time, and does not disclose a method for structuring unstructured information that transitions over time, taking into account time-series organization.

[0009] In addition, not only in electronic medical records but also in other big data, the same or common events may be entered multiple times periodically or irregularly in freely entered text information over time. For such text information, it is necessary to not only extract specific words and structure them by standardizing or standardizing them, but also to structure and analyze them by taking into account the context before and after. There is a need for a structure that makes it easy to understand the content and / or changes in content over time.

[0010] The present invention has been made in consideration of the above-mentioned problems, and one of its objectives is to provide an information processing system for structuring unstructured text information so that it is easy to grasp the structuring including the context and / or the changes in the content over time, etc. Another objective of the present invention is to provide an information processing system that structures unstructured text information entered into medical document information including electronic medical records (including documents such as medical certificates, referral letters, nursing care plans, instructions, etc., as well as unstructured text stored outside electronic medical records) and re-extracts it into a data format required for clinical research and clinical trials to make it easier to use it as registry data, and creates medical documents based on the structured information, making it easy to use important information when utilizing medical record information in another system or a text generation model using a neural network. [Means for solving the problem]

[0011] In order to solve the above problems, one of the information processing systems of the present invention comprises: An information processing system for structuring a plurality of unstructured text information in which descriptions of the same or a common phenomenon may exist, comprising: A structuring means and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to each of the structured words, which classifies the type of information of the words, and further assigning one of a plurality of labels classified according to the role or function of the words in the structure of a sentence; storing structured words included in the unstructured text information in a table including a plurality of cells corresponding to a plurality of labels, and storing the table as one data unit; The display processing means displays the structured words for each data unit, Furthermore, when identical or common structured words are contained among a plurality of data units, the plurality of data units are displayed in a line such that the identical or common structured words are aligned.

[0012] Furthermore, in the above information processing system, it is preferable that the plurality of labels includes a subject label.

[0013] Moreover, one of the information processing systems of the present invention comprises: 1. An information processing system for structuring unstructured text information, comprising: A structuring means and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to the structured word, which classifies the type of information of the word, and further assigning one of a plurality of labels including a subject label to the structured word; storing, among structured words included in the unstructured text information, words to which a label other than the topic label has been assigned (non-topic words) in association with one of the words to which the topic label has been assigned (topic words), the topic word and the associated non-topic word are stored as one data unit; The display processing means displays the structured words for each data unit.

[0014] Furthermore, in the above information processing system, the unstructured text information is the same or a common It is preferable that the information has a plurality of unstructured text information in which descriptions of an event may be contained, the structuring means stores the subject word and the associated non-subject words in a table including a plurality of cells corresponding to each of the plurality of labels, and the display processing means displays the plurality of data units in an order such that, when identical or common structured words are contained among the plurality of data units, the identical or common structured words are all lined up.

[0015] Furthermore, in the above information processing system, the display processing means may be configured to sort and display data units containing the most frequently occurring words in order of frequency with respect to the topic word, or to display the data units containing the most frequently occurring words separately.

[0016] Furthermore, in the above information processing system, the display processing means may be configured to remove data units including words that occur less frequently with respect to the topic word from the display targets, or to display data units including words that occur less frequently separately.

[0017] Furthermore, in the above information processing system, the unstructured text information is a part of medical document information, and the plurality of labels are: (1) a subject label; (2) a value label, a rating label, a frequency label, and / or a location label; and (3) a tense label and / or a date and time label; The signal processing unit 100 may be configured to include:

[0018] Furthermore, in the above information processing system, the display processing means may be configured to sort and display the data units in order starting from the data units containing the most frequently occurring words, or to distinguish and display the data units containing the most frequently occurring words.

[0019] Furthermore, in the above information processing system, the display processing means may be configured to rearrange and display the order of the data units, to distinguish and display the data units, or to delete the data units from the display targets based on display settings set for one or more of words, information types, labels, and data units.

[0020] Furthermore, in the above information processing system, major classifications including a plurality of information types are set for at least a portion of the information types, and the display processing means may be configured to display data units divided into the major classifications.

[0021] Furthermore, in the above information processing system, the display processing means may be configured to delete and display cells with labels that do not store words in all of the tables among the tables displayed side by side.

[0022] Furthermore, in the above information processing system, the display processing means may be configured to, when there are multiple data units in the same unstructured text information that contain the same or common structured words, integrate them into a single data unit and display them.

[0023] Moreover, one of the information processing systems of the present invention comprises: 1. An information processing system for structuring unstructured text information, comprising: The present invention includes a text division means, a structuring means, and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to each of the structured words, the information type of the word being classified; A method for storing structured words contained in the unstructured text information as one or more data units by associating the structured words with each other, Furthermore, if the unstructured text information includes a word indicating tense, the word is extracted and stored in association with other structured words included in the same unstructured text information; The display processing means displays the structured words for each data unit, Furthermore, when the word indicating the tense is future or past, the data unit including the word indicating the tense is displayed separately from other data units, or is deleted from the display targets.

[0024] Moreover, one of the information processing systems of the present invention comprises: An information processing system for structuring medical document information including a plurality of pieces of unstructured text information each having a different medical examination date, comprising: A structuring means and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to each of the structured words, the information type of the word being classified; storing structured words contained within the unstructured text information in association with one or more data units; The display processing means displays the structured words for each data unit, Furthermore, when identical or common structured words are contained among a plurality of data units, the plurality of data units are displayed in a line such that the identical or common structured words are aligned, or the plurality of data units in a line such that the identical or common structured words are aligned are displayed distinctly from other data units.

[0025] Furthermore, in the above information processing system, the information processing system may include a text segmentation means that segments the unstructured text information into text fragments that may include a plurality of semantically related words, and the structuring means may be configured to apply a structuring model to the text fragments to extract structured words. Effect of the Invention

[0026] In the information processing system of the present invention, instead of extracting each word independently, a structured model is applied to unstructured text information or divided text fragments that may contain multiple semantically related words to extract structured words, and the structured words contained in the unstructured text information or text fragments are treated as a set of data units, so that it is possible to associate and structure multiple semantically related words. Furthermore, when structuring multiple unstructured text information that may contain descriptions of the same or common phenomenon, if the same or common structured words are contained between different unstructured text information, the data units are displayed in a line so that the same or common structured words are aligned, so that related words in the unstructured text information or text fragments contained in the data units can be recognized together, and the changes and history of the same or common phenomenon can be easily understood based on the contents and changes of other related words. Furthermore, by assigning two different types of classifications, information type and label, to the structured words, it is possible to understand not only the type of information of the word but also the role or function of the word in the structure of a sentence. Furthermore, by setting a subject label, the gist of the content of the unstructured text information or text fragment can be grasped by the subject label, and the subject word and the associated non-subject words can be treated as a single data unit, which facilitates data management, understanding, and reconstruction of the structured text information. In addition, the structured words contained in the unstructured text information or text fragment are stored in a table containing multiple cells corresponding to multiple labels, and when the same or common structured words are contained between different unstructured text information, the table is treated as a single data unit and the cells are arranged so that the same or common structured words are all present. By displaying all the words in the same or common event, not only the words related to the same or common event but also the words before and after the same event contained in the unstructured text information or text fragment can be displayed in close proximity, making it easier to grasp the change in the event over time. In addition, by extracting words related to tense meaning present, future, or past and storing them in association with other structured words contained in the same unstructured text information or text fragment, it is possible to grasp whether the event was performed in the past, was performed when the unstructured text was input, or has not yet been performed. In particular, in electronic medical record information, it is important to be able to accurately grasp the effects of treatment and / or medicine, and for this purpose, it is necessary to accumulate facts that were actually performed when the information was input, and if past information and future information are mixed therein, the correct progress cannot be grasped and the reliability of the entire information is lost. Other effects will be described in the embodiment. [Brief description of the drawings]

[0027] [Figure 1] FIG. 1 is a schematic diagram showing an example of the overall configuration of a personal information anonymization system according to the present invention. [Diagram 2] A diagram showing an example of freely inputtable items in electronic medical record information [Diagram 3] A diagram showing an example of unstructured text information divided into text fragments. [Figure 4] A diagram showing an example of structuring a text fragment. [Diagram 5] FIG. 1 is a diagram showing an example of a table including a number of cells each corresponding to a number of labels. [Figure 6] An example of structuring text fragments and storing them in the table in Figure 5. [Figure 7] (A) to (C) are separate tables listing data units that contain frequently occurring words. [Figure 8] Data units containing words implying future or past tense have been removed. [Figure 9] A diagram showing data units sorted by topic word information type. [Figure 10]A diagram showing data units sorted by topic word information type. [Figure 11] A diagram showing data units sorted by topic word information type. [Figure 12] The diagram shows the data units that contain the most frequently occurring words sorted to the top, and the cells that do not contain any words deleted from all tables. [Figure 13] The figure shows that the label cells that do not contain words are deleted from all tables of the information type. [Figure 14] 1 is a flowchart showing the overall processing flow of an information processing system. [Figure 15] 1A and 1B are flowcharts showing an example of a structuring process. [Figure 16] Flowchart showing an example of processing from structuring processing to display processing DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] [Outline of information processing system] 1 is a schematic diagram showing an example of the overall configuration of an information processing system 1 of the present invention. The information processing system 1 is for structuring unstructured text information, and includes at least a structuring means 3 and a display processing means 4. The information processing system 1 may further include a text dividing means 2 for dividing the unstructured text information into text fragments that may include a plurality of semantically related words, and a text converting means 5 for converting characters from image data, audio data, video data, etc., that include characters into text, as necessary. In addition, it is preferable that the information processing system 1 includes an acquisition means and an output means (display means) so that a user can confirm, correct, and change information, and the structured information can be inspected. In the information processing system 1 of the present invention, when the acquired information is non-text information, the text conversion means 5 converts the information into text and generates unstructured text information. When the length of the unstructured text information exceeds the capacity of the structuring means 3 or the display processing means 4 or when accuracy is to be improved, the text division means 2 may divide the unstructured text information into text fragments that may contain multiple semantically related words. Thereafter, the structuring means 3 applies a structure model to the unstructured text information or the text fragments to extract structured words, and classifies the information type of the structured words. The structured words included in the unstructured text information or text fragments are associated with each other and stored as one or more data units, and the display processing means 4 displays the structured words for each data unit. In a first embodiment, in order to structure a plurality of pieces of unstructured text information that may contain descriptions related to the same or common phenomenon, the structuring means 3 applies a structuring model to the unstructured text information or text fragments to extract structured words, assigns to the structured words an information type that classifies the type of information of the words, and further assigns one of a plurality of labels classified according to the role or function of the words in the sentence structure, stores the structured words included in the unstructured text information or text fragments in a table including a plurality of cells corresponding to each of the plurality of labels, and the table is treated as one data unit. The display processing means 4 displays the structured words for each data unit, and further, when the same or common structured words are included among the plurality of data units, displays the plurality of data units in a line so that the same or common structured words are aligned. In a second embodiment, in order to structure unstructured text information, the structuring means 3 applies a structuring model to the unstructured text information or text fragments to extract structured words, assigns to the structured words an information type that classifies the type of information of the word, and further assigns one of a number of labels including a topic label, associates and stores, among the structured words contained in the unstructured text information or text fragments, words that have been assigned a label other than a topic label (non-topic words) with one of the words that have been assigned a topic label (topic words), treats the topic word and the associated non-topic word as one data unit, and the display processing means 4 displays the structured words for each data unit.In a third embodiment, in order to structure unstructured text information, a structuring means 3 applies a structuring model to unstructured text information or text fragments to extract structured words, assigns to the structured words an information type that classifies the type of information of the word, and associates and stores the structured words contained in the unstructured text information or text fragments as one or more data units. Furthermore, if the unstructured text information or text fragments contains a word indicating tense, the word is extracted and stored in association with other structured words contained in the same unstructured text information or text fragment. A display processing means 4 displays the structured words for each data unit. Furthermore, if the word indicating tense is future or past, the data unit containing the word indicating tense is displayed separately from other data units, or is deleted from the display targets. In the fourth embodiment, in order to structure medical document information (including documents such as medical certificates and unstructured text stored outside electronic medical records) including multiple unstructured text information with different consultation dates, the structuring means 3 applies a structuring model to the unstructured text information or text fragments to extract structured words, assigns information types to the structured words that classify the types of information of the words, associates the structured words included in the unstructured text information or text fragments and stores them as one or more data units, the display processing means 4 displays the structured words included in the unstructured text information or text fragments as one or more data units, and when the same or common structured words are included among the multiple data units, the display processing means 3 displays the multiple data units in a line so that the same or common structured words are aligned. Note that the first to fourth embodiments are referred to as the first, second, third, and fourth for convenience due to differences in processing contents, but multiple embodiments can be realized in one system, or each can be realized independently. It is also possible to appropriately combine the processing of different embodiments.

[0029] In this way, the information processing system 1 of the present invention does not extract individual words independently, but extracts structured words by applying a structuring model to unstructured text information or text fragments that may contain multiple semantically related words, and associates and handles the structured words contained in the unstructured text information or text fragments as one or more data units, making it possible to associate and structure multiple semantically related words. Furthermore, the structured words are classified into two different types, namely, information type and sentence structure. By assigning one of a plurality of labels classified according to the role or function of the word, not only the type of information of the word but also the role or function of the word in the structure of a sentence can be understood. In addition, among a plurality of pieces of unstructured text information, descriptions related to the same or common events may exist, but if the information is simply structured, the context with other unstructured texts is lost. However, when the same or common structured words are included among a plurality of data units, by displaying the plurality of data units in a line such that the same or common structured words are aligned, not only the words related to the same or common events but also the words before and after the same words included in the unstructured text information or text fragment can be displayed in close proximity, making it easier to understand changes in the event. In particular, by storing the structured words included in the unstructured text information or text fragment in a table including a plurality of cells corresponding to each of a plurality of labels, it is possible to provide a regularity in the arrangement of the labels within the data units, making it easier to understand the contents of the data units. Furthermore, if a subject label is set as one of the multiple labels, the subject label allows the gist of the content of the unstructured text information or text fragment to be understood, and the subject word and the associated non-subject words are treated as a single data unit, making it easier to manage, understand, and reconstruct the structured text information. Furthermore, if the unstructured text information or text fragment contains a word indicating tense, such as a past date or a word "last time" indicating "past," or a future date or a word "next time" or "schedule" indicating "future," the content of the unstructured text information or text fragment is past or future content, and is not information that was actually carried out at the time of inputting the unstructured text information or text fragment. Therefore, by displaying the content separately from other data units or deleting it from the display target, the possibility of selecting incorrect information can be reduced, and the chronological progress of facts can be easily understood.

[0030] The information processing system 1 can be realized by a computer including at least an arithmetic device such as a CPU that executes arithmetic processing of a program, and a storage device such as a RAM or a hard disk that stores information. The information processing system 1 may be realized by one computer, or may be realized by multiple computers with functions distributed. The computer includes various information processing devices such as a personal computer, a workstation, a server, a mainframe, a supercomputer, and a microcomputer, and a part or all of the functions of the information processing system may be realized by an arithmetic device such as a tablet terminal, a PDA (Personal Digital Assistant), or a multi-function mobile phone (smartphone, i-phone (registered trademark)). The computer may be in a cloud format. Furthermore, the computer may have a display device such as a display (screen), an input device such as a keyboard or a pointing device (mouse, numeric keypad, touch panel, etc.), and a communication device that transmits and receives information via a network such as the Internet or a LAN. Each means of the information processing system 1 can be realized by executing a program that executes the function in an arithmetic device, or by operating a module or device that executes the function.

[0031] Here, unstructured text information refers to information that includes at least unstructured character information, and is information to be structured as a target of the information processing system 1 of the present invention. Unstructured text information may include not only unstructured character information, but also structured data. Unstructured character information includes freely input text information, text information digitized by the text conversion means 5 from image data including characters, text information digitized by the text conversion means 5 from audio data, and text information digitized by the text conversion means 5 from the image and / or audio of video data. Here, the text conversion means 5 may be a device that performs optical character recognition (OCR (or Reader )) technology and voice recognition technology can be used. Furthermore, the target of the information processing system 1 of the present invention may be a plurality of pieces of unstructured text information in which descriptions relating to the same or common events may exist. Examples of unstructured text information include medical document information such as electronic medical record information having items that can be freely entered. Electronic medical record information may include multiple documents relating to the same patient. Diagnosis histories, treatment histories, etc. are stored, and records of the same or common events that change over time are present. In addition to electronic medical records, there are documents such as medical certificates, records registered in other systems, medical and nursing records that require follow-up observation, and even in non-medical cases, records of the same or common events that change over time are present in the minutes of projects that are assumed to be held multiple times, and response management at call centers that provide customer support.

[0032] In addition, structuring unstructured text information means converting unstructured character information into structured data or semi-structured data. Specifically, the information type to be structured is defined in advance according to the content of the information, and a word (target word) that corresponds to one of the information types is extracted from the words included in the unstructured character information, and the information type is associated with the target word. The target word included in the unstructured text information may be used as is after structuring, but in the case of unstructured text information, even if the words are synonymous, the target word may be written differently depending on the inputter, so the standard writing of the word (standard word) may be associated with the target word, or the target word may be replaced with a standard word. In this specification, the term "structured word" includes both the case where the target word included in the unstructured text information is used as is, and the standard word that replaces it. The standard word may be a formal name listed in a dictionary or a commonly used name, but it is preferable that the word be written in a common manner at least in the information processing system. Since the word structured in this way is associated with a predefined information type, it is suitable for data processing such as search, tabulation, and comparison, and is easy to use.

[0033] FIG. 2 is an example of freely inputtable items in electronic medical record information for a certain patient, and includes multiple pieces of unstructured text information with different consultation dates. In FIG. 2, the medical record ID ("ID" in FIG. 2), patient ID, consultation date, and unstructured text information are associated, but the electronic medical record information may include other information. For each medical record ID, the patient ID, consultation date, and unstructured text information are entered, and FIG. 2 shows unstructured text information for three consultations for the same patient with patient ID A001. The unstructured text information in the electronic medical record information includes information obtained during the consultation, such as the patient's chief complaint, the doctor's diagnosis, test contents and results, prescriptions, and treatment plans. In Figure 2, unstructured text information is entered using the SOAP method. The S (Subjective) items include subjective information such as the symptoms (chief complaint) complained by the patient, medical history, and information obtained from family and related parties. The O (Objective) items include objective information obtained from tests, examinations, and observations. The A (Assessment) items include evaluations and diagnoses derived by analyzing and considering the information in S and O. The P (Plan) items include the evaluations and diagnoses derived by analyzing and considering the information in S and O. The items mainly include treatment plans and follow-up observation plans. Electronic medical record information may be entered without using the SOAP method, and items other than SOAP may be added.

[0034] The text division means 2 is a means for dividing the unstructured text information into text fragments that may contain multiple semantically related words when the unstructured text information is too long for the processing capacity of the structuring means 3 or the display processing means 4, or when it is desired to improve the accuracy of structuring. Simply extracting important or necessary words independently from the unstructured text information makes it difficult to determine the meaning of the words in a sentence and to correctly structure the information. For this reason, the text division means 2 extracts text fragments that are combinations of words whose meaning can be determined. The text fragments may be clauses (a group of multiple clauses), sentences (separated by periods), paragraphs (a group of multiple sentences), items (e.g., each item of SOAP), or units separated by symbols such as periods, commas, commas, spaces, slashes, colons, and line breaks, but it is preferable that freely inputted unstructured text information can be divided into more flexible units, not necessarily into sentences with correct grammar. For example, even if the unstructured text information is a list of multiple words, some combinations of words may have meaning, and it is preferable to be able to select a combination of words that has meaning from the list of multiple words as a text fragment. Depending on the contents of the information, text fragments may become words. If the unstructured text information is originally of a length suitable for processing by the structuring means 3 and the display processing means 4, it is not necessary to provide the text dividing means 2. However, it is preferable to provide the text dividing means 2 because structuring is possible even if the processing capabilities of the structuring means 3 and the display processing means 4 are somewhat low, and because it is possible to increase the processing speed and improve accuracy.

[0035] The text division means 2 may divide the text into text fragments by dividing it by symbols such as periods, commas, commas, spaces, slashes, colons, and line breaks, or by item names, but in order to enable more flexible division, it is preferable to use a text division model constructed by a natural language processing model that divides the text by logic, machine learning, or neural network. For normal text, the context can be determined by a model (text division model) that employs a natural language processing algorithm, and semantically related text fragments can be extracted from the context. In addition, in cases where technical words, unique abbreviations, and syntax are included, such as in electronic medical records, or where the text does not follow accurate grammar, such as in notes, it is often difficult to determine the context or meaning as it is. For this reason, a text division model may be constructed in which the meaning of the original sentence is first decoded by a human so that the context can be determined, and the original sentence and the decoded sentence are matched and then learned as sample data, so that the context can be determined. Furthermore, a text segmentation model that can extract text fragments by learning the correspondence between the original sentence, the decoded sentence, and the text fragment classification as sample data may be constructed and used as text segmentation means. By learning the classification of text fragments in this way, text fragments can be segmented into any unit. Text fragments may also be extracted using a natural language processing model that has undergone deep learning using a neural network. Note that the text segmentation model can also function as a structured model if sample data is used to which word information types and labels are assigned.

[0036] Examples of multiple semantically related words include combinations of words such as "drug name" and "action (side effects)", "test name" and "test value", "symptoms" and "site", as well as "treatment method" and "scheduled date (which may be a combination of time and date)", and "treatment method" and "evaluation (improvement, deterioration)". The above is merely an example, and a labeling rule of A, B, C, D, and E can be formulated and generalized for sentences such as "how did event A become due to event B", "event A became state C", "event A was evaluated as D", and "what to do when event A becomes E", allowing general processing regardless of a specific case. The text division means 2 divides unstructured text information into text fragments that may contain multiple semantically related words, but is ultimately managed in data units. Data units do not need to correspond one-to-one with text fragments, and there is a process of associating words structured by a structured model afterwards, so semantic relevance does not need to be strictly determined in text division.

[0037] FIG. 3 is an example of the unstructured text information of FIG. 2 divided into text fragments. In FIG. 3, the medical record ID ("ID" in FIG. 2), the text fragment number ("TN" in FIG. 2), the consultation date, and the text fragment are associated with each other. In FIG. 3, the unstructured text information is divided into text fragments by a period and a line break. The text fragment number TN is a number for identifying a text fragment, and is assigned to each text fragment. Although the numbers are consecutive in FIG. 3, they do not have to be consecutive, and may be symbols or letters instead of numbers, or a combination of these. In FIG. 3, the text division means 2 divides the unstructured text information of ID1 into 7 text fragments, divides the unstructured text information of ID2 into 7 text fragments, and divides the unstructured text information of ID3 into 9 text fragments, resulting in a total of 23 text fragments.

[0038] The structuring means 3 applies a structuring model to unstructured text information or text fragments to construct the structure. The structured words are extracted, and for each structured word (target word and / or standard word), an information type that classifies the type of information of the word is assigned, and the structured word is associated and recorded. For example, structured words contained within unstructured text information or text fragments may be associated and stored as one or more data units; The structured words may be assigned one of a plurality of labels classified according to the role or function of the word in the sentence structure, the structured words included in the unstructured text information or the text fragment may be stored in a table including a plurality of cells corresponding to each of the plurality of labels, and the table may be stored as a single data unit; The structured words may be assigned one of a number of labels including a thematic label, and non-thematic words among the structured words contained in the unstructured text information or text fragment may be stored in association with one of the thematic words, and the thematic word and the associated non-thematic word may be stored as a single data unit; If the unstructured text information or text fragment contains a word that indicates tense, the word may be extracted and stored in association with other structured words contained within the same unstructured text information or text fragment. The data unit is a criterion for displaying structured words in the display processing means 4, and for example, whether to display or not and rearrangement are processed in data units.

[0039] The structured model can be constructed by learning unstructured text information or words of text fragments to which information types and labels are assigned as necessary as sample data. In particular, in cases where technical words, unique abbreviations and syntax are included, such as in electronic medical records, or where the text does not follow accurate grammar, such as in notes, it is often difficult to determine the context and meaning as is. For this reason, a structured model may be constructed in which the meaning of the original sentence is once decoded by a human so that the context can be determined, and the original sentence and the decoded sentence are matched and trained as sample data to determine the information type and label. Furthermore, in addition to the original sentence and the decoded sentence, information types, labels, and / or data units are further matched and trained as sample data to construct a structured model in which appropriate information types, labels, and / or data units can be selected. In addition, the structured model may refer to a dictionary database to determine whether or not a word is a target word, or may extract standard words. In particular, it is preferable to be able to refer to a dictionary database related to technical terms, such as "disease name", "test name", "medicine name", etc. Also, a structured model that determines information type and labels may be constructed using a natural language processing model (either a general-purpose natural language processing model or a natural language processing model specialized for medical documents) that has undergone deep learning using a neural network. There are also cases where it is better to use unstructured text from other systems or documents other than electronic medical records, and such information may be used to extract information using a structured model. Also, a system that provides input support to make structuring easier from the input stage of electronic medical records can be used.

[0040] The information type is set appropriately according to the contents of the unstructured text information, and the type of information that needs to be extracted from the unstructured text information is specified and set as the information type. For example, in the case of medical document information such as electronic medical record information, information types related to medical information include "disease name," "test name," "drug name," "symptoms," "treatment content," "regimen," "outcome," "action (side effects)," "number," "symbol," "evaluation," "site," "unit," "tense," "date and time," and further, information types other than medical information include "person's name," "facility name," "place name," "address," "telephone number," and "relationship." In the case of project minutes, etc., examples include "speaker," "request" and "task content" according to the contents of the project, and in the case of response management of a call center that provides customer support, examples include "customer name," "complaint," "response to complaint," and "means of communication." These information types are medical information as information to be accumulated and collected among the information entered in the freely inputtable fields of electronic medical record information, and personal information as information to be anonymized. When utilizing electronic medical record information as medical information such as registry data, what is important is the actual current event, and past and future events become noise. Therefore, it is preferable to structure words related to "tense" and associate them with related words, since it is possible to distinguish data units containing words related to "tense" in the past or future. However, the above information types are only examples, and "numerical value" and "unit" may be combined without being separated and distinguished as "weight", "concentration", "number of times", and "temperature", or "site" may be included in "symptoms" without being independent. In this way, the words to be structured are not limited to single words such as "numerical value" and "unit", and may include combinations of multiple words depending on the content of the information type, such as "weight". In addition, depending on the content of the information type, it is preferable to provide a dedicated dictionary database for the structured model to refer to. Furthermore, it is preferable to include an inspection process in which a human visually checks the correctness of the target word and information type, and it is more preferable to feed back the results of the inspection process to the structured model and allow it to re-learn.

[0041] FIG. 4 is an example of structuring the text fragment of FIG. 3. In FIG. 4, one or more combinations of a text fragment number ("TN" in FIG. 4), a data unit number ("DN" in FIG. 4), a target word included in the text fragment ("word"), a standard word of the target word ("standard"), and an information type of the target word ("type") are associated with each other. The data unit number DN is a number that identifies a data unit and is assigned to each data unit. The data unit number DN is a consecutive number in FIG. 4, but it does not have to be a consecutive number, and may be a symbol or a character instead of a number, or a combination of these. In FIG. 4, 29 data units are generated from 23 text fragments. One or more data units are generated by associating words that are semantically related from one text fragment. For example, in DN1, the words "asthma" and "visiting the hospital" are associated with each other, and it can be understood that the patient is visiting the hospital for asthma. Even if one text fragment contains multiple semantically different contents, it is preferable to make it into multiple data units. For example, when a single text fragment contains multiple words classified as the subject label, when it is a compound or complex sentence, when it contains multiple tenses, etc., it is preferable to treat each of these as separate data units. For example, the sentence "I feel fine, but I have been coughing since last year's pneumonia" in TN2 is a compound sentence consisting of two sentences, "I feel fine" and "I have been coughing since last year's pneumonia", and further, "last year's pneumonia" is a past event, and it is preferable to treat it as a separate data unit. In the sentence "Igg1100 has risen slightly, so I will have a re-examination next month just to be sure" in TN5, "Igg1100 has risen slightly" indicates a current event, but "re-examination next month" indicates a future event, and it is preferable to treat it as a separate data unit. In addition, in the sentence "Pet appointment for CT scan at Hospital A on 6 / 5" in TN9, there are two items that belong to the subject label, "ct" (type: examination name) and "pet" (type: examination name), and they are treated as separate data units. In TN18, there are two words in the subject label, "igg" (type: test name) and "treatment" (type: general treatment), so the data units are separated according to the meaning of the sentence. However, the structured model built by machine learning outputs a solution with a statistically high probability, so the above criteria do not need to be applied mechanically.In FIG. 4, even if there are multiple words classified as the subject label, they are treated as one data unit (DN1, 5, 9, 10, 17, 18, 25, 26). In addition, even if different text fragments are semantically related, they may be associated and treated as one data unit. For example, TN15 and 16 are divided into two text fragments because they are separated by a period, "My back hurts after all... It gets worse when I lie on my back." However, they are written in the same item, subjective information [S], and their content is semantically related, meaning "My back pain gets worse when I lie on my back," following the previous sentence, "My back hurts." For this reason, rather than treating them as separate data units from DN20 and 21 in FIG. 4, the two may be associated and treated as one data unit. In other words, structured words contained in multiple semantically related text fragments may be associated with each other and treated as one data unit. Such processing can be realized by a structured model that has been trained using multiple semantically related text fragments as sample data. In FIG. 4, for all target words contained in the text fragments, the target word, standardized word, and information type are assigned. The target words and the information types are stored in a single table in association with each other. However, the target words and the information types may be associated with each other, or the standard words and the information types may be associated with each other, or the association may be changed depending on the information type of the target words. Also, the target words, the standard words, and the information types may be stored separately in multiple tables associated with each other, for example, multiple tables linked by a common data unit ID. From Figure 4, it can be seen that the structured words contained in a text fragment are associated and stored as one or more data units, and that if a text fragment contains a word indicating tense, that word is extracted and stored in association with other structured words contained in the same text fragment.

[0042] In the present invention, the structured words may be given multiple labels classified according to the role or function of the word in the sentence structure in addition to the information type. The labels are set in advance according to the content of the information, but it is preferable to include at least the label of "theme", which is the central content of the data unit. In addition, in the case of electronic medical record information, it is preferable to include the label of "tense" so that it is possible to understand whether the content of the associated word is a past, present, or future event. The unstructured text information is divided into text fragments that may include multiple semantically related words by the text division means 2, and labels are assigned according to the meaning, role, and function of the corresponding word in the text fragment or unstructured text information. For example, in the case of electronic medical record information, since the text fragments are related to medical information, words of information types such as "disease name", "examination name", "medicine name", "symptoms", "general treatment", "treatment category", and "regimen" may be classified into the label of "theme". Labels other than the subject are determined according to the content of the information. For example, the labels may be two types, "the subject" and "non-the subject", or three types, "the subject", "tense" (including a word that means it is past, present, or future, date, and time), and "other", or "other" may be further subdivided. In the case of two types, "the subject" and "non-the subject", it may be realized by setting only the label "the subject", or in the case of three types, "the subject", "tense", and "other", it may be realized by setting two labels, "the subject" and "tense". In the case of electronic medical record information, as an example of typical multiple labels, all or some of "the subject", "value", "evaluation", "number of times", "site", "tense", and "date and time" may be set. The "theme" label is the central content of the data unit, and is given to words of information types such as "disease name", "test name", "medication name", "symptoms", and "regimen". When there are multiple topic words in a text fragment, it is preferable to generate multiple data units corresponding to the multiple topic words. The "value" label is mainly given to test values ​​and peaks. The "evaluation" label is mainly given to improved, worsened, recurrence, improvement, +, -, etc. The "number of times" label is mainly given to the number of treatments and units, etc. The "part" label is mainly given to words related to parts of the body.The "tense" label is mainly given to words that mean either the past, present, or future, and allows the user to determine when the associated words refer to an event. The "date and time" label is mainly given to dates and times. In addition, information types and labels do not correspond one-to-one, and may include multiple information types such as the label "subject". Different labels may be attached to words of the same information type depending on the contents of the sentence or combination with other words. For example, even if the information type is "numerical value", if there is a word of information type "test name" before or after it, it is judged to mean the test result and the label "value" is attached, and if there is a word of information type "unit" such as "mg" or "ml" after it, it is judged to mean the dosage and the label "value" is attached, but if there is a word of information type "unit" such as "times" after it, the label "number of times" is attached. In this way, "labels" may change depending on the context, and it is not sufficient to simply extract words listed in a dictionary, and it is necessary to apply a structured model to label (attach a label) based on the words before and after it. One label may be attached to one word, or one label may be attached to a combination of multiple words (for example, a "value" label may be attached to a combination of a word of "numerical value" and a word of "unit"). Furthermore, it is possible for a user to visually determine the label. It is preferable to include an inspection process for checking the accuracy of the label, and it is more preferable to feed back the results of the inspection process to the structured model for re-learning. The labels are classified from a different perspective than the information types, but the names of the labels and the names of the information types do not need to be different. For example, "date and time" and "evaluation" are common names for both the labels and the information types. From the perspective of grammar, the labels may be parts of speech such as "noun", "verb", "adjective", and "adverb", or the components of sentences such as "subject", "predicate", "object", and "modifier".

[0043] FIG. 5 is an example of a table (data unit) including a plurality of cells corresponding to a plurality of labels. In FIG. 5, the subject label displays the target word, standard word, and information type, the tense label displays the target word and information type, and the value, evaluation, number of times, part, and date and time labels are configured to display only standard words. This is because the subject is important information, and while checking the description in the actual unstructured text information by the target word, it is possible to recognize a general name by the standard word, and it is also possible to check which of the multiple information types included in the subject it belongs to. In addition, the tense label is also important information, and since there are many ambiguous expressions and additional information, it is preferable to be able to check the description in the actual unstructured text information by the target word. In addition, it is preferable to display the information type so that it is possible to know whether the tense is past, present, or future. For the other labels, only standard words are displayed, but the target word and information type may also be displayed.

[0044] FIG. 6 is an example in which the text fragment of FIG. 3 is structured and stored in the table of FIG. 5. In FIG. 6, the chart ID ("ID" in FIG. 6), the text fragment number ("TN" in FIG. 6), the data unit number ("DN" in FIG. 6), and the structured words stored in the label table of FIG. 5 are associated with each other. In FIG. 6, 38 data units are generated from 23 text fragments. Each cell in the table of FIG. 5 corresponds to multiple labels, and storing structured words makes it easier to understand the meaning of the structured words. In addition, when one text fragment contains multiple themes, it is preferable to separate tables for each theme and generate data units for each theme. In this case, the structuring means 3 may store one word in the cells of multiple data units. For example, the sentence "Contact if pain worsens or fever" in TN12 is a combination of two sentences, "Contact if pain worsens" and "Contact if fever occurs." The structuring means 3 separates the topic words "pain" and "fever" into cells of DN21 and DN22, and stores the word "if: future tense" in association with both DN21 and DN22. If the text fragment does not contain a word with a corresponding label, no word is stored in the cell with that label, and the cell is left blank. In addition, in unstructured text information, the topic is often omitted. For example, the topic cells of DN8, 10, 25, 27, 29, and 37 are blank. The structuring means 3 may complete the omitted topic word, or the user may visually inspect the result of the completion by the structuring means 3 and feed back the inspection result to the structured model. In addition, the user may input the omitted topic word, and the user's input result may be fed back to the structured model. For example, for DN8 and 25, the appointment date and time may be completed, and the word "visit" may be completed in the subject as in DN38. Also, for DN10, the name of the test to be retested may be completed, and for DN37, the regimen or drug name may be completed. Furthermore, for DN27, it may be integrated with DN26, and for DN29, the subject and evaluation of DN28 may be copied. It is preferable to change the color or blink the words completed or changed by the structured model to call the user's attention so that the user can confirm them. Also, at least for the cells of subject and tense, one cell is provided in one table, but for the cells of other labels, multiple cells of the same label may be provided and associated with one cell of subject or tense. Or, at least for the cells of subject and tense, one word is stored in one cell, but for the cells of other labels, multiple words may be stored in one cell. For example, DN28 and 29 are combined into one, and the number of cells of part is increased to two. Alternatively, "third rib" and "left ilium" may be stored, or the two words "third rib / left ilium" may be stored in the site cell. It can be seen from Fig. 6 that the structured words contained in the text fragment are stored in a table containing a number of cells corresponding to each of a number of labels, the table is stored as a single data unit, the structured words are assigned one of a number of labels including a topic label, a non-topic word among the structured words contained in the text fragment is stored in association with one of the topic words, and the topic word and the associated non-topic word are stored as a single data unit. It can also be seen from Fig. 6 that the structured words contained in the text fragment are associated and stored as one or more data units, and that if a word indicating tense is contained in the text fragment, the word is extracted and stored in association with other structured words contained in the same text fragment.

[0045] The display processing means 4 displays the structured words on the display device for each data unit, and may rearrange the data units, display some data units in a distinguished manner, or delete some data units from the display target so that the information is easy to see and use. For example, the display processing means 4 may rearrange the data units by information type. Furthermore, when the same or common structured words are included among a plurality of data units, the display processing means 4 may display the plurality of data units arranged so that the same or common structured words are aligned, or may display the plurality of data units arranged so that the same or common structured words are aligned, distinguished from other data units. Furthermore, the display processing means 4 may rearrange and display the data units in order starting from the data units including the most frequently occurring words, or may display the data units including the most frequently occurring words in a distinguished manner. The display processing means 4 may delete the data units including the least frequently occurring words from the display target, or may display the data units including the least frequently occurring words in a distinguished manner. Here, displaying in a distinct manner means displaying it in a manner different from other data units, for example by changing the character color, font, character size, cell color, etc., or by providing a partition between it and other data units, or by displaying it in a separate table from the other data units, or by displaying it on a separate screen (separate window).

[0046] In Fig. 7(A)-(C), the display processing means 4 extracts data units DN7, 14, and 23 in Fig. 4, which include the word "igg" that occurs frequently, rearranges them so that the word "igg" is consistent, rearranges them so that the "value" of the information type is consistent because it is common, and rearranges them so that the "evaluation" is consistent because the two are common, and creates a separate table. Fig. 7(A) shows a separate table of the combination of the target word, standard word, and information type as it is. Fig. 7(B) shows a combination of the target word, standard word, and information type only for the extracted word "igg", and displays the target word and information type for the other words. Fig. 7(C) shows a standard word only for the word "igg", and displays the target word for the other words. From Fig. 7(A)-(C), it can be seen that the test value of igg rose to 1100 on April 28, remained high at 1083 on May 24, and improved to 689 on June 21. Since displaying many items as in FIG. 7(A) requires a large display space, some information may be omitted to reduce the display space as in FIG. 7(B) and (C). Omitting one of the overlapping target words and standard words as in FIG. 7(B) can save display space without significantly reducing the amount of information displayed. Since the information type can be understood with a certain degree of knowledge and experience, it is also possible to further save space by omitting the information type as in FIG. 7(C). In FIG. 7, overlapping words are placed at the very end, and then sorted in descending order of the number of overlaps of the information type.

[0047] When the word indicating tense is future or past, the display processing means 4 displays the data unit including the word indicating tense separately from other data units, or deletes it from the display target. Fig. 8 shows a structured word list in Fig. 4 from which data units including words meaning future or past tense have been deleted. In Fig. 8, since those that are clearly past and future events have been deleted, it is possible to extract mainly those related to current events, and to grasp the progress of symptoms and changes in condition.

[0048] When a label is assigned to a structured word, the display processing means 4 may rearrange and display data units including those with a high frequency of occurrence in relation to the subject word, or may display data units including those with a high frequency of occurrence in a distinguished manner. The display processing means 4 may also delete data units including those with a low frequency of occurrence in relation to the subject word from the display target, or may display data units including those with a low frequency of occurrence in a distinguished manner. The display processing means 4 may also delete cells with labels in which no words are stored in all tables among the tables displayed in a line, and may also rearrange data units, display some data units in a distinguished manner, or delete some data units from the display target based on display settings set for one or more of the words, information types, labels, and data units. As the display settings, for example, "importance," "sort order," or "display position" may be set for some or all of the words, information types, labels, and / or data units, or "non-display" may be set for removing them from the display target. The display settings may be configurable by the user, or may be automatically configured by the structuring means 3 or the display processing means 4 based on the results of machine learning. However, if the settings can be configured automatically, it is preferable that the settings be changeable by the user.

[0049] Figures 9 to 11 show the data units (tables) of Figure 6 rearranged by information type of the topic word, and in Figures 9 to 11, the medical record ID ("ID") and text fragment number ("TN") are common, and in Figure 9, data units with topic words of "test name" and "disease name" are arranged, in Figure 10, data units with topic words of "symptoms" and "medicine name" are arranged, and in Figure 11, data units with topic words of "treatment" and other data units are arranged. By rearranging in this way, the data are organized by information type, and each data unit contains related words, making it easier to understand the content. Figure 12 shows that, for data units whose subject word is "Test Name," data units that contain the word "igg" that occurs frequently (data units included in TN5, 10, and 18) are extracted, rearranged to the beginning, and displayed in the same column so that the same word "igg" is aligned, and the cells (rows) labeled "Number of times," "Area," and "Date and time," which do not contain any words, are deleted from all of the tables. Figure 13 shows the display of the information types "Disease name", "Symptoms", "Drug name", and "Treatment" from Figures 9 to 11, with the deletion of cells with labels that do not store words in all tables for that information type. By deleting unnecessary cells in this way, space can be saved.

[0050] The display processing means 4 may integrate the same unstructured text information among the tables displayed side by side into one table for display. For example, in the "Symptoms" section of Fig. 10, the two data units including the word "lower back pain" included in TN9 and 11 have the same medical record ID (unstructured text information), and may be integrated into one table. In this case, all the words in the data unit of "lower back pain" in TN11 are included in the data unit of TN9, and TN9 further includes a word for tense, so the data unit of "lower back pain" in TN11 can be deleted and the data unit of TN9 can be left.

[0051] In addition, the structuring means 3 or the display processing means 4 may set a large classification including a plurality of information types for at least a part of the information types, particularly for all or a part of the information types belonging to the subject label, and the display processing means 4 may display the data units by dividing them into the large classifications. For example, The major categories may be displayed in descending order of frequency of occurrence, except for those that have been specified. By dividing into major categories, data units with the same meaning can be displayed close to each other, and it is visually easier to view the data units that appear most frequently, arranging them from left to right, except for those that are particularly important and specified separately. The major categories can be set or specified separately by using the display settings, and may be set in advance or re-set while viewing the extracted data. For electronic medical record information, major categories may be set, for example, as "diagnosis and progress," "treatment and medication," and "basic information," but the major categories are not limited to these three major categories and may be set from other perspectives. "Diagnosis and progress" classifies information types related to diagnosis contents and progress information, including information types such as "disease name," "symptoms," and "examination." "Treatment and medication" classifies information types related to treatment and medication, including information types such as "medication name," "regimen," and "action (side effects)." "Basic information" classifies basic information, including, for example, "height," "weight," "blood pressure," "pulse," and "medical history."

[0052] In this way, the information processing system of the present invention not only structures important words, but also structures the surrounding context and / or makes it easy to grasp changes in content over time, enabling the following uses: In clinical research etc., structured registries that include time series trends have been established, and by setting information types and / or labels corresponding to the registry, extraction can be performed in accordance with the registry. This refers to the extraction of medical information from electronic medical records in order for pharmaceutical companies, Contract Research Organizations (CROs), etc. to organize the data required for clinical trials. It will be utilized by certified businesses under the Next Generation Medical Infrastructure Act. Creating medical documents such as referral letters, medical information release forms, and other electronic medical record information from the structured text information of electronic medical record information.

[0053] [Specific processing] 14 is a flowchart showing the overall processing flow of the information processing system of the present invention. When the text division means 2 acquires unstructured text information (S11), it divides the unstructured text information into text fragments that may contain multiple semantically related words (S12). Next, the structuring means 3 applies a structuring model to the text fragments to extract structured words from the text fragments, assigns information types to the structured words that classify the types of information of the words, assigns labels as necessary, and stores them in association with each data unit (S13). The display processing means 4 displays the structured words for each data unit (S14).

[0054] 15(A) and (B) are flow charts showing an example of structuring processing. In FIG. 15(A), the structuring means 3 extracts target words from text fragments (S21), associates standard words as well as the structured words, assigns information types that classify the types of information of the words to the structured words (S22), associates them with each data unit, and stores them (S23). In FIG. 15(B), the structuring means 3 extracts target words from text fragments (S31), associates standard words as well as the structured words, assigns information types that classify the types of information of the words to the structured words (S32), further assigns labels (S33), and stores them (S34) in association with each data unit.

[0055] Fig. 16 is a flow chart showing an example of processes from structuring to display processing. In Fig. 16, the structuring means 3 extracts target words from text fragments (S41), associates standard words as well as the like as necessary, assigns information types to the structured words that classify the types of information of the words (S42), and further assigns labels (S43). Thereafter, the display processing means 4 deletes duplicate items within the same unstructured text information (S44), sorts them by importance and / or frequency of occurrence (S45), deletes unnecessary cells that are all blank (S46), and displays the results. The rearranged state may be stored in the storage means. [Explanation of symbols]

[0056] 1. Information Processing Systems 2. Text Segmentation Methods 3 Structuring means 4. Display Processing Means 5. Means of text conversion

Claims

1. An information processing system for structuring a plurality of unstructured text information pieces that may contain descriptions of the same or a common phenomenon, comprising: A structuring means and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to each of the structured words, which classifies the type of information of the words, and further assigning one of a plurality of labels classified according to the role or function of the words in the structure of a sentence; storing structured words included in the unstructured text information in a table including a plurality of cells corresponding to a plurality of labels, and storing the table as one data unit; The display processing means displays the structured words for each data unit, Furthermore, when the same or common structured words are included among a plurality of data units, the plurality of data units are displayed in a line such that the same or common structured words are aligned. Information processing system.

2. 2. The information processing system of claim 1, wherein the plurality of labels includes a subject label.

3. 1. An information processing system for structuring unstructured text information, comprising: A structuring means and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to the structured word, which classifies the type of information of the word, and further assigning one of a plurality of labels including a subject label to the structured word; storing, among structured words included in the unstructured text information, words to which a label other than the topic label has been assigned (non-topic words) in association with one of the words to which the topic label has been assigned (topic words), the topic word and the associated non-topic word are stored as one data unit; the display processing means displays the structured words for each data unit. Information processing system.

4. The unstructured text information includes a plurality of pieces of unstructured text information in which descriptions of the same or a common event may exist; the structuring means stores the topic words and the non-topic words associated therewith in a table including a plurality of cells corresponding to the plurality of labels, The information processing system according to claim 3, wherein the display processing means, when identical or common structured words are contained among a plurality of data units, displays the plurality of data units in an order such that the identical or common structured words are aligned.

5. 5. The information processing system according to claim 4, wherein the display processing means sorts and displays the data units in descending order of frequency of occurrence of the topic word, or displays the data units in descending order of frequency of occurrence of the topic word separately.

6. the display processing means deletes data units including those with a low frequency of occurrence with respect to the topic word from the display target, or displays the data units including those with a low frequency of occurrence separately; The information processing system according to claim 3 .

7. The unstructured text information is a part of medical document information, The plurality of labels include: (1) a subject label; (2) a value label, a rating label, a frequency label, and / or a location label; and (3) a tense label and / or a date and time label; The information processing system according to claim 1 , further comprising:

8. 7. The information processing system according to claim 1, wherein the display processing means sorts and displays data units in descending order of frequency of occurrence of words, or displays data units that include words that appear frequently in a distinguishing manner.

9. The information processing system according to any one of claims 1 to 6, wherein the display processing means rearranges and displays the order of the data units, displays the data units in a distinguished manner, or removes the data units from the display targets based on display settings set for one or more of words, information types, labels, and data units.

10. A major classification including a plurality of information types is set for at least a part of the information types, 7. The information processing system according to claim 1, wherein the display processing means displays the data units divided into the major categories.

11. 6. The information processing system according to claim 1, wherein the display processing means deletes cells with labels that do not store words from all of the tables displayed in the order.

12. The information processing system according to any one of claims 1, 2, 4 and 5, wherein when there are multiple data units in the same unstructured text information that contain the same or common structured words, the display processing means integrates them into a single data unit and displays them.

13. 1. An information processing system for structuring unstructured text information, comprising: The present invention includes a text division means, a structuring means, and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to each of the structured words, the information type of the word being classified; A method for storing structured words contained in the unstructured text information as one or more data units by associating the structured words with each other, Furthermore, if the unstructured text information includes a word indicating tense, the word is extracted and stored in association with other structured words included in the same unstructured text information; The display processing means displays the structured words for each data unit, Furthermore, when the word indicating the tense is future or past, a data unit including the word indicating the tense is displayed separately from other data units, or is deleted from the display target. Information processing system.

14. An information processing system for structuring medical document information including a plurality of pieces of unstructured text information each having a different medical examination date, comprising: A structuring means and a display processing means, The structuring means comprises: applying a structured model to the unstructured text information to extract structured words; assigning an information type to each of the structured words, the information type of the word being classified; storing structured words contained within the unstructured text information in association with one or more data units; The display processing means displays the structured words for each data unit, Furthermore, when the same or common structured words are included among a plurality of data units, the plurality of data units are displayed in a line such that the same or common structured words are aligned, or the plurality of data units in a line such that the same or common structured words are aligned are displayed in a manner that distinguishes them from other data units. Information processing system.

15. The information processing system includes a text segmentation means, The text segmentation means segments the unstructured text information into text fragments that may include a plurality of semantically related words; 15. The information processing system according to claim 1, wherein the structuring means applies a structuring model to the text fragment to extract structured words.

Images