JP2009140113A - Dictionary editing device, dictionary editing method, and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for enhancing an ontology label, and also to provide its method. <P>SOLUTION: A dictionary editing device is configured to perform the updating processing of dictionary data, e.g., an ontology dictionary, including relation information between the label being property information and the class being attribute information by class unit as a concept to express existence. A restriction condition as a text corpus retrieval condition is determined through the use of the label information of the class related to the class to be updated. The text corpus is retrieved according to the restriction condition so as to extract an association sentence including the label candidate of the class to be updated. An associated character string to be the label candidate of the class to be updated is extracted from the association sentence according to the restriction condition, thereby performing editing processing to set the extracted associated character string as the label of the class to be updated. By the above configuration, the ontology label is efficiently enhanced. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a dictionary editing device, a dictionary editing method, and a computer program. More specifically, a dictionary editing device for constructing and updating an ontology, which is a kind of dictionary of concepts representing existence (including vocabularies belonging to concepts), and describing a word relationship between a superordinate concept and a subordinate concept, The present invention relates to a dictionary editing method and a computer program.

  For example, a process for extracting a term applied in data processing from a natural language document such as a search key applied in a database search or an index as an index of a term dictionary is a technique required in various data processing fields. is there.

  A collection of various text data is called a text corpus. Research on extracting terms from documents included in a text corpus, that is, extracting terms as meaningful language units, has been performed. For example, if it is a common document such as [car runs on road], it is possible to extract morphemes such as [car], [road], and [run] by applying a general morphological analysis system. . The morpheme analysis system is known as a system that applies a predetermined dictionary for morpheme analysis and performs a part-of-speech recognition process by segmenting into morphemes that are semantic minimum units based on dictionary registered words. .

  Further, for example, a thesaurus or ontology is a dictionary that systematically summarizes terms used when performing data search. A thesaurus is a dictionary that systematically summarizes terms with similar meanings. Ontology means a classification system of knowledge and a set of inference rules, is a kind of concept dictionary representing existence, and is configured as a dictionary describing the word relationship between a superordinate concept and a subordinate concept. For example, a superordinate concept, a subordinate concept, a term (label) representing the concept, relation information between concepts, and the like are registered. In general, an ontology has a class hierarchy in which a connection relationship between nodes is defined with a class corresponding to a concept as a node. Specifically, the ontology describes a class corresponding to [concept], properties as class property information, and attributes as class attribute information.

  For example, Patent Literature 1 (Japanese Patent Laid-Open No. 2001-14166) and Patent Literature 2 (Japanese Patent Laid-Open No. 2006-309446) are disclosed as related arts that disclose the thesaurus and ontology construction and update processing.

  Patent Document 1 (Japanese Patent Application Laid-Open No. 2001-14166) uses a similarity of node attributes (node name, node type, node definition) between ontologies when there are a plurality of ontologies. A configuration is disclosed in which a node conversion rule corresponding to each data schema is prepared and associated with data automatically expressed in various formats such as XML, RDF, and DTD.

  Further, Patent Document 2 (Japanese Patent Application Laid-Open No. 2006-309446) discloses that, in a hierarchical database constituting an ontology, for example, when adding a class defined corresponding to [concept] registered in the ontology, Using the similarity of attributes (class and details field of the attribute), similar classes are extracted and presented to the user for confirmation. Ontology updates such as adding classes to the ontology and editing attributes of existing classes A configuration for improving the efficiency of ontology is proposed, such as using the past editing history that is held to present a candidate of an attribute value to the user so that the candidate can be selected.

  Further, Non-Patent Document 1 (Extended a thesaurus by classifying Words, (In Proceedings of the ACL / EACL Workshop on Automatic Information Extraction and Bild 19). When creating a feature vector using co-occurring words and adding a new word to the thesaurus, create a new feature vector for the newly added word and use the similarity of the feature vector with the existing word The technology for determining the location to add is disclosed.

As described above, there are a plurality of documents that disclose a system for constructing or updating a thesaurus or ontology. However, these prior arts do not show a configuration for enhancing properties (property information) that are lacking in existing ontologies. For example, Patent Document 2 provides support for class editing, but supports the use of past editing history, and is not suitable for enhancing properties that take values specific to classes such as labels. In order to perform advanced knowledge processing by linking text data and ontology, it is important that a label indicating the existence of the class in the text is appropriately set. This is because a class represents a concept in the ontology, and a label appears as an instance of the concept in the text. Therefore, enriching the label of the class set in the ontology increases the usefulness of the ontology.
Japanese Patent Laid-Open No. 2001-14166 JP 2006-309446 A Extending a thesaurus by classifying words. (In Proceedings of the ACL / EACL Workshop on Automatic Information Extraction and Building of Lexical Semantic Resources, 1997.)

  An object of the present invention is to provide a dictionary editing apparatus, a dictionary editing method, and a computer program for enhancing a label, which is a kind of property information (property) of a class set in an ontology.

  For example, when a certain ontology exists, the level of property fulfillment may differ depending on the class. This is remarkable when, for example, an ontology created in a different language (such as English) is translated into Japanese. Even if this is not the case, depending on the bias in the knowledge of the ontology builder, the label notation for each class may be biased, and the label for a particular class may be insufficient. In the present invention, it is possible to enrich a label by using a text corpus including a specific class label and a related class label, for example, a text highly relevant to an ontology. Although the present invention describes enhancement of class labels, the enhancement method described in the present invention can be applied to other properties having properties similar to labels.

The first aspect of the present invention is:
A dictionary editing device that sets a concept representing existence as a class, and executes update processing of dictionary data in which data including class-specific property information and data including class-related attribute information is included. Yes,
A related class extraction unit that extracts a related class of the update target class from the dictionary data;
A constraint condition extraction unit that determines a constraint condition as a search condition of a text corpus based on label information corresponding to a related class;
A related sentence extraction unit that searches a text corpus according to the constraint condition and extracts a related sentence including a candidate label of the update target class;
In accordance with the constraint condition, a related character string extraction unit that extracts a related character string that is a label candidate of the update target class from the related sentence extracted by the related sentence extraction unit;
An edit processing unit that sets the related character string extracted by the related character string extraction unit as a label of the update target class;
The dictionary editing apparatus is characterized by comprising:

  Furthermore, in one embodiment of the dictionary editing device of the present invention, the dictionary editing device further includes a related character string presentation unit that presents the related character string extracted by the related character string extraction unit to the editing operator. The edit processing unit executes a process of setting the selected related character string as a label of the update target class in response to input of selection information of the related character string by the editing operator.

  Furthermore, in one embodiment of the dictionary editing apparatus of the present invention, the dictionary data is dictionary data in which a plurality of classes are set as a hierarchical structure, and the related class extraction unit is configured to use a super class ( A related class of at least one of a parent class), a subclass (child class), and a sibling class (sibling class) is extracted, and the constraint condition extraction unit includes a super class (parent class), a subclass (child class), and a sibling class ( The constraint condition is determined based on label information corresponding to two related classes selected from the sibling class.

  Furthermore, in one embodiment of the dictionary editing apparatus of the present invention, the constraint condition extraction unit determines a character string as a positive constraint condition and a character string as a negative constraint condition, and the related sentence extraction unit includes: A process of extracting from the text corpus text that includes a character string of the same type as a character determined as a positive constraint condition and does not include a character string of the same type as a character determined as a negative constraint condition.

  Furthermore, in one embodiment of the dictionary editing apparatus according to the present invention, the constraint condition extraction unit includes a super class (parent class), a sub class (child class), and a sibling class (sibling class) extracted as related classes of the update target class. The longest common part of the label information corresponding to the two selected related classes is determined as a positive constraint condition.

  Furthermore, in one embodiment of the dictionary editing apparatus according to the present invention, the constraint condition extraction unit deletes a negative constraint condition included in a positive constraint condition extracted by referring to a label of a related class of the update target class. The process is executed to determine the final constraint condition.

  Furthermore, in one embodiment of the dictionary editing apparatus of the present invention, the related character string extraction unit extracts a continuous character string including the character string as the positive constraint condition from the related sentence extracted by the related sentence extraction unit. , And extracting as a related character string to be a label candidate of the update target class.

  Furthermore, in an embodiment of the dictionary editing apparatus according to the present invention, the related character string extraction unit performs a morphological analysis of the related sentence extracted by the related sentence extraction unit, and includes a character string as the positive constraint condition. A continuous part of nouns is extracted as a related character string that is a label candidate of the update target class.

  Furthermore, in one embodiment of the dictionary editing apparatus according to the present invention, the dictionary is an ontology dictionary configured by registration information of classes having a hierarchical structure.

Furthermore, the second aspect of the present invention provides
A dictionary editing method for executing dictionary data editing processing in an information processing device,
The dictionary data to be edited is a dictionary data in which a concept representing existence is set as a class, and data including a label that is property information corresponding to a class and data including relationship information between classes that is attribute information corresponding to a class,
A related class extraction unit extracts a related class of the update target class from the dictionary data; and
A constraint condition extraction step in which a constraint condition extraction unit determines a constraint condition as a text corpus search condition based on label information corresponding to the related class;
A related sentence extracting unit performs a search of a text corpus according to the constraint condition, and extracts a related sentence including a label candidate of the update target class; and
A related character string extraction unit that extracts a related character string that becomes a label candidate of the update target class from the related sentence extracted by the related sentence extraction unit according to the constraint condition;
An edit processing unit that sets the related character string extracted in the related character string extraction step as a label of the update target class; and
A dictionary editing method characterized by comprising:

  Furthermore, in one embodiment of the dictionary editing method of the present invention, in the dictionary editing method, the related character string presentation unit further presents the related character string extracted in the related character string extraction step to the editing operator. A related character string presenting step, wherein the editing processing step executes a process of setting the selected related character string as a label of the update target class in response to input of selection information of the related character string by the editing operator It is characterized by doing.

  Furthermore, in an embodiment of the dictionary editing method of the present invention, the dictionary data is dictionary data in which a plurality of classes are set as a hierarchical structure, and the related class extracting step includes a super class ( A related class of at least one of a parent class), a subclass (child class), and a sibling class (sibling class) is extracted, and the constraint extraction step includes a super class (parent class), a subclass (child class), and a sibling class ( The constraint condition is determined based on label information corresponding to two related classes selected from the sibling class.

  Furthermore, in one embodiment of the dictionary editing method of the present invention, the constraint condition extraction step is a step of determining a character string as a positive constraint condition and a character string as a negative constraint condition, and the related sentence extraction The step includes executing a process of extracting, from the text corpus, text that includes a character string of the same type as the character determined as the positive constraint condition and does not include the character string of the same type as the character determined as the negative constraint condition. And

  Furthermore, in one embodiment of the dictionary editing method according to the present invention, the constraint condition extraction step includes a super class (parent class), a sub class (child class), and a sibling class (sibling class) extracted as related classes of the update target class. The longest common part of the label information corresponding to the two selected related classes is determined as a positive constraint condition.

  Furthermore, in one embodiment of the dictionary editing method of the present invention, the constraint extraction step deletes a negative constraint included in a positive constraint extracted by referring to a label of a related class of the update target class. The process is executed to determine the final constraint condition.

  Furthermore, in one embodiment of the dictionary editing method of the present invention, the related character string extraction step includes a continuous character string including the character string as the positive constraint condition from the related sentence extracted in the related sentence extraction step. , And extracting as a related character string to be a label candidate of the update target class.

  Furthermore, in one embodiment of the dictionary editing method of the present invention, the related character string extraction step performs a morphological analysis of the related sentence extracted in the related sentence extraction step, and includes a character string as the positive constraint condition. A continuous part of nouns is extracted as a related character string that is a label candidate of the update target class.

  Furthermore, in an embodiment of the dictionary editing method of the present invention, the dictionary is an ontology dictionary configured by registration information of classes having a hierarchical structure.

Furthermore, the third aspect of the present invention provides
A computer program for executing dictionary data editing processing in an information processing apparatus;
The dictionary data to be edited is a dictionary data in which a concept representing existence is set as a class, and data including a label that is property information corresponding to a class and data including relationship information between classes that is attribute information corresponding to a class,
A related class extraction step for causing the related class extraction unit to extract a related class of the update target class from the dictionary data;
A constraint condition extraction step for causing the constraint condition extraction unit to determine a constraint condition as a text corpus search condition based on the label information corresponding to the related class;
A related sentence extraction unit that performs a search of a text corpus according to the constraint condition and causes a related sentence extraction unit to extract a related sentence including a label candidate of the update target class; and
A related character string extraction step for causing the related character string extraction unit to extract a related character string that is a candidate label for the update target class from the related sentence extracted by the related sentence extraction unit in accordance with the constraint condition;
An editing processing step for causing the editing processing unit to set the related character string extracted in the related character string extracting step as a label of the update target class;
There is a computer program characterized by comprising:

  The computer program of the present invention is, for example, a computer program that can be provided by a storage medium or a communication medium provided in a computer-readable format to a general-purpose computer system that can execute various program codes. . By providing such a program in a computer-readable format, processing corresponding to the program is realized on the computer system.

  Other objects, features, and advantages of the present invention will become apparent from a more detailed description based on embodiments of the present invention described later and the accompanying drawings. In this specification, the system is a logical set configuration of a plurality of devices, and is not limited to one in which the devices of each configuration are in the same casing.

  According to the configuration of one embodiment of the present invention, update processing is performed for dictionary data, for example, an ontology dictionary, which includes labels that are property information in terms of classes as a concept representing existence and relationship information between classes that is attribute information. In this configuration, the constraint condition as the text corpus search condition is determined using the label information of the related class of the update target class, the text corpus is searched according to the constraint condition, and the related sentence including the label candidate of the update target class is obtained. Further, an editing process is performed in which a related character string that is a candidate label for the update target class is extracted from the related sentence according to the constraint condition, and the extracted related character string is set as a label of the update target class. With this configuration, it is possible to efficiently execute the ontology label enhancement.

  Details of a dictionary editing apparatus, a dictionary editing method, and a computer program according to an embodiment of the present invention will be described below with reference to the drawings.

  FIG. 1 shows a configuration diagram of a dictionary editing apparatus according to an embodiment of the present invention. The dictionary editing apparatus according to the present invention corresponds to the dictionary data stored in the ontology storage unit 111 which is a dictionary storage unit, in this example, the [class] to be processed from the ontology, that is, the concept representing the presence registered in the ontology. [Class] is selected, and a related class stored in the ontology storage unit 111 is extracted, and an update of a label recorded as property information (property) of [Class] to be processed, for example, a new label Perform additional processing.

  As shown in FIG. 1, a dictionary editing apparatus 100 according to an embodiment of the present invention includes a class selection unit 101, a related class extraction unit 102, a constraint condition extraction unit 103, a related sentence extraction unit 104, and a related character string extraction unit 105. A related character string presentation unit 106 and an ontology editing processing unit 107. Further, an ontology storage unit 111 that stores an ontology to be edited and a text corpus storage unit 112 that stores a text corpus used in the related sentence extraction unit 104 are used.

  Before describing the processing using the dictionary editing device 100, the ontology as dictionary data stored in the ontology storage unit 111, which is a dictionary storage unit, and the text corpus stored in the text corpus storage unit 112 will be described with reference to FIGS. This will be described with reference to FIG.

  First, the ontology stored in the ontology storage unit 111 will be described with reference to FIG. As described above, the ontology is a kind of concept dictionary representing existence, and is configured as a dictionary describing the relationship between the superordinate concept and the subordinate concept. For example, it has a hierarchical structure in which a relationship between a superordinate concept and a subordinate concept of a concept is defined by a connection relationship between nodes. In the ontology, a class corresponding to [concept] and a property as property information of the class are described as registration information.

  FIG. 2 shows an information configuration example of one registered word described according to OWL known as an ontology description language. In the ontology storage unit 111, registration information as shown in FIG. 2 is recorded for each class corresponding to a concept that is a registered word.

As shown in FIG. 2, the ontology includes, for example, the following information:
(A) a class corresponding to [concept],
(B) properties as class property information;
(C) Attributes as class attribute information (relationship information with other classes, such as parent-child relationships),
These pieces of information are registered.

  [Class] is a field for recording a concept name representing registered existence, and [Property information (property)] records information on a class such as a label. Furthermore, relation information with other classes is also recorded as class attribute information (attributes).

  The relationship with other classes will be described with reference to FIG. The ontology records the registration information shown in FIG. 2 in units of classes corresponding to concepts, and has a configuration that allows each class to be identified as one node and the relationship between classes to be identified as a connection relationship between nodes. As relations between classes, three relations are defined when roughly classified. The class relationship with the class A 201 will be described with the class A 201 shown in FIG. 3 as the class of interest.

First, the class P211 is a superordinate concept class corresponding to the class A201.
This class P211 is a superclass (parent class) of class A201.

The class A 201 includes a class S 221 and a class T 222 as subordinate concept classes corresponding to the class A 201.
The class S221 and the class T222 are subclasses (child classes) of the class A201.

Further, classes registered as subordinate concepts of class P211 include class B212 and class C213 in addition to class A201. Class B212 and class C213 have the same superclass (parent class) = class P211 as class A201.
The class B212 and the class C213 are sibling classes (sibling classes) of the class A201.

In this way, the ontology class relationship is
(A) Super class (parent class)
(B) Subclass (child class)
(C) Sibling class (sibling class)
These class relationships are defined. Furthermore, a subclass of a subclass (grandchild class) and the like are also defined.

  In the dictionary editing apparatus of the present invention, one class stored in the ontology storage unit 111 is selected as a processing target, and a related class stored in the ontology storage unit 111 is extracted, and the processing target [class] is selected. The label recorded as the property information (property) is updated, for example, a process of adding a new label is performed to enrich the label.

  Next, an example of a text corpus stored in the text corpus storage unit 112 will be described with reference to FIG. The text corpus storage unit 112 is a database that stores various texts. As shown in FIG. 4, an ID as an identifier is set for each text.

  Hereinafter, a specific processing example using the dictionary editing apparatus 100 shown in FIG. 1 will be described. In the present embodiment, an example of processing using an ontology storage unit that stores medical field terms and a text database that collects medical field texts will be described.

First, in the class selection unit 101, a user (ontology editor) inputs a designation of a class to be enriched from the ontology stored in the ontology storage unit 111.
The designated class is [Class A].
For example, the class A designated as the processing target has registration information similar to that described with reference to FIG.

  The related class extraction unit 102 determines super classes (parent classes), subclasses (child classes), sibling classes (sibling classes) as the related classes of class A from the registration information of class A, and determines the determined related classes as Extracted from the ontology storage unit 111.

  For example, in the ontology described in OWL shown in FIG. 2, the superclass can be extracted by referring to the “SubClassOf” property of class A. As for the subclass, if the entire superclass structure is extracted, the specified class can be extracted as a class having the superclass. The sibling class can also be extracted as a subclass of the superclass of class A. Note that as a related class to be extracted, a subclass of a subclass (grandchild class) or the like may be further extracted.

  Next, the constraint condition extraction unit 103 determines a constraint condition for searching the text corpus in the text corpus storage unit 112 from the label of the related class extracted as the related class of class A.

What are constraints?
It is a constraint condition when searching for a text corpus in the text corpus storage unit 112, and determines [positive constraint condition] and [negative constraint condition]. [Positive constraint] and [Negative constraint] are used as follows when the related sentence extraction unit 104 searches for a text corpus in the text corpus storage unit 112.
A text search process of the related sentence extraction unit 104 = a character string of the same type as a character determined as a negative constraint including a character string of the same type as a positive constraint from the text corpus of the text corpus storage unit 112 Extract non-contained text.

The constraint condition extraction unit 103 sets the constraint condition for searching the text corpus in the text corpus storage unit 112 from the label of the related class extracted as the related class of the class A in a different manner according to the class relationship with the class A. decide.
Below,
(A) Constraint extraction by label comparison between sibling (sibling) class and sibling (sibling) class (b) Constraint extraction by label comparison between super (parent) class and sibling (sibling) class (c) Super (parent ) Constraint extraction by label comparison between class and sub (child) class (d) Constraint extraction by label comparison between sub (child) class and sub (child) class

  An example of processing of these four patterns will be described. Note that the constraint condition determination pattern described below is merely an example, and other than the processing described below, a configuration may be adopted in which a constraint condition is set by setting another similar pattern between the extracted related classes. Hereinafter, each of the processes (a) to (d) will be described.

(A) Constraint extraction by label comparison between sibling class and sibling class First, two sibling classes and sibling class corresponding to class A to be processed for label enrichment An example of a constraint condition extraction process based on a label comparison will be described with reference to FIG.

  In the constraint extraction process by comparing the sibling class with the sibling class, the labels of the two sibling classes corresponding to the class A to be processed for label enrichment are compared, and the longest common Extract the part as a positive constraint. Further, a difference between labels including the common part is extracted as a negative constraint condition.

  For example, as shown in FIG. 5, there are sibling classes 1 and 303 and sibling classes 2 and 304 as sibling (sibling) classes that share the super (parent) class 302 in the class A 301 to be processed for label enhancement.

In such a setting, labels as property information of the sibling classes 1 and 303 and the sibling classes 2 and 304 are extracted, and the labels are compared. In the example shown in FIG.
In sibling class 1,303,
* Lung cancer * Lung cancer These labels are registered.
In sibling class 2,304,
* Breast cancer * Breast cancer These labels are registered.

The labels of these two sibling classes are compared, and the longest common part is extracted as a positive constraint. Further, a difference between labels including the common part is extracted as a negative constraint condition. In the example shown in FIG.
"Cancer" and "Cancer" as positive constraints
Extract these labels,
"Lung" and "milk" as negative constraints
Extract these labels.

(B) Extraction of constraint condition by label comparison between super (parent) class and sibling (sibling) class Next, a super (parent) class and sibling (sibling) class corresponding to class A to be processed for label enhancement An example of constraint condition extraction processing based on label comparison will be described with reference to FIG.

  In the constraint condition extraction process by comparing the labels of the super (parent) class and sibling (sibling) class, the labels of these classes are compared, and the longest common part is extracted as a positive constraint condition. Also, the difference between sibling class labels including the common part is extracted as a negative constraint.

  For example, as shown in FIG. 6, there are a super (parent) class 312 and a sibling (sibling) class 313 in the class A 311 to be processed for label enhancement.

In such a setting, labels as property information of the super (parent) class 312 and sibling (sibling) class 313 are extracted, and the labels are compared. In the example shown in FIG.
Super (parent) class 312 includes
* Cancer * Cancer * Carcinoma These labels are registered.
In the sibling class 313,
* Lung cancer * Lung cancer These labels are registered.

The labels of these two classes are compared, and the longest common part is extracted as a positive constraint. Further, a difference between labels including the common part is extracted as a negative constraint condition. In the example shown in FIG.
"Cancer" and "Cancer" as positive constraints
Extract these labels,
"Lung" as a negative constraint
Extract these labels.

(C) Constraint extraction by label comparison between super (parent) class and sub (child) class Next, a super (parent) class and a sub (child) class corresponding to class A to be processed for label enrichment An example of constraint condition extraction processing based on label comparison will be described with reference to FIG.

  In the constraint extraction process based on label comparison between the super (parent) class and sub (child) class, the labels of these classes are compared, the longest common part is extracted as a positive constraint, and the common part is further extracted. Superclass labels that contain are also extracted as positive constraints. Negative constraints are not extracted.

  For example, as shown in FIG. 7, the class A 321 to be processed for label enhancement includes a super (parent) class 322 and a sub (child) class 323.

In such a setting, labels as property information of the super (parent) class 322 and the sub (child) class 323 are extracted, and the labels are compared. In the example shown in FIG.
Super (parent) class 322 has
* Tumor * Tumor * Mass * Abnormal growth These labels are registered.
The sub (child) class 323 includes
* Mesothelioma These labels are registered.

By comparing the labels of these two classes, the longest common part is extracted as a positive constraint, and the label of the superclass including the common part is also extracted as a positive constraint. Negative constraints are not extracted. In the example shown in FIG.
"Tumor", "Tumor", "Tumor" as positive constraints
Extract these labels.

(D) Extraction of constraint conditions by label comparison between sub (child) class and sub (child) class Next, constraint by label comparison of two sub (child) classes corresponding to class A to be processed for label enrichment An example of condition extraction processing will be described with reference to FIG.

  In the constraint condition extraction process by comparing the labels of the sub (child) class and the sub (child) class, the labels of these classes are compared, and the longest common part is extracted as a positive constraint condition. Negative constraints are not extracted.

  For example, as shown in FIG. 8, the class A 331 to be processed for label enhancement includes a sub (child) class 332 and a sub (child) class 333.

In such a setting, labels as property information of the sub (child) class 332 and the sub (child) class 333 are extracted, and the labels are compared. In the example shown in FIG.
The sub (child) class 332 includes
* Cancer * Cancer * Cancer type * Abnormal growth These labels are registered.
Sub (child) class 333 includes
* Adenoma * Adenoma These labels are registered.

The labels of these two classes are compared, and the longest common part is extracted as a positive constraint. Negative constraints are not extracted. In the example shown in FIG.
"Tumor" as a positive constraint
Extract this label.

  The above is the constraint condition extraction processing executed by the constraint condition extraction unit 103 shown in FIG. As described above, the constraint condition extraction unit 103 determines the constraint condition for searching the text corpus in the text corpus storage unit 112 from the label of the related class extracted as the related class of class A.

Further, the constraint condition extraction unit 103 extracts a constraint condition from among all the related classes by the above-described process, and further determines a final constraint condition by using the extracted positive / negative constraint condition.
Specifically, a process of deleting a negative constraint included in a positive constraint is performed.
For example,
"Tumor" is extracted as a positive constraint,
If "tumor" was extracted as a negative constraint,
Negative constraint "tumor" is removed.

After the positive / negative constraint conditions are determined in this way, the related sentence extraction unit 104 performs a text search from the text corpus storage unit 112.
The related sentence extraction unit 104 extracts a set of sentences that include a positive constraint condition and does not include a negative constraint condition.

  For example, if there are “cancer” and “cancer” as positive constraints, and “lung” and “milk” as negative constraints, a sentence like “… is likely esophageal cancer…” is extracted. However, it does not extract sentences such as "... probable lung cancer ...".

  The text extracted by the related sentence extraction unit 104 from the text corpus storage unit 112 using the positive constraint condition and the negative constraint condition is input to the related character string extraction unit 105.

  The related character string extraction unit 105 extracts a related character string from the set of sentences extracted by the related sentence extraction unit 104 from the text corpus storage unit 112 using a positive constraint condition or the like. This related character string extraction process can be executed using, for example, one of the following two methods (method a) and (method b), or a combination thereof.

(Method a) Extracting a continuous character string including characters of the same type as a positive constraint condition Related character string by a method of extracting a continuous character string including characters of the same type as a positive constraint condition from a set of extracted sentences The extraction process will be described.

From a set of sentences extracted from the text corpus storage unit 112,
Select a character string that contains the same type of character as the positive constraint as the first character, and extract consecutive characters that contain that character. Similarly, a character that contains the same type of character as the last character Select a column and extract consecutive strings that contain that string

For example,
The positive constraint is "tumor"
If there is a sentence collection of “… and infarcts, neoplastic lesions…” and “I am looking at a bladder tumor…”
"Neoplastic focus" and "bladder tumor"
These character strings are extracted.
As the character type, kanji, hiragana, katakana, alphanumeric characters, symbols, and the like are used.

(Method b) Performing morphological analysis on a sentence and extracting a series of morphemes including positive constraint conditions For a set of sentences extracted from the text corpus storage unit 112, for example, a bowl (known as a morphological analysis algorithm) http://chasen.naist.jp/hiki/ChaSen/) etc., morphological analysis is performed, the extracted sentence is divided, and a set of consecutive nouns containing the same kind of word as a positive constraint is used as a related character string. Extract.

For example, a positive constraint is "tumor"
As a sentence extracted from the text corpus storage unit 112 and separated by morphological analysis
"... / ya (particle) / infarction (noun) /, (symbol) / neoplastic (noun) / lesion (noun) / a (particle) / ...", "bladder (noun) / tumor (noun) / Particle) / see (verb) / te (particle) / is (verb) / ... "
As a set of consecutive nouns that contain words of the same kind as positive constraints,
“Tumor lesion” and “bladder tumor” are extracted as related character strings.

  If there are many sentences in which a positive constraint appears, the number of related character strings to be extracted tends to increase. Therefore, ranking of related character strings may be performed to extract the top n (n is an arbitrary number). For example, the entropy of the related character string in the text corpus can be obtained, and the top n having the highest value can be taken. A method for obtaining entropy of a character string is generally known, and a method described in the literature “Kitakenji. Stochastic language model. The University of Tokyo Press, 1999” or the like can be applied.

  With the above processing, the related character string extraction unit 105 extracts a related character string from the set of sentences extracted by the related sentence extraction unit 104 from the text corpus storage unit 112 using a positive constraint condition or the like. The extracted related character string is input to the related character string presentation unit 106, and the extracted related character string is presented to the user (ontology editor) using a display or the like. In addition to display output, audio output and the like can be applied, and the presentation method to the user is not particularly limited. Moreover, when ranking of related character strings is performed, the ranking and score may be presented together.

  The ontology editing processing unit 107 is a processing unit that executes editing of the label as the property information of the class A for which the label is to be enriched. For example, when the user (ontology editor) determines whether the presented related character string is valid as a label as the property information of class A to be enriched, and determines that it is valid Then, the related character string is additionally written as a class A label.

  In addition, it is good also as a structure which performs a process without a user (ontology editor) intervention, In that case, the process which adds the related character string extracted in the related character string extraction part 106 as a class A label is performed. What is necessary is just to be the structure to do. In this case, the process follows the dotted arrow shown in FIG. 1, and the related character string presentation unit 106 can be omitted.

  Next, a processing sequence using the dictionary editing apparatus 100 shown in FIG. 1 will be described with reference to the flowchart shown in FIG.

  First, in step S101, a class to be enriched with a label is selected from the ontology stored in the ontology storage unit 111. This is the processing of the class selection unit 101 shown in FIG. 1, for example, may be performed as a selection by the user (ontology editor), or the classes are sequentially selected from the ontology stored in the ontology storage unit 111. It is good also as composition to do. The processing target class [class A] has registration information similar to that described with reference to FIG.

  Next, in step S102, a related class of the processing target class [class A] is extracted. This is the processing of the related class extraction unit 102 shown in FIG. 1, and the super class (parent class), subclass (child class), sibling class (sibling class) are selected as the related class of class A from the registration information of class A. The determined related class is extracted from the ontology storage unit 111.

  Next, in step S103, a constraint condition that is a search condition for a text corpus used to extract a related character string applied to the label of the processing target class [class A] is extracted. This process is the process of the constraint condition extraction unit 103 shown in FIG.

  As described above, the constraint condition is a constraint condition when searching the text corpus in the text corpus storage unit 112, and includes a [positive constraint condition] and a [negative constraint condition]. It is used when searching a text corpus, and the extraction of text that includes the same type of character string as the positive constraint and does not include the same type of character as the negative constraint Become.

  As described above, the constraint condition extraction process in step S103 is determined in a different manner according to the class relationship of the related classes of the processing target class [class A]. That is, as described above with reference to FIGS. 5 to 8, the constraint condition is determined by the following processing mode.

(A) Extraction of constraints by label comparison between sibling class and sibling class Compare the labels of two sibling classes and extract the longest common part as a positive constraint. Further, a difference between labels including the common part is extracted as a negative constraint condition.

(B) Extraction of constraint conditions by label comparison between super (parent) class and sibling (sibling) class Compare the labels of these classes and extract the longest common part as a positive constraint condition. Also, the difference between sibling class labels including the common part is extracted as a negative constraint.

(C) Constraint extraction by label comparison of super (parent) class and sub (child) class Compares the labels of these classes, extracts the longest common part as a positive constraint, and further extracts the common part The superclass labels that it contains are also extracted as positive constraints. Negative constraints are not extracted.

(D) Extraction of constraint condition by label comparison between sub (child) class and sub (child) class The labels of these classes are compared, and the longest common part is extracted as a positive constraint condition. Negative constraints are not extracted.

Further, in step S103, after extracting the constraint conditions between all the related classes, the negative constraint condition included in the positive constraint condition is deleted, and the final constraint condition is determined.
For example,
"Tumor" is extracted as a positive constraint,
If "tumor" was extracted as a negative constraint,
Negative constraint "tumor" is removed.

After the positive and negative constraint conditions are determined in this way, in step S104, text search is performed from the text corpus storage unit 112 by applying the determined constraint conditions. This process is the process of the related sentence extraction unit 104 shown in FIG.
The related sentence extraction unit 104 extracts from the text corpus storage unit 112 a set of sentences that include positive constraint conditions and does not include negative constraint conditions.

  Next, in step S105, a process of extracting a related character string that is a candidate to be added as a class A label from the set of sentences extracted from the text corpus storage unit 112 is executed. This process is the process of the related character string extraction unit 105 shown in FIG.

As described above, the related character string extraction unit 105 uses the following (Method a), (Method b), any one of these methods, or a combination thereof as a candidate related character to be added as a class A label. Extract columns.
(Method a) Extracting a continuous character string including characters of the same type as a positive constraint (Method b) Performing morphological analysis on a sentence to extract a series of morphemes including a positive constraint.

  As mentioned above, the number of related character strings to be extracted tends to increase if there are many sentences in which a positive constraint appears. Therefore, ranking of related character strings is performed, and the top n (n is an arbitrary number) is extracted. May be.

  Next, in step S106, the extracted related character string is presented to the user (ontology editor) using a display or the like. This is the process of the related character string presentation unit 106 shown in FIG. When ranking related character strings, the ranking and score may be presented together.

  Finally, in step S107, label editing is performed in which the presented related character string is additionally written as property information of class A to be enriched. This process is the process of the ontology editing processing unit 107 shown in FIG.

  For example, when the user (ontology editor) determines whether the presented related character string is valid as a label as the property information of class A to be enriched, and determines that it is valid Then, the related character string is additionally written as a class A label.

  As described above, the process may be executed without the intervention of the user (ontology editor). In this case, the related character string extracted in the related character string extraction process in step S105 is directly used as the class A. Execute the process of adding as a label. In this case, the related character string presentation process in step S106 is omitted.

  Finally, a hardware configuration example of an information processing apparatus that constitutes a dictionary editing apparatus that executes the above-described processing will be described with reference to FIG. The information processing apparatus that constitutes the dictionary editing apparatus can be realized as hardware by, for example, a PC, and can execute data search and display data generation and output by executing a program that executes the above-described processing. A CPU (Central Processing Unit) 701 is a process corresponding to an OS (Operating System), a class selection process, a related class extraction process, a constraint condition extraction process, a related sentence extraction process, and a character string extraction process described in the above embodiment. , Related character string presentation processing, ontology editing processing, and the like. These processes are executed according to a computer program stored in a data storage unit such as a ROM or a hard disk of each information processing apparatus.

  A ROM (Read Only Memory) 702 stores programs used by the CPU 701, calculation parameters, and the like. A RAM (Random Access Memory) 703 stores programs used in the execution of the CPU 701, parameters that change as appropriate during the execution, and the like. These are connected to each other via a host bus 704 including a CPU bus.

  The host bus 704 is connected to an external bus 706 such as a PCI (Peripheral Component Interconnect / Interface) bus via a bridge 705. A keyboard 708 and a pointing device 709 are input devices operated by the user. The display 710 is composed of a liquid crystal display device, a CRT (Cathode Ray Tube), or the like, and displays various types of information as text and images.

  An HDD (Hard Disk Drive) 711 includes a hard disk, drives the hard disk, and records or reproduces a program executed by the CPU 701 and information. The hard disk is used as storage means such as an ontology, a text corpus, and a dictionary, and further stores various computer programs such as a data processing program.

  The drive 712 reads data or a program recorded on a removable recording medium 721 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, and reads the data or program as an interface 707 and an external bus 706. , The bridge 705, and the RAM 703 connected via the host bus 704.

  The connection port 714 is a port for connecting the external connection device 722 and has a connection unit such as USB or IEEE1394. The connection port 714 is connected to the CPU 701 and the like via the interface 707, the external bus 706, the bridge 705, the host bus 704, and the like. The communication unit 715 is connected to a network and executes data search by communicating with an external database 801, for example.

  The hardware configuration example of the information processing apparatus shown in FIG. 10 is an example of an apparatus configured by applying a PC, and the dictionary editing apparatus of the present invention is not limited to the configuration shown in FIG. Any configuration can be used as long as the processing described in the above can be executed.

  The present invention has been described in detail above with reference to specific embodiments. However, it is obvious that those skilled in the art can make modifications and substitutions of the embodiments without departing from the gist of the present invention. In other words, the present invention has been disclosed in the form of exemplification, and should not be interpreted in a limited manner. In order to determine the gist of the present invention, the claims should be taken into consideration.

  The series of processing described in the specification can be executed by hardware, software, or a combined configuration of both. When executing processing by software, the program recording the processing sequence is installed in a memory in a computer incorporated in dedicated hardware and executed, or the program is executed on a general-purpose computer capable of executing various processing. It can be installed and run. For example, the program can be recorded in advance on a recording medium. In addition to being installed on a computer from a recording medium, the program can be received via a network such as a LAN (Local Area Network) or the Internet, and installed on a recording medium such as a built-in hard disk.

  Note that the various processes described in the specification are not only executed in time series according to the description, but may be executed in parallel or individually according to the processing capability of the apparatus that executes the processes or as necessary. Further, in this specification, the system is a logical set configuration of a plurality of devices, and the devices of each configuration are not limited to being in the same casing.

  As described above, according to the configuration of one embodiment of the present invention, dictionary data including a label that is property information and a relationship information between classes that are attribute information in units of classes as a concept representing existence, for example, In the configuration that executes the update process of the ontology dictionary, the constraint information as the text corpus search condition is determined using the label information of the related class of the update target class, the text corpus is searched according to the constraint condition, and the update target class The related sentence including the label candidate is extracted, and further, the related character string that becomes the label candidate of the update target class is extracted from the related sentence according to the constraint condition, and the extracted related character string is set as the update target class label. . With this configuration, it is possible to efficiently execute the ontology label enhancement.

It is a block diagram shown about the structural example of the dictionary editing apparatus which concerns on one Example of this invention. It is a figure explaining the example of ontology utilized in the dictionary editing apparatus concerning one Example of this invention. It is a figure explaining the example of ontology utilized in the dictionary editing apparatus concerning one Example of this invention. It is a figure explaining the example of the text corpus utilized in the dictionary editing apparatus concerning one Example of this invention. It is a figure explaining the example of an extraction process of the constraint conditions performed in the dictionary editing apparatus concerning one Example of this invention. It is a figure explaining the example of an extraction process of the constraint conditions performed in the dictionary editing apparatus concerning one Example of this invention. It is a figure explaining the example of an extraction process of the constraint conditions performed in the dictionary editing apparatus concerning one Example of this invention. It is a figure explaining the example of an extraction process of the constraint conditions performed in the dictionary editing apparatus concerning one Example of this invention. It is a figure which shows the flowchart explaining the process sequence performed in the dictionary editing apparatus which concerns on one Example of this invention. It is a figure explaining the hardware structural example of the information processing apparatus which comprises the dictionary editing apparatus which concerns on one Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Dictionary editing apparatus 101 Class selection part 102 Related class extraction part 103 Restriction condition extraction part 104 Related sentence extraction part 105 Related character string extraction part 106 Related character string presentation part 107 Ontology edit processing part 111 Ontology storage part 112 Text corpus storage part 201 Class A
211 to 213 Class 221 to 222 Class 301 Class A
302 Super class 303,304 Sibling class 311 Class A
312 Super class 313 Sibling class 321 Class A
322 Super class 323 Subclass 331 Class A
332,333 Subclass 701 CPU (Central Processing Unit)
702 ROM (Read-Only-Memory)
703 RAM (Random Access Memory)
704 Host bus 705 Bridge 706 External bus 707 Interface 708 Keyboard 709 Pointing device 710 Display 711 HDD (Hard Disk Drive)
712 Drive 714 Connection port 715 Communication unit 721 Removable recording medium 722 Externally connected device 801 Database

Claims (19)

A dictionary editing device that sets a concept representing existence as a class, and executes update processing of dictionary data in which data including class-specific property information and data including class-related attribute information is included. Yes,
A related class extraction unit that extracts a related class of the update target class from the dictionary data;
A constraint condition extraction unit that determines a constraint condition as a search condition of a text corpus based on label information corresponding to a related class;
A related sentence extraction unit that searches a text corpus according to the constraint condition and extracts a related sentence including a candidate label of the update target class;
In accordance with the constraint condition, a related character string extraction unit that extracts a related character string that is a label candidate of the update target class from the related sentence extracted by the related sentence extraction unit;
An edit processing unit that sets the related character string extracted by the related character string extraction unit as a label of the update target class;
A dictionary editing apparatus comprising: The dictionary editing device further includes:
A related character string presenting unit that presents the related character string extracted by the related character string extracting unit to an editing operator;
The edit processing unit
The dictionary editing apparatus according to claim 1, wherein a process of setting the selected related character string as a label of the update target class is executed in response to input of selection information of the related character string by the editing operator. . The dictionary data is dictionary data in which a plurality of classes are set as a hierarchical structure,
The related class extraction unit includes:
As a related class of the update target class, extract at least one of the super class (parent class), sub class (child class), sibling class (sibling class),
The constraint condition extraction unit
The processing for determining the constraint condition is performed based on label information corresponding to two related classes selected from a super class (parent class), a subclass (child class), and a sibling class (sibling class). Item 4. The dictionary editing device according to item 1. The constraint condition extraction unit
Determine a string as a positive constraint and a string as a negative constraint,
The related sentence extraction unit includes:
A process for extracting from the text corpus text that includes a character string of the same type as a character determined as a positive constraint and does not include a character string of the same type as a character determined as a negative constraint. Item 4. The dictionary editing device according to item 3. The constraint condition extraction unit
Positive constraint on the longest common part of the label information corresponding to two related classes selected from the super class (parent class), subclass (child class), sibling class (sibling class) extracted as the related class of the update target class The dictionary editing apparatus according to claim 4, wherein the dictionary editing apparatus is determined as a condition. The constraint condition extraction unit
5. A final constraint condition is determined by executing a process of deleting a negative constraint condition included in a positive constraint condition extracted by referring to a label of a related class of an update target class. The dictionary editing device described in 1. The related character string extraction unit includes:
A continuous character string including a character string as the positive constraint condition is extracted as a related character string to be a label candidate of the update target class from the related sentence extracted by the related sentence extraction unit. Item 5. The dictionary editing device according to item 4. The related character string extraction unit includes:
Performing a morphological analysis of the related sentence extracted by the related sentence extraction unit, and extracting a continuous part of a noun including a character string as the positive constraint condition as a related character string serving as a label candidate of the update target class. 5. The dictionary editing apparatus according to claim 4, wherein   The dictionary editing apparatus according to claim 1, wherein the dictionary is an ontology dictionary configured by registration information of a class having a hierarchical structure. A dictionary editing method for executing dictionary data editing processing in an information processing device,
The dictionary data to be edited is a dictionary data in which a concept representing existence is set as a class, and data including a label that is property information corresponding to a class and data including relationship information between classes that is attribute information corresponding to a class,
A related class extraction unit extracts a related class of the update target class from the dictionary data; and
A constraint condition extraction step in which a constraint condition extraction unit determines a constraint condition as a text corpus search condition based on label information corresponding to the related class;
A related sentence extracting unit performs a search of a text corpus according to the constraint condition, and extracts a related sentence including a label candidate of the update target class; and
A related character string extraction unit that extracts a related character string that becomes a label candidate of the update target class from the related sentence extracted by the related sentence extraction unit according to the constraint condition;
An edit processing unit that sets the related character string extracted in the related character string extraction step as a label of the update target class; and
A dictionary editing method characterized by comprising: The dictionary editing method further includes:
A related character string presentation unit has a related character string presentation step of presenting the related character string extracted in the related character string extraction step to the editing operator;
The editing process step includes:
The dictionary editing method according to claim 10, wherein a process of setting the selected related character string as a label of the update target class is executed in response to input of selection information of the related character string by the editing operator. . The dictionary data is dictionary data in which a plurality of classes are set as a hierarchical structure,
The related class extracting step includes:
As a related class of the update target class, extract at least one of the super class (parent class), sub class (child class), sibling class (sibling class),
The constraint condition extraction step includes:
The processing for determining the constraint condition is performed based on label information corresponding to two related classes selected from a super class (parent class), a subclass (child class), and a sibling class (sibling class). Item 11. The dictionary editing method according to Item 10. The constraint condition extraction step includes:
Determining a string as a positive constraint and a string as a negative constraint;
The related sentence extracting step includes:
A process for extracting from the text corpus text that includes a character string of the same type as a character determined as a positive constraint and does not include a character string of the same type as a character determined as a negative constraint. Item 13. The dictionary editing method according to Item 12. The constraint condition extraction step includes:
Positive constraint on the longest common part of the label information corresponding to two related classes selected from the super class (parent class), subclass (child class), sibling class (sibling class) extracted as the related class of the update target class The dictionary editing method according to claim 13, wherein the dictionary editing method is determined as a condition. The constraint condition extraction step includes:
14. A final constraint condition is determined by executing a process of deleting a negative constraint condition included in a positive constraint condition extracted by referring to a label of a related class of an update target class. The dictionary editing method described in 1. The related character string extraction step includes:
A continuous character string including a character string as the positive constraint condition is extracted as a related character string to be a label candidate of the update target class from the related sentence extracted in the related sentence extracting step. Item 14. The dictionary editing method according to Item 13. The related character string extraction step includes:
Performing morphological analysis of the related sentence extracted in the related sentence extracting step, and extracting a continuous part of nouns including the character string as the positive constraint condition as a related character string serving as a label candidate of the update target class. The dictionary editing method according to claim 13, wherein:   The dictionary editing method according to claim 10, wherein the dictionary is an ontology dictionary configured by registration information of a class having a hierarchical structure. A computer program for executing dictionary data editing processing in an information processing apparatus;
The dictionary data to be edited is a dictionary data in which a concept representing existence is set as a class, and data including a label that is property information corresponding to a class and data including relationship information between classes that is attribute information corresponding to a class,
A related class extraction step for causing the related class extraction unit to extract a related class of the update target class from the dictionary data;
A constraint condition extraction step for causing the constraint condition extraction unit to determine a constraint condition as a text corpus search condition based on the label information corresponding to the related class;
A related sentence extraction unit that performs a search of a text corpus according to the constraint condition and causes a related sentence extraction unit to extract a related sentence including a label candidate of the update target class; and
A related character string extraction step for causing the related character string extraction unit to extract a related character string that is a candidate label for the update target class from the related sentence extracted by the related sentence extraction unit in accordance with the constraint condition;
An editing processing step for causing the editing processing unit to set the related character string extracted in the related character string extracting step as a label of the update target class;
A computer program characterized by comprising:

Images