JP2001195400A - Method and device for structuralizing document context - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the document context structuralizing method and device which can divide the intra-document information for every written object and can describe the inter-attribute relation for identifying the reverting destination of every attribute when every semantic attribute appears in plural numbers in a document semantic structuralization mode. SOLUTION: The attributes are given to the character strings included in an inputted still non-structuralized document. The stile non-structuralized document is decided as a document that is written about plural objects among the sets of attributes when the same attribute name appear in plural numbers. Then the structuralization is carried out in plural contexts.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for structuring a document context, and more particularly, to performing a structuring based on a semantic context in a document, thereby semantically retrieving information considering the context. The present invention relates to a method and apparatus for structuring a document context for integrating and classifying information that is semantically similar.

2. Description of the Related Art In structuring a document, in order to represent a relationship between attributes, a structuring apparatus divides information in a document sometime based on a predetermined rule, and divides information between the divided areas. The relation structure between the relation structure and the attribute in the area is extracted, and the relation between the attributes is expressed.

[0002] When a document is structured with its surface structure, for example, information in the document is divided at the position of the document and the size of the paragraph (large paragraph, middle paragraph, small paragraph) is determined. Is used to show the relational structure between paragraphs, and also to define the relation of attributes to show the relation between attributes in one paragraph.

[0003] When a document is structured by its logical structure, for example, information in the document is divided at the position of a conjunction, and
The logical structure indicated by the conjunction is used to indicate the logical structure of the divided area, and the attribute relation is defined to indicate the relation between the attributes in one paragraph.

[0004]

However, in the above-mentioned conventional method, the logical structure and the formal structure of one document are constant in both the logical structure and the formal structure.
There is only one result structured based on a certain structuring rule, and there is a problem that structuring cannot be performed based on a plurality of semantic contexts of a document.

The present invention has been made in view of the above points, and a plurality of semantic attributes may appear in the semantic structuring of a document.
It is an object of the present invention to provide a document context structuring method and apparatus capable of dividing information in a document for each written object and describing a relationship between attributes.

Another object of the present invention is to divide a document into a plurality of contexts when dividing information in a document, thereby enabling a document having a semantic structuring that enables retrieval in various contexts. It is to provide a context structuring method and apparatus.

[0007]

FIG. 1 is a diagram for explaining the principle of the present invention.

The present invention (Claim 1) performs a structuring based on a semantic context in a document, thereby performing semantic search of information in consideration of the context, integration and classification of semantically similar information. In the document context structuring method for performing the above, an attribute is assigned to the character string in the unstructured document (step 1), and when a plurality of the same attribute names appear in a set of the attributes (step 2) It is determined that the unstructured document is a document in which a plurality of objects are written, and structuring is performed in a plurality of contexts (step 3).

According to the present invention (claim 2), when a plurality of the same attribute names appear, an attribute name sequence is created by arranging the plurality of attribute names in the order in which they appear in the document, and is repeated in the attribute name sequence. One or more attribute patterns are extracted, an attribute name string is divided for each attribute pattern, and the structured information in the document is divided into a plurality of segments.

According to the present invention (claim 3), the attribute pattern
The data is stored in the database in advance in association with the concept (class) of the context, and the database is referred to based on the extracted attribute pattern, and a class is assigned to the document.

According to the present invention (claim 4), a class and a segment are assigned to all documents, a class ID and a segment ID are assigned to all attributes in the document, and each attribute is uniquely identified. Output.

According to the present invention (claim 5), when nesting of a pattern is found from an attribute name string, the result of segmentation according to a plurality of attribute patterns is stored hierarchically, and Then, a document ID and a class ID are assigned, and a hierarchical relationship is output for each individual segment.

According to the present invention (claim 6), an attribute given to information in one document, an attribute name string created by arranging the attributes, and an attribute pattern extracted from the attribute name string are one or more. One or more classes are assigned to one document, and one document is structured in one or more contexts.

FIG. 2 is a diagram showing the principle of the present invention.

The present invention (Claim 7) provides a structure based on a semantic context in a document, thereby performing a semantic search of information in consideration of the context, integration of semantically similar information, and classification. A document context structuring device for inputting an unstructured document, an attribute assigning unit 13 for assigning an attribute to a character string in the unstructured document, and a set of attributes When the same attribute name appears more than once, it is determined that the document is a document written for a plurality of objects,
Pattern dividing means 16 for structuring in a plurality of contexts.

According to the present invention (claim 8), when a plurality of the same attribute names appear, an attribute name sequence is created by arranging the plurality of attribute names in the order in which they appear in the document, and is repeated in the attribute name sequence. Pattern extracting means 14 for extracting one or more attribute patterns. The pattern dividing means 16 divides an attribute name string for each attribute pattern, and divides structured information in a document into a plurality of segments. Have means to do so.

According to the present invention (claim 9), the pattern extracting means 14 has means for storing the attribute pattern in the database 12 in advance in association with the concept (class) of the context. It has means for referring to the database 12 based on the attribute pattern to be given and assigning a class to the document.

According to the present invention (claim 10), the pattern dividing means 16 has means for assigning a class and a segment to all documents and assigning a class ID and a segment ID to all attributes in the document. And an output unit 17 for uniquely identifying and outputting each attribute.

According to the present invention (claim 11), when a pattern nest is found from an attribute name string in the pattern dividing means 16, the result of dividing into segments according to a plurality of attribute patterns is stored hierarchically. Then, it has a means for assigning a document ID and a class ID to each segment, and the output means 17 has a means for outputting a hierarchical relationship for each individual segment.

According to the present invention (claim 12), in the pattern dividing means 16, an attribute given to information in one document, an attribute name string created by arranging attributes, and an attribute extracted from the attribute name string There are one or more patterns, one or more classes assigned to one document, and a means for structuring one document in one or more contexts.

As described above, the present invention enables semantic structuring in structuring within a document. Here, structuring means giving attributes to information in a document and showing a relationship between the attributes. The present invention describes the relationship between attributes.
It shows the relationship between attributes according to the context. As a result, in the semantic structuring of a document, a plurality of each semantic attribute may appear.
It is possible to divide the information in the document for each written object and to describe the relationship between the attributes.

[0022]

DESCRIPTION OF THE PREFERRED EMBODIMENTS First, a plurality of classes are defined as a concept representing a context. Each class has a corresponding semantic attribute. Also, a hierarchical relationship between the classes is defined. Next, a semantic attribute is added to the information in the document. At this time, by assigning semantic attributes corresponding to all possible classes, semantic attributes according to the context can be assigned. That is, a plurality of semantic attributes are given to certain information.

When a plurality of one semantic attribute names appear in one document, this document contains information on a plurality of objects. In order to uniquely identify each of these overlapping attributes, a set of attributes is divided for each target. The divided part is defined as a segment, and a corresponding class is defined from the arrangement pattern of the attributes in the segment. Since a plurality of semantic attributes are assigned to certain information, a plurality of attributes are arranged, so that a plurality of classes are defined. Therefore, a single document can be structured in a plurality of contexts.

Each of the divided portions is provided with a class ID and a segment ID for identifying the divided portion. By writing a class ID and a segment ID together for a certain attribute as a structured result, the attribute in the document can be uniquely represented.

When a segment further includes information on a plurality of objects, the segment is divided into lower segments and a lower class ID and a lower segment ID are assigned.
This makes it possible to determine the hierarchical relationship for each segment. Since each attribute is given a class ID and a segment ID, defining a hierarchical relationship between segments is synonymous with defining a hierarchical relationship between attributes.

[0026]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 3 shows the configuration of a document context structuring apparatus according to an embodiment of the present invention.

The document context structuring apparatus 1 shown in FIG. 1 includes a document input unit 11, a document storage unit 12, an attribute assignment unit 13, a pattern extraction unit 14, a pattern division unit 16, an output unit 17, an attribute dictionary 22, a pattern matcher. It comprises a rule 23 and a class database 24.

The document input section 11 is a section for inputting an unstructured document.

The document storage section 12 stores an unstructured document. The unstructured document is converted after receiving each processing, and stored again.

The attribute assigning unit 13 assigns an attribute to the character string in the unstructured document with reference to the attribute dictionary 22 and the pattern matching rule 23.

The attribute dictionary 22 stores character string attributes (such as names).
Is a dictionary for extracting.

The pattern matching rule 23 is a set of rules for extracting pattern (telephone number, time, etc.) attributes.

The pattern extracting unit 14 arranges the attributes assigned by the attribute assigning unit 13 in the order of appearance, creates an attribute name sequence,
Extract the attribute pattern that is repeated.

The class database 24 stores the correspondence between classes and attribute patterns in the classes in advance. The class stores the correspondence between attribute patterns (arrangement of attributes) and classes. In addition, a hierarchical structure between classes is stored.

The pattern dividing section 16 includes the pattern extracting section 1
4 and the class database 24
This is a part that matches attribute patterns in the, extracts a corresponding class, and divides in-document information to which an attribute is assigned into segments based on the attribute pattern.

The output unit 17 outputs attribute information to which the class information and the segment information extracted by the pattern division unit 16 are added.

Next, the operation in the above configuration will be described.

FIG. 4 is a flowchart showing the operation of the document context structuring apparatus according to one embodiment of the present invention.

Step 101) An unstructured document is input to the document input unit 11, and the unstructured document is stored in the document storage unit 12
To be stored.

Step 102) The attribute assigning unit 13 assigns a semantic attribute to the character string in the document by using the attribute dictionary 22 and the pattern matching rule 23.

For example, by matching with the attribute dictionary 22,
An attribute “shop name” is added to the character string “XX shop”. In the pattern matching rule 23, "XXX-XXX
An attribute “telephone number” is added to the symbol string “−XXXX”. There are a plurality of semantic attributes assigned to a certain character string or symbol string. Assign all possible semantic attributes.

For example, "carrot" is provided with a "produce" attribute, a "material" attribute, a "vegetable" attribute, a "fiction name" attribute, and the like. Since a co-occurrence rule and an exclusion rule are applied between these attributes, the number of a plurality of assigned semantic attributes is narrowed down. For example, if a co-occurrence rule is defined between the value of the "industry" attribute "agriculture" and the "agricultural product" attribute, if the "industry" attribute is "agriculture", the character string "carrot" An attribute of "agricultural product" is given.
Similarly, if a co-occurrence rule is defined between the value of the "industry" attribute "restaurant" and the "material" attribute, if the "industry" attribute is "restaurant", "carrot" An attribute “material” is provided.

Step 103) Next, the pattern extraction unit 14 classifies a case where a plurality of identical semantic attributes appear and a case where one semantic attribute name appears only once.

Step 104) Pattern extractor 14
In step 103, in step 103, a plurality of attribute names are arranged in the order in which they appear in the document, and an attribute name sequence is created. Since a plurality of attributes may be given to certain information, there may be a plurality of attribute continuous patterns. For each attribute continuous pattern, a repeated attribute pattern is extracted.

An example of the extraction method will be described below.

A list of attributes appearing twice or more in the attribute name string is extracted. It is assumed that there is no duplication of attributes in the sequence of attributes. The arrangement may not be adjacent to each other, and unspecified zero or more attributes that do not overlap with the two attributes may appear between the two attributes. For example, a case where unspecified zero or more attributes that do not overlap with the two attributes appearing between the attribute A and the attribute B appears as “A * B”. A sequence of these attributes is defined as an attribute pattern. For example, if the attribute name column is "ABC
In the case of “ABCABCB” (assuming that one character is one attribute name), patterns of two or more adjacent attributes without duplication are “AB”, “BC”, “CA”, “AB”.
C, "BCA", and "CAB", and the patterns that are not adjacent without being overlapped are "A * C", "B * A", and "C *".
Here, * means that 0 or more unspecified attributes appear between them. Therefore, if these are collectively described, "A * C" and "B * A" Since only a specific character string appears, it is absorbed by “ABC” and “BCA”, and the attribute pattern candidates are “AB”, “BC”, “CA”,
“ABC”, “BCA”, “CAB”, and “C * B”. These candidates are sequentially matched from the attribute name sequence, and the pattern that matches the most and has the largest number of attributes is set as the attribute pattern. In the example, “ABC” is the attribute pattern. The number of extracted patterns is not limited to one, and a plurality of candidates may be extracted.

Step 105) Pattern dividing section 16
Means that if multiple candidates are extracted as attribute patterns,
Determine the relationship between each attribute pattern.

For example, “ABCCDCDDABABC
In the case of "D", the attribute patterns "AB" and "CD"
It is assumed that the lower layer pattern is nested with the upper layer pattern. Since the attribute patterns “AB” and “CD” are associated with the classes, respectively, the hierarchical structure is determined. Is a hierarchical relationship.

Step 106) Pattern dividing section 16
Is the extracted attribute pattern and class database 24
The attribute patterns in are matched.

Step 107) As a result of the matching, a corresponding class is extracted. Step 108) The pattern division unit 16 divides the attribute name string at the appearance position of the attribute pattern, and assigns a segment ID to each divided portion. A class ID and a segment ID are assigned to all attributes.

Step 109) The segment to be divided is set as a nested lower segment, a hierarchical relationship is given between the segments, and the process proceeds to step 104. If a plurality of the same semantic attributes appear in the segment, the process proceeds to step 104. The processing of 105 and 106 is further performed.

Step 110) If the same attribute does not appear in Step 103, the pattern dividing unit 1
6 performs matching between the appearing attribute name set and the class database 24 to extract a corresponding class. There is one segment for a class.

Step 111) The output unit 17 outputs a class ID and a segment ID to all the attribute information. For example, as shown in FIGS. 5A and 5B, an XML document is output for each class or stored in a database. In the figure, (A) and (B) show the same document I.
D. In other words, they are based on the same document, but have different structured results. This is because the class information indicating the context is different. Since the structured result changes according to the class information, as shown in (A) and (B),
Different output information is generated.

The output unit 17 outputs the class I to the hierarchical relationship held in step 105 and step 108.
The hierarchical relationship is output for each area identified by D and the segment ID. The output image is stored in a database table as shown in FIG. In the figure, the database stores a child unique segment ID and a parent unique segment ID for one segment ID.

Since the above steps 104 to 111 are performed for all the attribute name strings, the result of structuring one document with a plurality of classes can be output.

Although the above embodiment has been described with reference to the configuration of FIG. 3 and the flowchart of the operation of FIG. 4, the operation of FIG. 4 is constructed as a program and used as a document context structuring apparatus. The present invention can be easily realized by storing it in a portable storage medium such as a disk device connected to a computer, a floppy disk, or a CD-ROM, and installing it when implementing the present invention.

It should be noted that the present invention is not limited to the above-described embodiment, but can be variously modified and applied within the scope of the claims.

[0059]

As described above, according to the present invention, a plurality of each semantic attribute may appear in the semantic structuring of a document, but information in the document is written in order to identify the destination of each attribute. Can be divided for each target, and the relationship between attributes can be described.

Further, according to the present invention, a search in various contexts can be performed by dividing information in a document in a plurality of contexts.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a configuration diagram of a document context structuring apparatus according to an embodiment of the present invention.

FIG. 4 is a flowchart showing an operation of the document context structuring apparatus according to one embodiment of the present invention.

FIG. 5 is an example in the case of outputting a structured document using XML according to an embodiment of the present invention.

FIG. 6 is a diagram illustrating a relationship between a class “area” and a class “shop” according to an embodiment of the present invention.

[Explanation of symbols]

 1 Document Context Structuring Device 11 Document Input Unit, Document Input Unit 12 Database, Document Storage Unit 13 Attribute Assigning Unit, Attribute Assigning Unit 14 Pattern Extracting Unit, Pattern Extracting Unit 16 Pattern Dividing Unit, Pattern Dividing Unit 17 Output Unit, Output Unit 22 Attribute Dictionary 23 Pattern Match Rule 24 Class Database

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Katsumi Takahashi 2-3-1 Otemachi, Chiyoda-ku, Tokyo F-term in Nippon Telegraph and Telephone Corporation (reference) 5B009 QA09 SA14 VA09

Claims (12)

[Claims] 1. A document context structure for performing a semantic search of information considering a context, integrating and classifying semantically similar information by performing structuring based on a semantic context in a document. In the structuring method, an attribute is assigned to a character string in the input unstructured document, and when the same attribute name appears in a set of the attributes, the unstructured document is written for a plurality of objects. A document context structuring method characterized by determining that the document is a document, and structuring in a plurality of contexts. 2. When a plurality of the same attribute names appear, create an attribute name sequence in which the plurality of appearing attribute names are arranged in the order in which the attribute names appear in the document, and at least one attribute name sequence is repeated in the attribute name sequence. The attribute pattern is extracted, the attribute name string is divided for each of the attribute patterns, and the structured information in the document is divided into a plurality of segments.
Document context structuring method of description. 3. The attribute pattern is stored in a database in association with a concept (class) of a context in advance, and the database is referred to based on the extracted attribute pattern to assign a class to a document. Document context structuring method. 4. A class and a segment are assigned to all documents, a class ID and a segment ID are assigned to all attributes in the document, and each attribute is uniquely identified and output. Or the document context structuring method according to 3. 5. When a pattern nest is found from the attribute name string, the result of segmentation is hierarchically retained according to a plurality of attribute patterns, and a document ID and a class ID are assigned to each segment. The document context structuring method according to claim 2, wherein the hierarchical relation is output for each individual segment. 6. An attribute assigned to information in one document, an attribute name string created by arranging the attributes, and one or more attribute patterns extracted from the attribute name string are assigned to one document. Is one or more classes, and for one document,
A structuring in one or more contexts, wherein the structuring is performed in one or more contexts.
3. The document context structuring method according to 3. 7. A document context structure for performing a semantic search of information considering a context, integrating and classifying semantically similar information by performing structuring based on a semantic context in a document. A document input unit for inputting an unstructured document; an attribute assigning unit for assigning an attribute to a character string in the unstructured document; A document context structuring device, comprising: a pattern division unit configured to determine that the document is a document written for a plurality of objects when a plurality of documents appear, and to perform structuring in a plurality of contexts. 8. When a plurality of the same attribute names appear, create an attribute name sequence in which the plurality of attribute names appear in the order in which they appear in the document, and create at least one attribute name sequence repeated in the attribute name sequence. Pattern extracting means for extracting an attribute pattern of the attribute pattern, wherein the pattern dividing means has means for dividing the attribute name string for each of the attribute patterns and dividing structured information in a document into a plurality of segments. The document context structuring device according to claim 7. 9. The pattern extracting means has means for storing an attribute pattern in a database in advance in association with a concept (class) of a context, and the pattern dividing means has a function based on the attribute pattern to be extracted. 9. The document context structuring apparatus according to claim 8, further comprising means for referring to said database and assigning a class to a document. 10. The pattern dividing means has means for assigning a class and a segment to all documents and assigning a class ID and a segment ID to all attributes in the document. 10. The document context structuring apparatus according to claim 7, further comprising an output unit for identifying and outputting. 11. When the pattern nesting is found from the attribute name string, the pattern dividing means hierarchically stores the result of segmenting according to a plurality of attribute patterns, and Document I
9. The document context structuring apparatus according to claim 8, further comprising: means for assigning D and a class ID; and wherein said output means includes means for outputting a hierarchical relationship for each individual segment. 12. The pattern dividing means may include at least one attribute given to information in one document, an attribute name string created by arranging attributes, and one or more attribute patterns extracted from the attribute name string. 10. The method according to claim 7, wherein one or more classes are assigned to one document, and a means for structuring one document in one or more contexts is provided.
Document Context Structuring Device.