Simulation information extraction device

Abstract

The present invention comprises: a collection part (11) that collects a book related to a ground phenomenon; a conversion processing part (12) that converts sentences of the book into text; and a construction part (14) that constructs, by using the text as an input and using a base model, a machine learning model for extracting simulation information for use in simulation of a ground phenomenon model. The present invention further comprises an extraction part (32) that inputs a received keyword into the machine learning model and extracts the simulation information from the book related to the ground phenomenon.

Description

Simulation Information Extraction Device

[0001] The present disclosure relates to a simulation information extraction device.

[0002] Simulations are being conducted to elucidate terrestrial phenomena such as meteorological changes, changes in natural phenomena, and ecosystem changes. For example, research is being conducted to analyze meteorological phenomena occurring on Earth by conducting simulations using meteorological models (terrestrial phenomenon models) such as existing numerical weather prediction models and atmosphere-ocean circulation models. In order for users new to the field of analyzing such meteorological phenomena to conduct simulations using a meteorological model, they must install the meteorological model and learn how to operate it, which can require a long learning process.

[0003] Furthermore, the data format and setting conditions required for input differ depending on the target to be clarified and the weather model used, and the knowledge required to prepare these is also required. This presents a major barrier for new users.

[0004] "Academia.edu," "ResearchGate," and "J-STAGE," which are disclosed in Non-Patent Documents 1 to 3, provide a service in which a user inputs a search keyword and papers and manuals (hereinafter collectively referred to as "books") containing words related to the search keyword are presented. Non-Patent Document 1 also provides a service that presents papers with content related to the search keyword in addition to the search keyword. Users refer to the presented papers to extract various conditions and setting parameters (hereinafter collectively referred to as "simulation information") used in various weather model simulations, and learn how to operate and install the weather model.

[0005] Academia.edu "https: / / www.academia.edu / " ResearchGate "https: / / www.researchgate.net / " J-STAGE "https: / / www.jstage.jst.go.jp / browse / -char / ja"

[0006] However, the above-mentioned Non-Patent Documents 1 to 3 merely present books containing terms related to the search keywords entered by the user. Therefore, in order to analyze the simulation information (conditions, setting parameters, etc.) used in the simulation of meteorological models such as numerical weather prediction models and atmospheric and ocean circulation models, the user must read each of the presented books. Furthermore, even if the main simulation information used in the simulation of meteorological models is described in the books, it is difficult to understand all of this information.

[0007] The present disclosure has been made in consideration of the above circumstances, and its purpose is to provide a simulation information extraction device that is capable of extracting simulation information to be used in simulating a ground phenomenon model in response to input of a search keyword.

[0008] A simulation information extraction device according to one aspect of the present disclosure includes a collection unit that collects books related to terrestrial phenomena, a conversion processing unit that converts the text in the books into text, a construction unit that uses the text as input and constructs a machine learning model using a base model to extract simulation information to be used in simulating a terrestrial phenomenon model, and an extraction unit that inputs received keywords into the machine learning model and extracts the simulation information from books related to terrestrial phenomena.

[0009] According to the present disclosure, it is possible to present simulation information for a terrestrial phenomenon model in response to the input of a search keyword.

[0010] FIG. 1 is a block diagram showing the configuration of a simulation information extraction device according to an embodiment. FIG. 2 is a flowchart showing the procedure for constructing a machine learning model for extracting simulation information. FIG. 3 is a flowchart showing the procedure for inputting keywords into the machine learning model and extracting simulation information. FIG. 4 is an explanatory diagram showing an example of a screen displayed on the display unit when a user inputs "WRF, 3dvar, parameter" into the input unit. FIG. 5 is a block diagram showing the hardware configuration of this embodiment.

[0011] Hereinafter, an embodiment will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a simulation information extraction device according to an embodiment. The simulation information extraction device according to this embodiment extracts simulation information to be used in a simulation of a ground phenomenon model in response to a user input of keywords related to ground phenomena such as meteorological fluctuations, fluctuations in natural phenomena, and ecosystem fluctuations. In the following, meteorological fluctuations will be described as an example of a ground phenomenon.

[0012] As shown in FIG. 1, the simulation information extraction device 100 includes a learning model generation unit 1, a storage unit 2, a response unit 3, an input unit 4, and a display unit 5.

[0013] The learning model generation unit 1 includes a collection unit 11, a conversion processing unit 12, a first preprocessing unit 13, and a construction unit 14. The memory unit 2 includes an information storage unit 21, a conversion data accumulation unit 22, and a learning model memory unit 23.

[0014] The collection unit 11 collects meteorological papers (books) posted on the web, and manuals (books) of models used in meteorological analysis, such as numerical weather prediction models and atmosphere-ocean circulation models (hereinafter referred to as "weather models"). Papers and manuals are examples of "books." That is, the collection unit 11 collects meteorological books. Meteorological models are an example of ground phenomenon models.

[0015] Web articles exist in various formats, such as data created using a word processor or other document creation application, data in Portable Document Format (PDF), image data, etc. In addition to the above data, manuals are also displayed in HyperText Markup Language (HTML) format.

[0016] The collection unit 11 stores files containing papers and manuals, and URLs (Uniform Resource Locators) where the manuals are posted, in the information storage unit 21. In addition, for papers, by cooperating with a service that provides these, only the posted URLs may be stored in the information storage unit 21 without directly storing the files.

[0017] The conversion processing unit 12 reads various data and files stored in the information storage unit 21. The conversion processing unit 12 executes a process of converting various book sentences created in various formats into text. That is, the conversion processing unit 12 converts book sentences into text. Data created with a word processing application and data in PDF format can be converted into text. Image files can also be converted into text using a character recognition application. HTML files can also be converted into text using an application such as "html-to-text" (https: / / rapids-tech.com / react-html-to-text / ). The conversion processing unit 12 stores these converted texts in the conversion data accumulation unit 22.

[0018] The first preprocessing unit 13 reads out text stored in the conversion data storage unit 22. The first preprocessing unit 13 performs preprocessing on the read out text. Specifically, the first preprocessing unit 13 performs text cleaning such as deleting unnecessary characters, tokenization to divide sentences into words and phrases, deletion of characters that do not affect the sentence itself such as " " and "", and stemming and lemmatization to convert words into their basic forms (see, for example, "https: / / seonorth.ca / ja / nlp / stemming-and-lemmatization / "). In other words, the first preprocessing unit 13 performs preprocessing on the text. The first preprocessing unit 13 stores the preprocessed text in the conversion data storage unit 22.

[0019] The construction unit 14 reads the preprocessed text from the conversion data storage unit 22. Based on the read text, the construction unit 14 uses a base model such as a neural network or a large-scale language model to construct a machine learning model for extracting various conditions and setting parameters (hereinafter collectively referred to as "simulation information") to be used in simulating a weather model. The construction unit 14 stores the constructed machine learning model in the learning model storage unit 23. In other words, the learning model storage unit 23 stores the machine learning model. The machine learning model constructed by the construction unit 14 does not initially store information related to books such as papers and manuals. The neural network and the large-scale language model are examples of base models.

[0020] By executing the processing using the above-mentioned base model, simulation information such as setting parameters contained in papers and manuals is accumulated in the machine learning model. The construction unit 14 updates the machine learning model as appropriate when the text to be processed is changed, such as when a paper is added or a manual is supplemented or corrected. That is, the construction unit 14 uses the text preprocessed by the first preprocessing unit 13 to construct a machine learning model for extracting simulation information to be used in simulating a weather model using a base model such as a neural network or a large-scale language model.

[0021] The input unit 4 is, for example, a keyboard, a mouse, or a touch panel. By operating the input unit 4, the user can input desired keywords related to the weather model or conditions that the user uses to the response unit 3. That is, the input unit 4 accepts the input of keywords by the user.

[0022] The response unit 3 includes a second preprocessing unit 31 , an extraction unit 32 , and a formatting unit 33 .

[0023] The second preprocessing unit 31 converts the keyword input via the input unit 4 into text if it is not text. The second preprocessing unit 31 cleans the text input via the input unit 4, tokenizes it, deletes characters that do not affect the sentence itself, stems it to convert it into its basic form, and converts it into remnant matter. The second preprocessing unit 31 performs preprocessing on the keyword text input via the input unit 4.

[0024] The extraction unit 32 inputs words contained in the text processed by the second preprocessing unit 31 into the machine learning model stored in the learning model storage unit 23. The extraction unit 32 extracts the meteorological model output by inputting the words into the machine learning model, as well as information, conditions, setting parameters, and their contents of papers targeting the meteorological model.

[0025] As a specific example, the extraction unit 32 extracts descriptions of setting parameters and conditions from a manual for a meteorological model described in a paper. For example, Weather Research & Forecasting (WRF) discloses a next-generation mesoscale numerical weather prediction system developed for both atmospheric research and meteorological forecasting. In the case of WRF, the parameters in "Namelist Variables" are relevant (see "https: / / www2.mmm.ucar.edu / wrf / users / wrf_users_guide / build / html / namelist_variables.html").

[0026] The WRF manual is available at https: / / www2.mmm.ucar.edu / wrf / users / wrf_users_guide / build / html / index.html. The contents of the namelist configuration file are available at https: / / www2.mmm.ucar.edu / wrf / users / wrf_users_guide / build / html / namelist_variables.html.

[0027] Furthermore, the extraction unit 32 classifies the collected papers into the multiple weather models that they use, and extracts sections related to the implementation of simulations using the weather models, such as the conditions and parameters of each weather model. In this case, it is preferable to extract paragraphs containing words rather than individual words. It is also possible to specify specific parameter names for each weather model. That is, the extraction unit 32 inputs the keywords preprocessed by the second preprocessing unit 31 into a machine learning model to extract simulation information from books related to weather.

[0028] The formatting unit 33 formats the data extracted by the extraction unit 32 into a format suitable for providing to the user, and outputs the format to the display unit 5. That is, the formatting unit 33 formats the various conditions and setting parameters extracted by the extraction unit 32 into a format suitable for presentation to the user.

[0029] Furthermore, when there are multiple weather models, the formatting unit 33 compiles the paper name, URL, conditions, and described parameter settings for each weather model in accordance with the parameter list included in the manual. For example, when a user refers to a paper search site and inputs the weather model and related keywords they want to use through the input unit 4, a list of conditions and setting parameters for various weather models is output to the display unit 5 along with the paper information and presented to the user.

[0030] The display unit 5 displays the data output from the formatting unit 33 on a display or the like.

[0031] Next, the operation of the simulation information extraction device 100 according to this embodiment will be described with reference to the flowcharts shown in FIGS.

[0032] Fig. 2 is a flowchart showing a procedure for constructing a machine learning model for extracting simulation information. Fig. 3 is a flowchart showing a procedure for inputting keywords into the machine learning model and extracting simulation information. In step S11 shown in Fig. 2, the collection unit 11 accesses a paper search site and collects papers and manuals.

[0033] In step S12, the conversion processing unit 12 converts sentences contained in books such as papers and manuals into text. As described above, data created by a word processing application and sentences contained in PDF files can be converted into text. Image files can also be converted into text using a character recognition application. HTML files can also be converted into text using an application such as "html-to-text." The conversion processing unit 12 stores the converted text in the converted data storage unit 22.

[0034] In step S13, the first preprocessing unit 13 performs preprocessing on the text converted by the conversion processing unit 12. Specifically, as described above, the first preprocessing unit 13 performs text cleaning, tokenization to divide sentences into words and phrases, deletion of characters that do not affect the sentence itself, stemming, and remattering. The first preprocessing unit 13 stores the preprocessed text in the conversion data storage unit 22.

[0035] In step S14, the construction unit 14 constructs a machine learning model using a base model such as a neural network or a large-scale language model based on the text preprocessed by the first preprocessing unit 13. The construction unit 14 stores the constructed machine learning model in the learning model storage unit 23. In this way, a machine learning model for extracting simulation information to be used in simulating a weather model can be constructed.

[0036] Next, a process for extracting simulation information to be used in a weather model simulation using the machine learning model constructed by the process shown in FIG. 2 will be described with reference to the flowchart shown in FIG.

[0037] 3, the input unit 4 accepts a keyword input by the user. The keyword input by the input unit 4 is input to the second preprocessing unit 31 of the response unit 3. The keyword is input as text. If the keyword is not text, the second preprocessing unit 31 converts it into text.

[0038] In step S32, the second preprocessing unit 31 performs text preprocessing. Specifically, the second preprocessing unit 31 cleans the text input via the input unit 4, tokenizes it, deletes characters that do not affect the sentence itself, stems it to convert it to its base form, and converts it into rematter.

[0039] In step S33, the extraction unit 32 inputs the text output by the second preprocessing unit 31 into the machine learning model stored in the learning model storage unit 23. As a result, machine learning is performed using the machine learning model. The extraction unit 32 extracts simulation information (various conditions and setting parameters) output from the machine learning model.

[0040] In step S34, the formatting unit 33 formats the simulation information extracted by the extraction unit 32 into a format that can be provided to the user.

[0041] In step S35, the shaping unit 33 displays the shaped simulation information on the display unit 5. As a result, the user can recognize the simulation information.

[0042] As a result, the results shown in Figure 4 are obtained. Figure 4 is an explanatory diagram showing an example of a screen displayed on the display unit 5 when the user inputs "WRF, 3dvar, parameter" into the input unit. In the example shown in Figure 4, three papers containing the keywords "WRF" (symbol a), "3dvar" (symbol b), and "parameter" (symbol c) are presented, and their URLs are listed. By looking at this display, the user can recognize the various conditions and setting parameters to be input into the meteorological model.

[0043] As described above, the simulation information extraction device 100 according to this embodiment includes a collection unit 11 that collects books related to ground phenomena (e.g., meteorological changes), a conversion processing unit 12 that converts the sentences in the books into text, a construction unit 14 that receives the text as input and uses a base model to construct a machine learning model for extracting simulation information to be used in simulating a ground phenomenon model, and an extraction unit 32 that inputs received keywords into the machine learning model and extracts simulation information from books related to ground phenomena.

[0044] According to this embodiment, a user can obtain detailed simulation information for weather models constructed by past researchers by specifying keywords related to the simulation they wish to perform and a weather model for elucidating meteorological phenomena on Earth. This makes it possible for even users with little experience using weather models to easily perform a simulation using a weather model in a short amount of time.

[0045] Furthermore, even experienced users can set simulation information such as setting parameters and various conditions related to the weather phenomenon they wish to verify all at once without having to look at each case in related papers, making it possible to use the system to confirm the validity of the simulation information.

[0046] Furthermore, since the books collected by the collection unit 11 include papers and manuals, it is possible to build a machine learning model suited to the collected papers and manuals. Furthermore, since the simulation information includes various conditions and setting parameters contained in the text of the papers and manuals, it is possible to build a highly accurate machine learning model that is suited to the various conditions and setting parameters.

[0047] Furthermore, the simulation information extraction device 100 according to this embodiment has the function of extracting and presenting information necessary to carry out simulations of meteorological models, such as numerical weather prediction models and atmospheric-ocean circulation models, from meteorological-related papers, i.e., it is possible to summarize and output the papers.

[0048] In addition, the simulation information extraction device according to the present disclosure can be applied to various models other than meteorological models, such as natural phenomenon models and ecosystem models, which perform simulations to clarify phenomena that may occur on the ground.

[0049] The simulation information extraction device 100 of the present embodiment described above can be, for example, a general-purpose computer system including a CPU (Central Processing Unit, processor) 901, a memory 902, a storage 903 (HDD: Hard Disk Drive, SSD: Solid State Drive), a communication device 904, an input device 905, and an output device 906, as shown in Figure 5. The memory 902 and the storage 903 are storage devices. In this computer system, the CPU 901 executes a predetermined program loaded on the memory 902, thereby realizing each function of the simulation information extraction device 100.

[0050] The simulation information extraction device 100 may be implemented on one computer or on multiple computers. Furthermore, the simulation information extraction device 100 may be a virtual machine implemented on a computer.

[0051] The program for the simulation information extraction device 100 can be stored in a computer-readable recording medium such as a HDD, SSD, USB (Universal Serial Bus) memory, CD (Compact Disc), DVD (Digital Versatile Disc), or can be distributed via a network. The computer-readable recording medium is, for example, a non-transitory recording medium.

[0052] The present disclosure is not limited to the above-described embodiments, and various modifications are possible within the scope of the gist of the present disclosure.

[0053] REFERENCE SIGNS LIST 1 Learning model generation unit 2 Storage unit 3 Response unit 4 Input unit 5 Display unit 11 Collection unit 12 Conversion processing unit 13 First pre-processing unit 14 Construction unit 21 Information storage unit 22 Conversion data accumulation unit 23 Learning model storage unit 31 Second pre-processing unit 32 Extraction unit 33 Shaping unit 100 Simulation information extraction device

Claims

1. A simulation information extraction device comprising: a collection unit that collects books related to terrestrial phenomena; a conversion processing unit that converts the sentences in the books into text; a construction unit that uses the text as input and constructs a machine learning model using a base model to extract simulation information to be used in simulating a terrestrial phenomenon model; and an extraction unit that inputs received keywords into the machine learning model and extracts the simulation information from books related to terrestrial phenomena.

2. The simulation information extraction device according to claim 1, further comprising: a first pre-processing unit that performs pre-processing on the text converted by the conversion processing unit; and a second pre-processing unit that performs pre-processing on input keywords, wherein the construction unit constructs the machine learning model using the text pre-processed by the first pre-processing unit, and the extraction unit extracts the simulation information using the keywords pre-processed by the second pre-processing unit.

3. The simulation information extraction device according to claim 2, wherein the first preprocessing unit and the second preprocessing unit perform at least one of the following: text cleaning, sentence tokenization, deletion of characters that do not affect the sentence itself, word stemming, and word remattering.

4. The simulation information extraction device according to claim 1, further comprising a formatting unit that formats the various conditions and setting parameters extracted by said extraction unit into a format that can be presented to a user.

Images