JP2023063180A - Data management system, data management method, and data management program - Google Patents

Abstract

To provide a data management system which can represent a time-series relation between nodes of graph data and high value-added information such as tags of the nodes, clearly on a user interface, a data management method, and a data management program.SOLUTION: A data management system manages multiple pieces of data by using the data as nodes, and defining a relation between the nodes as edges. A group of nodes is managed as a dataset. The dataset is displayed in a matrix structure having a first system and a second system that intersects with the first system, on a user interface. The first system and the second system have a tree structure in accordance with the types of tags of the systems. The types of the tags of the first and second systems include time series and/or data types, and can be selected and modified. The systems can be rearranged in accordance with the selected type of the tag.SELECTED DRAWING: Figure 1

Description

The present invention relates to a data management system, data management method and data management program.

A data management method using graph theory is adopted in a data management system that enables information retrieval. For example, Patent Literature 1 describes a graph data generation device that generates new graph data for information retrieval based on given graph data.

Patent No. 6293335

In Patent Document 1, new graph data for searching can be automatically generated, but each data is limited to simple fixed data, and the generated graph data consists of multiple nodes and input edges between them. Despite the simple structure of graph data consisting of only output edges, it contains many loops and multiple edges. It has been difficult to express high value-added information such as chronological relationships and tags of each node on the user interface.

The present invention provides a data management system, data management method, or data management program capable of expressing time-series relationships between nodes of graph data and high value-added information such as tags of each node in an easy-to-understand user interface. intended to provide

A data management system according to an aspect of the present invention is a data management system that defines data as nodes, defines relationships between the nodes as edges, and manages a plurality of data, wherein groups of the nodes are managed as datasets. and the data set is displayed in a user interface as a matrix-like structure having a first series and a second series intersecting with the first series, the first series and the second series being tags for each series. wherein each tag type of the first series and the second series includes at least a time series and/or a data type and can be selectively changed, and each series is selected It is characterized in that it can be rearranged according to the type of tag.

According to the data management system according to an aspect of the present invention, it is possible to express high-value-added information such as time-series relationships between nodes in graph data and tags of each node in an easy-to-understand user interface.

FIG. 1 is a block diagram of a data management system according to Embodiment 1. FIG. FIG. 2 is an explanatory diagram of the relationship between graph data and reports according to the first embodiment. FIG. 3 is an explanatory diagram of map display of reports and data according to the first embodiment. FIG. 4 is an explanatory diagram of data display in a report. FIG. 5 is an explanatory diagram of the report creation operation. FIG. 6 is an explanatory diagram of a data citation operation for a report. FIG. 7 is an explanatory diagram of registration display of a new report. FIG. 8 is an explanatory diagram of the data reference display of the report. FIG. 10 is an explanatory diagram of a text search result display; FIG. 10 is an explanatory diagram of the display of data citation relationships. FIG. 11 is an explanatory diagram of the procedure for creating a report. FIG. 12 is an explanatory diagram of the data input procedure. FIG. 13 is an explanatory diagram of map display of reports and data according to the second embodiment.

Hereinafter, a data management system, a data management method, and a data management program according to embodiments of the present invention will be described with reference to the drawings. However, the embodiments shown below are examples of a data management system, a data management method, and a data management program for embodying the technical idea of the present invention, and the present invention is not limited to these. It is equally applicable to other embodiments within the scope of the claims.

[Embodiment 1]
A data management system, a data management method, and a data management program according to the first embodiment will be described with reference to FIGS. 1 to 12. FIG.

[About the system block diagram]
1 is a block diagram of a data management system according to one aspect of the present disclosure; FIG. The data management system of this embodiment can handle graph data consisting of nodes and edges that define the relationship between each node. Nodes and edges are tagged. Since the tag schema is defined according to the data type, data management can be facilitated by attaching a tag suitable for the data type. For example, the tag schema differs depending on whether the file type is a CSV file or an ING file, and for images and videos, depending on the resolution and size. The essential information of the tag includes, but is not limited to, information such as date and time, data type, and person in charge.

The central processing system 10 has a login function section 11 , a machine learning section 12 and a central processing section 13 . The central processing system 10 is also connected to user terminals 45 via the user management system 30 . The login function unit 11 communicates with an external SSO (Single Sign-On) 16, so that when a user performs a login operation from a user terminal 46 via the user management system 30, central processing is performed using the external SSO 16. All web services, cloud services, applications, etc. accessed from the system 10 can be logged into. The machine learning unit 12 can perform calculations such as classification model, vector distance, edge generation, report missing value interpolation, and the like, for example, using a learning model.

The central processing unit 13 has a user session management unit 14 and a graph data management unit 15 . The central processing unit 13 is connected to the metadata storage unit 20 and the sub-metadata storage unit 25 . The metadata storage unit 20 is a storage unit that tags and manages data groups, and is a database that has a document node table 21 and a data node table 22 . The sub-metadata storage unit 25 is a database that holds information related to each data of the data group as edge information, and has an edge table 26 . The user session management unit 14 performs management according to operation instructions from each user corresponding to each user management system 30 and management of operation history. The graph data management unit manages the tagged data and the relationship between the data as edges using the metadata stored in the metadata storage unit 20 and the sub-metadata stored in the sub-metadata storage unit 25. We have realized that "all data is connected".

The central processing unit 13 is also connected to an external cloud storage 17 . In the central processing system 10, only the tag data is stored in the metadata storage unit 20 and the sub-metadata storage unit 25, and actual data requiring a huge storage capacity, such as image data, moving image data, text data, etc., is stored in an external cloud. It is stored in the storage 17. Specifically, the central processing system 10 stores the node data in the metadata storage unit 20 and stores the edge data in the sub-metadata storage unit 25, while storing the actual data itself, such as video data, image data, text data, etc. etc. are stored in the cloud storage 17 . In addition, the amount of data processing can be further reduced by performing calculations after narrowing down to only necessary data using the filtering function. As a result, the data management system itself of this embodiment stores only tag data to reduce the amount of data to be stored. graph data, for example, 100 million or more data, high-speed calculation is realized. Therefore, the data management system of this embodiment can be constructed using a general-purpose PC. In addition, the data management system of this embodiment realizes a random walk function for digging deep into edge data, which could not be realized with conventional spreadsheet software, for example, by using a tab called "cycle". For example, for the node data narrowed down by the filter function, a node group is managed as a data set, and the data set is displayed on the user terminal 45 by setting the horizontal axis to the time series and the vertical axis to the data type. It is possible to dig deeper into the edge data from each node in the asset or data set, and trace the edges one after another, such as the development code, price information, purchaser information, etc. linked to each node.

The user management system 30 includes a user information management section 31, a map viewer 32, a file wrapper 33, a user interface 35, a BI tool 36, a report creation section 37, and the like. User management system 30 is connected to one or more user terminals 45 and data mart 40 . It is also connected to the data mart 40 and the user terminal 45 .

The user information management unit 31 manages the user's login state by passing the login information from the user to the login function unit 11 of the central processing system 10 . A map viewer 32 generates a MAP View from the graph data obtained from the central processing unit 13 . The file wrapper 33 has a function of converting data input from the user terminal 45 . The file wrapper 33 has a file uploader 34 , and the file uploader 34 processes and mines data input from the user terminal 45 to convert the data into data that can be handled by the central processing unit 13 . The user interface 35 performs input/output management of the user terminal 45 . The BI tool 36 is software that analyzes and visualizes data and uses it for management and operations. The data mart 40 extracts necessary information in the user management system from information accumulated in the central processing system 10, the metadata storage unit 20, and the sub-metadata storage unit 25. It is a database stored in an easy-to-use form, and has a user attribute management table 41 for managing user attribute information and one or more master tables 42 for managing information for each type.

The user terminal 45 can use the user interface 35 and the functions of the BI tool 36 and the report creation unit 37, so that the user himself or herself can analyze data and create reports. FIG. 1 shows a configuration having a user interface 35, a BI tool 36, and a report creation unit 37 in the user management system 30, but this embodiment is not limited to this. 36 and the report creation unit 37 may be integrated with the user interface 35, or the user interface 35 may be provided as a common function on the central processing system 10 side.

In the past, data groups were stored in a file server, so they lacked versatility and reusability. This has improved reusability and search performance. Also, by holding related information as edge information, the relationship between each data is similarly saved. Edges connect all data by holding not only physical names such as vehicle models, parts names, specification names, and values, but also additional information such as evidence and version management information for specification values. realized. A node corresponding to the actual file is written to the metadata storage unit 20, and the meta information is held as a property. Also, the sub-metadata storage unit 25 writes out sub-metadata information (evidence, knowledge, version management) as an edge, thereby linking each node. Then, the actual file itself, which has a huge amount of data, is stored in the cloud storage 17 . As a result, the data management system of the present embodiment connects each data, so that all data can be managed cross-sectionally, and the computational load is reduced.

The fill wrapper 33 of the user management system 30 processes and mines the data of each system and converts it into data that can be handled by the central processing system. created in

The central processing unit 13 of the central processing system 10 has the core algorithm of the system of this embodiment. to perform high-speed searches and genealogy management. The central processing unit 13 is also connected to the graph data analysis cloud 18, and as described later, it is possible to use the functions of the graph data analysis cloud 18 for genealogy management, display maps, and the like. As described above, the data management system of the present embodiment uses an external cloud to reduce the processing load of the system itself, and realizes high-speed calculation for a large amount of graph data.

The data mart 40 stores data necessary for the user interface 35, the BI tool 36, the report creation section 37, and the like.

The document node table 21 is a table for grouping and holding data nodes in the form of reports and documents. The data node table 22 is a table for nodes for holding various data, and the data includes various information such as graph data. The edge table 26 is a table for holding linking information between data. In this embodiment, linking information is held according to various conditions such as date, revision, and evidence.

It is also possible to centralize the user interface 35 on the central processing system 10 . File uploader 34 accepts data uploaded in a fixed format, such as CSV or TSV. The user interface 35 includes, for example, a Node Editor & viewer, which allows single addition, deletion, and property editing of nodes and edges, and can be displayed on the user terminal 45 as a graph image.

The file wrapper 33 shapes and converts the user's data format so that it can correspond to the file upload function of the central processing system 10 .

The central processing system 10 prepares various APIs. For example, Search is an API for executing simple search processing without writing queries, and QueryAPI is a general-purpose API for querying graphs. The file wrapper 33 uses the query API of the central processing system 10 to convert each data into the data format of the central processing system and register it.

The user interface 35 also has the following functions. Report maker provides screens for creating and editing reports. The Report Viewer & Search screen is a screen for searching and duplicating reports, and for outputting to the data mart. The Report maker is a screen for creating and editing reports, and the contradiction detection crawler notifies the user of missing reports. Based on the learning model trained by the machine learning unit 12, discrepancies, deficiencies, typographical errors, etc. are judged.

[Explanation of terms]
Here, the terms used in this embodiment will be explained.
(1) Documents A "document" means a unit managed by a file including not only document formats such as Word and Excel, but also data to be input/output to/from MATLAB (registered trademark). There are two types of documents: "report" and "data". "Data" is the result of measurement, simulation, prediction, etc., stored as objective facts, and may be a single image or chart, or a file containing them. The "report" is a description of the thought process and the content of the discussion, including objective considerations, based on these data.

Various tags are assigned to nodes and edges. Nodes are given tags such as creator, date and time of creation, object ID, type, and name. Tags such as creator, date and time of creation, type, data number, and comment are attached to edges. As for the types of nodes, for example, plans, test plans, actual measurement data, specification values, simulation results, models, and the like can be cited. Regarding edges, for example, types include report attachment, evidence, next process, data use, evidence, etc. For example, data numbers include codes such as 1, 2, 3, . Examples include the creation of charts from data and reports at the time of reconfirmation.

(2) Report screen Reports are generally created using Word, Excel, etc., rather than being created on a web screen. For a report created using Word, Excel, etc., the relationship between the report and data is held as the relationship between nodes and edges, making it possible to trace back to the data later. Therefore, at least the data included in the report can be registered. The web UI that displays a list of the data registered in each report enables tasks such as "searching for related reports." In addition, report files created using Word, Excel, etc. can be downloaded and viewed as needed. When a report is registered and edited on each view, node data and edge data are automatically added to and changed in the metadata storage section 20 and sub-metadata storage section 25 . When displaying a report, target subgraphs are extracted from the graph data, and data for screen drawing display is created by the graph data analysis cloud 18, for example, a graphic client (Oracle). It is also possible to display the citation relationship of the data used in the report. Furthermore, it is possible to access the data directly from the object ID.

(3) The inference engine and machine learning inference engine applies logical expressions to derive logical entailments. for example,
A => B, B => C then A => C
A logical conclusion is obtained. Using machine learning, a learning model is generated by training based on past data, and by applying this, results can be obtained without going through the logical formulas of the inference engine. For example, a learning model can provide operations such as classification models, vector distances, edge generation, report missing value interpolation, and the like.

(4) Advantages of focusing on genealogy Conventionally, when reports and data are searched and databased to improve reusability, the granularity and degree of normalization of the database differ depending on the requirements for searching. For example, database requirements differ depending on requirements such as whether aggregation is necessary or whether full-text search is sufficient. The data management data of the present embodiment enables a database of flexible relationships between reports and data. Database structuring includes full-text search (text), metadata (tags), graphs (tags and relationships), structured documents (XML/JSON), relational (tables), semantics (RDF), etc. considered.

(5) Nodes and edges Write actual files as nodes in the graph DB and retain meta information as properties. In addition, sub-meta information (evidence, knowledge, version control) is written out as an edge, which links each node. As a result, each data is connected, and all data can be managed cross-sectionally.

[File Wrapper and Report Wrapper]
"File wrapper" provides a visually easy-to-understand file structure similar to EXCEL, with the horizontal axis (horizontal system) as "time series tree structure" and the vertical axis (vertical series) as "relationship tree structure". do. Conventional graph data could automatically create edges only for fixed and simple data structures such as weather data. Weather data is simple fixed form data consisting of date, area, temperature, humidity, and the like. Automatic edge generation was only possible for standard data that could be handled as simple, so-called big data, such as weather data.

The "report wrapper" (report creating unit 37) registers "report" as a data node group. Heterogeneous data processing is required. In other words, it is necessary to be able to utilize it in a practical system in which data with various different attributes are mixed.

[About learning]
Learning algorithms that are simpler than deep learning are available for sticking edges by learning. For example, edges corresponding to nodes can be pasted by performing shortest path search learning using missing value interpolation. As another example of learning, reinforcement learning in an N-order polynomial non-Markov process can be used to suggest places where the operator has not yet entered. For example, a correction suggestion is made such as "Although the previous related file contains α as specification value data, is it okay to not include α in this file?"

[Outline of the system of this embodiment]
The system of this embodiment, for example, has a time-series tree structure for the horizontal system and a relevance tree structure for the vertical system, and has a visually easy-to-understand file structure. can be implemented more effectively than The data list is displayed in a map display (a map display consisting of horizontal and vertical series) instead of a list display, and the display is easy to understand visually by displaying the data after associating it with each other. By adding a full-text search function for attribute values, you can search for data as if you were using a familiar search engine, such as EXCEL.

The data management system of this embodiment eliminates the inconvenience of registering data and enables simple registration. The data to be registered takes into account the user's attribute value, etc., and the attribute value registration based on the default setting has been realized. Enable registration with minimal input items. The user's operation from the user terminal 45 is performed by Drag & Drop, so that anyone can operate it in an easy-to-understand manner. An automatic registration function for information such as attribute values is provided through inter-system cooperation. Graph-based (using graph theory) PaaS (Platform as a Service) can be used for map data that realizes visual representation. By using PaaS, it is possible to use the execution environment necessary for application development of the data management system of this embodiment, such as the runtime of the language necessary to execute the application code and middleware such as a database, on the cloud. can. This service also includes a document management function. As a graph data analysis cloud 18, for example, the Oracle analysis cloud can be used for data visualization and reporting. Geospatial analysis can be performed using graph analytics functions in Oracle Analytics Cloud. For example, you can calculate the distance and number of hops between two vertices.

The business-specific attribute values in the user management system 30 are developed in the Add-on function to minimize the amount of development. The data content reference function assumes a general Self Service BI tool and allows users to select an easy-to-use tool. That is, in the data management system of the present embodiment, the user himself/herself can analyze data and create reports using BI tools (Business Intelligence tools) 36 .

[Relationship of graph data on the display screen]
In the system of the present embodiment, a report creation function is added as an alternative to Excel in order to convert all report systems that were created and managed in Excel conventionally into a Web format. The user terminal can display data on the database and upload local data and content.

FIG. 2 is an explanatory diagram of the relationship between graph data and reports. In graph data, data are represented as nodes, and relationships between data are represented as edges. A plurality of related data are collectively managed as data sets, and edges are formed between the data sets. Forming an edge as a group for the data set facilitates management of the data group, and facilitates collective management of data for each simulation or each experiment, for example.
(1) The file wrapper 33 takes data from each user management system 30 into the central processing system 10, extracts the data, generates the necessary nodes and edges, and sends them to the central processing system 10 via API.
(2) Regarding creation of a new node, when a new report is created, a new report node is automatically generated.
(3) Regarding addition of content, when new content (for example, an image) created by the user is added to the report, the content is added as a node and an edge is automatically generated between the content and the report.
(4) For citing existing data, existing data obtained from other series is quoted in the report (If you drag and drop the existing data on the user terminal, the node corresponding to the report and the existing data will Edges are automatically generated.
(5) Relations represented by edges, when adding a chart based on existing data acquired from multiple data nodes to a report, for example, when two data are used in the same chart, the two Edges are automatically generated between nodes corresponding to the data.

[Mode of display on user interface]
FIG. 3 is an explanatory diagram of map display of reports and data. The horizontal series is a time-series tree, and the vertical series is a relevance tree. Attribute value narrowing conditions are displayed at the top of the screen. Only nodes with this attribute value are displayed. In the example of FIG. 3, attribute value filtering conditions are input in advance with default values assigned to user information, but the present embodiment is not limited to this. Input from the user terminal 45 is also possible.

In FIG. 3, the horizontal series is a time-series tree and the vertical series is a relevance tree. Data sets and data point tags are classified by data type, and can also be classified by the nested structure of the data set. Data sets can be managed as data chunks by forming a nested structure and using data entity chunks as datasets. Data relationships are established in units of datasets. On the other hand, if specified, it is possible to link individual data. In addition, since the data is classified by data type, it is easy to manage and search the graph data.

For example, information about test vehicles and simulation models is registered by the system administrator (the system administrator's terminal is not shown) or is registered in cooperation with each management system (not shown). It should be noted that the system manager terminal allows the test/simulation executor to select only which test vehicle (model) was used, thereby establishing a relationship between the test vehicle and the performance information. Associations between different tests/simulations are also possible by linking test vehicle and model version information as basic information.

FIG. 4 is an explanatory diagram of data display in a report. The file contains multiple pieces of data. For data with KPIs (key performance indicators), KPIs as single numerical values and their values are retained as data points. KPIs are indicators that define the degree of goal achievement, and are used to evaluate and improve specific processes and progress. In FIG. 4, by clicking on each entity of data, it is possible to break down to the data point of the data alone.

FIG. 5 is an explanatory diagram of the report creation operation. For the report creation operation, first, on the display of the user terminal 35, right-click the entity of the node for which the report is to be created, and select "create report" from the pull-down menu. From the report creation selection screen, it is also possible to select a model list, an automatic format, etc., and it is possible to select a predetermined model or use an automatically created report format. When using the automatic format, it is possible to automatically create an appropriate report format by using a learning model trained in the machine learning unit 12 . In this case, reports created in the past, data sets, data, citation relationships, and the like used therein are used for training in the machine learning unit.

FIG. 6 is an explanatory diagram of a data citation operation for a report. Clicking on the entity of the data you want to cite zooms in on the dataset you want to cite and shows the data groups contained within. A report is created by dragging and dropping data to be quoted into the report format selected in FIG. You can retrieve data from other files and paste it into your report by dragging it. When creating a report, a template for each report is automatically recommended as described above, or it can be selected from a list of templates. Furthermore, if there is an input omission or the like, the AI assists in inputting missing values. In the report part predicted to be missing by AI, the part that seems to be missing is displayed in reverse video, and data candidates to be complemented can be displayed. Although the AI is not particularly limited, for example, shortest path search learning using missing value interpolation, reinforcement learning in the N-order polynomial non-Markov definition, and the like are used. It should be noted that it is also possible to use deep learning for learning CNN using history data stored in the past as AI.

FIG. 7 is an explanatory diagram of registration display of a new report. Each data is automatically edged by the reporting operation. The report is added as a new dataset node, and this node is added as a new entity on the user terminal 45 display.

FIG. 8 is an explanatory diagram of the data reference display of the report. On the display of the user terminal 45, right-clicking a report or data entity to be analyzed and selecting "data reference" from the pull-down menu proceeds to an analysis screen for data reference. If you have already built a reference screen, go to the configured screen. On the other hand, if the reference screen has not yet been constructed, proceed to the default screen of the analysis project that loaded this data. To create and edit the reference screen, select "edit reference screen" from the pull-down menu and follow the same procedure as for creating the report described above.

FIG. 9 is an explanatory diagram of the result display of the text search. Files within all data and all reports are text searchable. A full-text search can be performed by entering a search expression and a search word in the "Search" column at the top of FIG. Depending on the search formula, conditional searches such as proximity searches and data category specified searches are also possible. A list of search results is displayed on the display of the user terminal 45 as "Text Search Result".

FIG. 10 is an explanatory diagram of the display of data citation relationships. Since edge search is possible, it is possible to display the linking relationship of each data. "Parent Node Connection Search" in the upper row is the result of searching for nodes related to the parent node one level above. "Direct Node Connection Search" in the middle row is the search result of the node (other than the parent node) that is directly connected to itself. "Browse Log Node Connection Search" in the lower part is the search result of the node browsed via the node (the node of the target data itself).

FIG. 11 is an explanatory diagram of the procedure for creating a report. By dragging and dropping existing data, you can insert the data you want to cite into the report you are creating. For example, when a parent node selected in "Parent Node Connection Search" is clicked, a list of performance information datasets associated therewith is displayed (see the upper left diagram in FIG. 11). By opening the report creation screen as described above (operation in the upper right of FIG. 11), the data of the citation source can be pasted on the data pasting position on the report by dragging and dropping (lower diagram in FIG. 11). You can compare data by dragging and dropping different data points onto the report canvas. In addition, by drawing various data points side by side, the status of various items can be checked at the same time. Using the graph drawing function, graphs can be drawn by dragging and dropping data points onto the report canvas. In this way, all data and files in all reports can be text-searched, and a report can be created using the data extracted as a result of the edge search.

FIG. 12 is an explanatory diagram of the data input procedure. When entering new data, click the "new registration button" (not shown) or select an experiment node to be linked. Attribute values can be specified from the user terminal 45 in the "attribute value selection area" (see the attribute value display portion on the left side of the upper diagram in FIG. 12), but the initial setting values are preset from the settings given to the user. It is in selected state. The "connection node" basically associates the setting value with the previous result. Similar nodes are automatically selected as connection target nodes in view of the previous registration operation. For the "registration file", for example, the user drags and drops the file to be registered from the explorer on the user terminal 45 as it is. If you want to add a selection node as a connection target node, click the "Select New Node" button as shown in the upper part of Fig. 12 to display a list of node candidates that match the conditions in the "node selector". display. From the list of candidates, select the node you want to newly connect and click the registration button (see the middle diagram in Fig. 12). Data is registered. When the data is registered, the entity is displayed on the display of the user terminal 45 and reflected in the data management system together with the edge set by the corresponding node (see the lower diagram in FIG. 12).

[Description of internal operation]
An internal operation regarding file registration in the data management system of this embodiment will be described. File registration is registered on the assumption that it is associated with the last registered node unless otherwise specified. If you select a file or dataset, it will be generated as a node linked to that node. Nodes with the same attributes are automatically connected on a chronological basis. It is also possible to discriminate the inside of the data, disassemble the information related to the inside, and generate individual nodes. Attributes possessed by the user are basically retained as they are. Files that hold data basically have pointers that indicate internally divided data for each individual piece of data (column), and manage them by graphing them as data points. Being able to operate on a per-datapoint basis means that individual datapoints can be compared and presented in reports. In addition, the Parse function (data structure conversion function) and file extraction function are provided in order to examine the contents of the file data and access individual columns.

[Association method (1) Master node connection method]
Among data set nodes, information on master nodes (test vehicles/simulation models) is strictly versioned at the time of data registration. Only at the start of the project, there will be no connected nodes. Other dataset nodes are always registered in association with some master node. As a result, information is always connected through the masternode.

[Association method (2) Similar attribute linking]
A search is made for items that have a certain attribute value equal or close to equal, and the degree of similarity is scored. Automatically concatenate high score values.
<Example of scoring algorithm>
Attribute values are classified into four categories a to d below, and after judging whether or not the value matches with a), c), and d), b) is scored (counting the number of matches) and the edge is determined. Generate automatically.
a) "Must be exactly the same" (important string attribute) Equal to hit b) "Not necessarily the same" (unimportant string attribute) Count the number of matches c) "Within range can be considered identical if
d) Is it within range?

[Association method (3) Linking by data reference history]
Keep a log of all the times a user has moved from one node to another and viewed data. Record as a log which node flew to which node. A new edge is registered when it is recorded more than a certain number of times.

[Association method (4) Manual combination]
Based on the edges drawn by (1) to (3) above, it is also possible to manually browse the Graph space and draw edges where necessary based on human judgment. In addition, when setting the edge manually in this way, a browsing function, which is normally available only to the administrator, is used.

[Examples of application to specific cases]
The data that should be obtained through system linkage includes the information of the person in charge in the company and the project in charge, the project list, the automatic registration of the data set to be registered, the image reading of the numerical values in the photograph, and the identification of the data content. mentioned.

Information necessary for presetting, such as personnel information and assigned project information, can be automatically retrieved from a peripheral system (not shown) by the file wrapper 33 . For customer-specific matters, it is necessary to set up the user management system in advance. User information definitions are held, for example, in a general RDB format. However, for example, for projects in charge, business areas in charge/performance in charge, final experiments/simulations, final reports, etc., a JSON format attribute function is provided so that various attribute information can be added.

A full-text search is possible because a full-text search index is assigned to the property values stored in the node table and edge table.

[Embodiment 2]
A data management system, a data management method, and a data management program according to the second embodiment will be described with reference to FIG. FIG. 13 is an explanatory diagram of map display of reports and data according to the present embodiment. The same numbers are given to the same configurations as those in FIGS. 1 to 12, and the description thereof is omitted.

As a basis for consolidation, it is also possible to visually display the version of the generated report. Arrange chronological data on the horizontal axis to clarify the version. On the vertical axis, measurement/simulation types such as NV, fuel efficiency, and steering stability are arranged as related items. This clarifies the citation relationship between the items on the vertical axis, making it easy to grasp the version upgrade status.

In this embodiment, an example of the vertical axis is shown, but arbitrary map display is possible by freely selecting the type of tag for both the vertical axis and the horizontal axis so that the user can easily view the map. At least, it is possible to select the types of tags such as time series and data types. In addition, various tags such as person in charge, product code, version information, experiment conditions, and simulation conditions can be selected as the types of tags for each axis.

The data management system, data management method, and data management program according to the embodiments of the present invention have been described above. These are examples of programs, and the present invention is not limited to these, and can be equally applied to other embodiments. Modifications and combinations of aspects of each embodiment are possible.

10... Central processing system 11... Login function unit 12... Machine learning unit 13... Central processing unit 14... User session management unit 15... Graph data management unit 16... External SSO
17... Cloud storage 18... Graph data analysis cloud 20... Metadata storage unit 21... Document node table 22... Data node table 25... Sub metadata 26... Edge table 30... User management system 31... User information management unit 32... Map viewer 33... File wrapper 34... File uploader 35... User interface 36... BI tool 40... Data mart 41... User attribute management table 42... Master table 45... User terminal

A data management system according to an aspect of the present invention is a data management system that defines data as nodes, defines relationships between the nodes as edges, and manages a plurality of data, wherein groups of the nodes are managed as datasets. and the data set is displayed in a user interface as a matrix-like structure having a first series and a second series intersecting with the first series, the first series and the second series being tags for each series. wherein each tag type of the first series and the second series includes at least a time series and/or a data type, and can be changed according to a selection input from a user. , each series can be rearranged according to the type of the selected tag.

The central processing system 10 has a login function section 11 , a machine learning section 12 and a central processing section 13 . The central processing system 10 is also connected to user terminals 45 via the user management system 30 . The login function unit 11 communicates with an external SSO (Single Sign-On) 16, so that when a user performs a login operation from the user terminal 45 via the user management system 30, central processing is performed using the external SSO 16. All web services, cloud services, applications, etc. accessed from the system 10 can be logged into. The machine learning unit 12 can perform calculations such as classification model, vector distance, edge generation, report missing value interpolation, and the like, for example, using a learning model.

The central processing unit 13 has a user session management unit 14 and a graph data management unit 15 . The central processing unit 13 is connected to the metadata storage unit 20 and the sub-metadata storage unit 25 . The metadata storage unit 20 is a storage unit that tags and manages data groups, and is a database that has a document node table 21 and a data node table 22 . The sub-metadata storage unit 25 is a database that holds information related to each data of the data group as edge information, and has an edge table 26 . The user session management unit 14 performs management according to operation instructions from each user corresponding to each user management system 30 and management of operation history. The graph data management unit 15 uses the metadata stored in the metadata storage unit 20 and the sub-metadata stored in the sub-metadata storage unit 25 to define the tagged data and the relationship between the data as edges. It manages and realizes that "all data is connected".

The central processing unit 13 is also connected to an external cloud storage 17 . In the central processing system 10, only the tag data is stored in the metadata storage unit 20 and the sub-metadata storage unit 25, and actual data requiring a huge storage capacity, such as image data, moving image data, text data, etc., is stored in an external cloud. It is stored in the storage 17. Specifically, the central processing system 10 stores the node data in the metadata storage unit 20 and the edge data in the sub-metadata storage unit 25, while storing the actual data itself, such as video data, image data, text data. etc. are stored in the cloud storage 17 . In addition, the amount of data processing can be further reduced by narrowing down to only necessary data by the filtering function and performing the calculation. As a result, the data management system itself of this embodiment stores only tag data to reduce the amount of data to be stored. graph data, for example, 100 million or more data, high-speed calculation is realized. Therefore, the data management system of this embodiment can be constructed using a general-purpose PC. In addition, the data management system of this embodiment realizes a random walk function for digging deep into edge data, which could not be realized with conventional spreadsheet software, for example, by using a tab called "cycle". For example, for the node data narrowed down by the filter function, a node group is managed as a data set, and the data set is displayed on the user terminal 45 by setting the horizontal axis to the time series and the vertical axis to the data type. It is possible to dig deeper into the edge data from each node in the asset or data set, and trace the edges one after another, such as the development code, price information, purchaser information, etc. linked to each node.

The user information management unit 31 manages the user's login state by passing the login information from the user to the login function unit 11 of the central processing system 10 . A map viewer 32 generates a MAP View from the graph data obtained from the central processing unit 13 . The file wrapper 33 has a function of converting data input from the user terminal 45 . The file wrapper 33 has a file uploader 34 , and the file uploader 34 processes and mines data input from the user terminal 45 to convert the data into data that can be handled by the central processing unit 13 . The user interface 35 performs input/output management of the user terminal 45 . The BI tool 36 is software that analyzes and visualizes data and uses it for management and operations. The data mart 40 extracts necessary information in the user management system 30 from the information accumulated in the central processing system 10, the metadata storage unit 20, and the sub-metadata storage unit 25, aggregates, and manages It is a database stored in an easy-to-use form, and has a user attribute management table 41 for managing user attribute information and one or more master tables 42 for managing information for each type.

The file wrapper 33 of the user management system 30 processes and mines the data of each system and converts it into data that can be handled by the central processing system 10. In order to correspond to customer attribute data, each user management system 30 created separately.

The central processing system 10 prepares various APIs. For example, Search is an API for executing simple search processing without writing queries, and QueryAPI is a general-purpose API for querying graphs. The file wrapper 33 uses the query API of the central processing system 10 to convert each data into the data format of the central processing system 10 and register it.

The user interface 35 also has the following functions. Report maker provides screens for creating and editing reports. The Report Viewer & Search screen is a screen for searching and duplicating reports, and for outputting to the data mart 40 and the like. The Report maker is a screen for creating and editing reports, and the contradiction detection crawler notifies the user of missing reports. Based on the learning model trained by the machine learning unit 12, discrepancies, deficiencies, typographical errors, etc. are determined.

[File Wrapper and Report Wrapper]
"File wrapper" ( 33) is a visually easy-to-understand file like EXCEL with the horizontal axis (horizontal system) as "time series tree structure" and the vertical axis (vertical series) as "relationship tree structure". Provide structure. Conventional graph data could automatically create edges only for fixed and simple data structures such as weather data. Weather data is simple fixed form data consisting of date, area, temperature, humidity, and the like. Automatic edge generation was only possible for standard data that could be handled as simple, so-called big data, such as weather data.

[Relationship of graph data on the display screen]
In the system of the present embodiment, a report creation function is added as an alternative to Excel in order to convert all report systems that were created and managed in Excel conventionally into a Web format. The user terminal 45 can display data on the database and upload local data and content.

FIG. 2 is an explanatory diagram of the relationship between graph data and reports. In graph data, data are represented as nodes, and relationships between data are represented as edges. A plurality of related data are collectively managed as data sets, and edges are formed between the data sets. Forming an edge as a group for the data set facilitates management of the data group, and facilitates collective management of data for each simulation or each experiment, for example.
(1) The file wrapper 33 takes data from each user management system 30 into the central processing system 10, extracts the data, generates the necessary nodes and edges, and sends them to the central processing system 10 via API.
(2) Regarding creation of a new node, when a new report is created, a new report node is automatically generated.
(3) Regarding addition of content, when new content (for example, an image) created by the user is added to the report, the content is added as a node and an edge is automatically generated between the content and the report.
(4) As for citing existing data, existing data acquired from other series is cited in the report (when existing data is dragged and dropped on the user terminal 45 , the node corresponding to the report and the existing data is , edges are automatically generated.
(5) Relations represented by edges, when adding a chart based on existing data acquired from multiple data nodes to a report, for example, when two data are used in the same chart, the two Edges are automatically generated between nodes corresponding to the data.

FIG. 5 is an explanatory diagram of the report creation operation. For the report creation operation, first, on the display of the user terminal 45 , right-click the entity of the node for which the report is to be created, and select "create report" from the pull-down menu. From the report creation selection screen, it is also possible to select a model list, an automatic format, etc., and it is possible to select a predetermined model or use an automatically created report format. When using the automatic format, it is possible to automatically create an appropriate report format by using a learning model trained in the machine learning unit 12 . In this case, reports created in the past, data sets, data, citation relationships, and the like used therein are used for training in the machine learning unit 12 .

Information necessary for presetting, such as personnel information and assigned project information, can be automatically retrieved from a peripheral system (not shown) by the file wrapper 33 . For customer-specific matters, the user management system 30 must be set in advance. User information definitions are held, for example, in a general RDB format. However, for example, for projects in charge, business areas in charge/performance in charge, final experiments/simulations, final reports, etc., a JSON format attribute function is provided so that various attribute information can be added.

Claims (5)

A data management system that defines data as nodes and relationships between the nodes as edges, and manages a plurality of data,
managing the group of nodes as a dataset;
displaying the data set in a user interface as a matrix-like structure having a first series and a second series intersecting the first series;
The first series and the second series are tree structures according to the types of tags in each series,
the types of tags of each of the first series and the second series include at least a time series and/or a data type, and are selectable and changeable;
A data management system, wherein each series can be rearranged according to the type of tag selected. The edge is given in units of the data set,
The nodes and edges are tagged,
the node tag schema is defined according to a data type; and/or
2. The data management system according to claim 1, wherein a classification by type and/or a classification having a nest structure are defined as tags of the nodes. said dataset is represented as an entity in a user interface;
3. The data management system of claim 1 or 2, wherein the data set is manipulated using at least one of the following functional groups by manipulation of the user interface.
displaying or manipulating data belonging to said dataset;
downloading data belonging to said dataset;
viewing, creating or editing a report of said dataset;
reading, creating or editing a reporting format for said data set;
adding data to said dataset;
search for reports, datasets or data;
searching for said node;
referencing or citing data from other datasets;
registration of new data,
Operation of citing new data or data of other datasets by drag and drop,
setting tags for said edges;
switching display of nodes of the data set;
display switching of lower nodes of the data set;
management of KPIs held by the data;
searching, displaying or analyzing relevance of said dataset or said nodes;
Setting of display filter conditions, selection operation of pull-down menu,
selecting the entity or selecting an operation menu;
breakdown or selection of data points of data units from said entity;
login function,
the ability to create reports on said dataset;
A function to quote data of other nodes or new data, or
Missing or erroneous report correction assistance function,
A group of functions consisting of A data management method for managing a plurality of data by defining data as nodes and relationships between the nodes as edges,
a management step of managing the group of nodes as a dataset;
The data set is provided in a user interface as a matrix having a first series, which is a tree structure corresponding to the type of tag, and a second series, which intersects with the first series and is a tree structure corresponding to the type of tag. displaying as a structure;
a step in which the types of tags of each of the first series and the second series receive a selection change;
each series sorting according to the type of tag selected;
with
A data management method, wherein the types of tags for each of the first series and the second series include at least a time series and/or a data type. A data management program for executing the data management method according to claim 4 by a computer.

Images