JP2007102717A - Bayesian network configuration support apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a Bayesian network configuration support apparatus that facilitates configuring of a Bayesian network. <P>SOLUTION: A label object named "expression" that is to be a candidate for a node name or for a domain name is defined and displayed in a conceptual map editor. A user defines the relation between labels by performing a mouse operation to the label objects arranged on the conceptual map editor, and creates node names and domain names. Here, the "expression" is created by text mining, by using a document and a dictionary ("expression" creation rule) specified by the user. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an apparatus for supporting the construction of a Bayesian network model.

In order to construct an expert system, a Bayesian network has been used (see, for example, Patent Documents 1 and 2). A Bayesian network is a probability model expressed in a graph structure in which random variables are represented by nodes and linked between variables that have dependent relationships such as causality and correlation. This means that the path following this link is represented by an acyclic directed graph in which the path does not circulate. A Bayesian network is composed of such a graph structure and a set of conditional probabilities assigned to each node. In the case of a discrete variable, the conditional probability related to the parent node of the child node is represented by a “Conditional Probability Table” in which conditional probabilities in all states are arranged.
JP 2002-202890 A JP 2001-117776 A

By the way, creating a Bayesian network (hereinafter abbreviated as BN) requires a lot of labor and a long time, such as hearings from domain experts. The person in charge of creating the BN must be familiar with the knowledge of the BN, and it is very difficult for a domain expert to create the BN alone.
This invention is made | formed by the said situation, The objective is to provide the Bayesian network construction assistance apparatus which can produce a Bayesian network easily.

  In order to achieve the above object, according to one aspect of the present invention, there is provided a Bayesian network construction support apparatus that supports editing of nodes and their domain constituting a Bayesian network, and a plurality of expressions related to a problem area are represented by: An expression database that stores each expression associated with each other in association with each other, and displays the expression as a label object on a GUI (Graphical User Interface) screen, and when any of the displayed label objects is selected, When a display control means for reading a label object related to the selected label object from the expression database and displaying it on the GUI screen and a plurality of label objects on the GUI screen are designated as mutual relations, Configure a Bayesian network based on the relationship on the GUI screen. Bayesian network configuration supporting apparatus, characterized by comprising a construction unit that is provided.

  By taking such means, a GUI environment for constructing a Bayesian network is provided. Then, the user generates a label object whose name is “expression” that is a candidate for the node name or domain name on the GUI screen, and the relationship between the labels (causal relationship or proposition) is displayed on the label object arranged on the screen. Node names and domain names can be created by defining (Relation) with the mouse. When the operator specifies two types of relationships (causal relationship between nodes or relationship between nodes and domain) with a GUI operation for a set of label objects on the GUI screen, the burden of thinking work of the worker As well as the work time.

  Further, in creating the expression database, “expression” is generated by text mining using a document and a dictionary (so-called “expression” creation rule) designated by the user. As a result, highly correlated expressions can be easily made into a database.

  According to the present invention, it is possible to provide a Bayesian network construction support apparatus that can easily create a Bayesian network.

  FIG. 1 is a functional block diagram showing a Bayesian network (hereinafter referred to as BN) construction support apparatus according to the present invention. The BN construction support apparatus 115 includes an input unit 108 such as a mouse and a keyboard, a storage medium 110, an output unit 112 such as a display, a storage unit 114, and a processor 116. Among these, the storage unit 114 includes an internal data holding unit 117 in a predetermined storage area. The internal data holding unit 117 is provided with an object buffer 118, an expression buffer 120, and a correlation buffer 122 between expressions.

  The storage medium 110 is provided with a BN construction support program 100, a conceptual map data storage unit 104, an expression correlation data storage unit 105, a document data storage unit 106, and a dictionary data storage unit 107. Each of the BN construction support programs 100 includes a concept map generation unit 101, an expression / correlation degree generation unit 102, and a screen control unit 103 as program blocks. Further, the expression / correlation level generation unit 102 includes a text mining unit 102a and a correlation level calculation unit 102b.

  In the above configuration, the concept map generation unit 101 generates a label object whose name is “expression” that is a candidate for a Bayesian network node name and domain name on a screen called a concept map. The range of states that a certain node can take is called the domain of that node. Hereinafter, a concept map creation operation performed by a user using a GUI (Graphical User Interface) screen will be described. FIG. 2 is a flowchart showing the procedure.

  In FIG. 2, when the BN construction support program 100 is activated, the screen control unit 103 displays a conceptual map editor (step S201). FIG. 3 shows the initial screen. . Next, the expression correlation data is read from the expression correlation data storage unit 105, and an expression list editor as shown in FIG. 4 is displayed in a separate window (step S203). A file menu 301, an edit menu 303, and a window menu 305 are displayed on the toolbar of this window.

  When “display expression list” in the window menu 305 is selected and an expression list file is specified, the screen control unit 103 reads the expression correlation data from the expression correlation data storage unit 105, and the degree of correlation between the expression buffer 120 and the expression Store in the buffer 122. Next, the screen control unit 103 displays the expression list editor that reads the buffer data from the expression buffer 120 and displays the expression in the expression list field 401. FIG. 4 shows this state.

  In the next step S205, when an expression is input to the expression input field 403 and the add button 404 is pressed, the added expression is registered in the expression correlation data and stored in the expression correlation data storage unit 105 (step S207). Thereafter, the process returns to step S203. In this series of procedures, the screen control unit 103 reads an expression from the expression input field 403 and stores it in the expression buffer 120. When the expression list editor (FIG. 4) is closed, the screen control unit 103 reads the expression buffer 120 and stores the contents in the expression correlation data storage unit 105.

  If the add button 404 is not pressed in step S205, the processing procedure proceeds to step S209. In this step, when an expression in the expression list field 401 is selected and the mapping button 405 is pressed, the screen control unit 103 reads the expression selected in the expression list field 401 as an object and stores it in the object buffer 118. The conceptual map generation unit 101 generates a label object in the conceptual map field 307 of the conceptual map editor using the read object as a label object. FIG. 5 shows a conceptual map editor at this time.

  Next, the designation of the relationship between the label objects is confirmed by a user operation on the editor window of FIG. 5 (step S211). That is, for example, when another label object is inserted into the label object of the conceptual map field 307 by a mouse operation (Yes in step S211), the conceptual map generation unit 101 inserts the inserted label object into the domain object. The previous label object is changed to a node object (step S213). Further, when a link is established between the label objects in the conceptual map field 307 using the mouse, a link object is created and displayed. Here, the screen control unit 103 reads the link of the conceptual map field 307 as a link object and stores it in the object buffer 118. The conceptual map generation unit 101 changes a label object connected via a link to a node object. This state is shown in FIG.

  In the next step S215, when a certain node is selected in the screen of FIG. 6 and “input CPT (conditional probability table)” in the edit menu 303 is selected (Yes in step S215), as shown in FIG. Then, the conditional probability table input editor of the selected node is opened. When a value is input to the CPT input field 701 on this editor screen and the input button 703 is pressed, the processing procedure proceeds to step S217. In this step, the screen control unit 103 reads the value input in the input field 701 and stores the value in the object buffer 118 as the CPT of the object.

  On the other hand, when “preliminary input” in the edit menu 303 is selected in a state where a node is selected, the prior probability input editor for the selected node shown in FIG. 8 is opened. When a value is input to the prior probability input field 801 and the input button 803 is pressed, the processing procedure proceeds to step S217. In this step, the screen control unit 103 reads the value input in the input field 801 and stores the value in the object buffer 118 as the object prior probability.

  On the other hand, when the input operation of the random variable is completed, “Save” in the file menu 301 is selected (No in step S215). Then, the conceptual map generation unit 101 reads the buffer from the object buffer 118 and stores the created conceptual map as data in the conceptual map data storage unit 104 (step S219).

  Next, a procedure for creating an expression list will be described with reference to FIGS. FIG. 9 is a flowchart showing a procedure for creating an expression list. This procedure is performed by the user using the GUI screen. When “Create Expression / Correlation” in the window menu 305 in the window of FIG. 3 is selected and a document file is designated, the screen control unit 103 activates the expression / correlation degree generation unit 102. The description content of the designated document file is called a problem area.

  Of the activated programs, the text mining unit 102a reads the document data from the document data storage unit 106 and the dictionary data from the dictionary data storage unit 107, and creates an expression by a text mining technique (step S901). Text mining is a technology for discovering useful information from a large amount of text information. Specifically, words are extracted from text information by morphological analysis, and these words are statistically processed to determine their appearance frequency and simultaneous appearance relationships. It is a technique to analyze.

  The created representation is stored in the representation buffer 120. In the next step S903, the correlation degree calculation unit 102b reads the expression buffer from the expression buffer 120 and the document data from the document data storage unit 106, and calculates the number of expressions existing between the expressions. This calculated value indicates the degree of correlation between expressions, and the calculated correlation value is stored in the correlation degree buffer 122 (step S903).

  Next, a specific procedure for creating a label object will be described. This processing is related to step S209 in FIG. 2, and is performed by processing using the user's GUI screen. In step S209, the mapped expression is read, a label object is created in the conceptual map editor, and then a display label is selected. When “correlation expression list display” in the window menu 305 of the conceptual map editor is selected in this state, the screen control unit 103 extracts the selected label name. Then, the expression and correlation are read from the expression buffer 120 and the correlation buffer 122 between expressions, and a correlation expression list window in which expressions having a correlation with the selected label are displayed in the expression list field 1001 is displayed. This window is shown in FIG.

In FIG. 10, when an expression in the expression list field 1001 is selected with the mouse and the mapping button 1005 is pressed, the screen control unit 103 reads the expression selected in the expression list field 1001 as an object, and the object buffer 118. The conceptual map generation unit 101 generates a label object in the conceptual map field 307 of the conceptual map editor using the read object as a label object. As described above, in this embodiment, “expression” that is a candidate for a node name or a domain name. Is defined and displayed in the conceptual map editor. A user defines a relationship between labels by operating a mouse on a label object arranged on the conceptual map editor, thereby creating a node name and a domain name. At this time, a “expression” is generated by text mining using a document and a dictionary (“expression” creation rule) designated by the user. Further, for each generated “expression”, an “expression” having a high correlation with the expression is calculated. In this way, by providing a GUI environment called a concept map editor, a user can easily construct a Bayesian network. In other words, two types of relationships (causal relationship between nodes or relationship between nodes and domain) can be specified with GUI operation for a set of object labels, so that the burden of thought work and work time of an expert can be reduced. Can be reduced. Furthermore, since the label names are created in advance in a database and displayed on the screen, the burden of work for the expert to consider the label names can be reduced. Furthermore, the label newly added by the expert can be added to the label candidate list (expression list), whereby the knowledge of the expert can be accumulated in the label candidate list. Such a label candidate list can be reused even when another expert constructs a BN with a conceptual map, so that a more user-friendly system can be constructed. Accordingly, it is possible to provide a Bayesian network construction support device that can easily create a Bayesian network.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

The block diagram which shows the structure of the BN construction | assembly assistance apparatus concerning this invention. The flowchart which shows the preparation procedure of a concept map. The figure which shows the GUI window of a concept map editor. The figure which shows the GUI window of an expression list editor. Diagram showing the conceptual diagram editor with the label object generated The figure which shows the conceptual diagram editor by which the node, the definition area, and the link object were produced | generated. The figure which shows the CPT input editor window of a node. The figure which shows the prior probability input editor window of a node. The flowchart which shows the preparation procedure of an expression and correlation list. The figure which shows a correlation expression list window.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 100 ... BN construction support program, 101 ... Concept map generation part, 102 ... Expression and correlation degree generation part, 102a ... Text mining part, 102b ... Correlation degree calculation part, 103 ... Screen control part, 104 ... Concept map data storage part, DESCRIPTION OF SYMBOLS 105 ... Expression correlation data storage part, 106 ... Document data storage part, 107 ... Dictionary data storage part, 108 ... Input part, 110 ... Storage medium, 112 ... Output part, 114 ... Storage part, 116 ... Processor, 117 ... Internal data Holding unit, 118 ... object buffer, 120 ... expression buffer, 122 ... correlation buffer between expressions, 301 ... file menu, 303 ... edit menu, 305 ... window menu, 307 ... conceptual map field, 401 ... expression list field, 403 ... Expression input field, 404 ... Add button, 405 ... Mapin Button, 701 ... CPT input field, 703 ... input button, 801 ... prior probability input field, 803 ... input button, 1001 ... representation list field, 1005 ... mapping button

Claims (2)

A Bayesian network construction support device that supports editing of a node and its domain constituting a Bayesian network,
An expression database that stores a plurality of expressions related to the problem area in association with each expression related to each other,
The representation is displayed as a label object on a GUI (Graphical User Interface) screen, and when any of the displayed label objects is selected, a label object related to the selected label object is displayed from the representation database. Display control means for reading and displaying on the GUI screen;
Bayesian network construction support, comprising: construction means for constructing a Bayesian network based on the relationship on the GUI screen when a mutual relationship is specified between the plurality of label objects on the GUI screen. apparatus. 2. The Bayesian network construction support apparatus according to claim 1, wherein the expression database is generated by a text mining method using document data and dictionary data prepared in advance.

Images