Patent Grant
6853992
1. Field of the Invention
The present invention relates to an apparatus for searching a document structure and document contents at high speed from a large number of structured documents, such as SGML documents, registered in a data base. More particularly, the present invention relates to a structured-document search apparatus which has means to convert a query of structure and contents to a Boolean expression which has been used in a conventional full-text search engine, to thereby enable utilization of the high-speed search performance of the full-text search engine.
2. Description of the Related Art
As a result of recent popularization of word processors and development of OCRs (Optical Character Readers), a huge volume of electronic documents have been created and accumulated. As the amount of accumulated documents becomes increasingly huge, demand for searching a necessary document at high speed becomes stronger and stronger.
In order to satisfy such demand, there have been developed a full-text search engine as described in, for example, Japanese Patent Application Laid-Open No. 10-27183 (Data Registration Method and Apparatus) and Japanese Patent Application Laid-Open No. 8-249354 (Word Index, Word-index Creation Apparatus, and Document Search Apparatus). Such a full-text search engine is designed to search the entirety of each document, and has indexes for referring at high speed to documents which include a designated search key. Each of the full-text search engines described in these publications eliminates the necessity of adding to each document keywords for searching and does not cause oversight during searching. However, since the entirety of each document is searched in a uniform or fixed manner, a search request with designation of a document structure cannot be processed.
Meanwhile, with explosive popularization of the Internet, there have been created a large volume of documents each having a structure (hereinafter referred to as “structured documents”), such as HTML (Hypertext Markup Language) documents and XML (Extensible Markup Language) documents. Further, in enterprises, SGML (Standard Generalized Markup Language) documents have been created and accumulated for the document management and re-use of documents. In relation to search of such structured documents, there has been increasing demand for a technique which does not only search the entirety of each structured document uniformly but also enables a user to designate search conditions for each part of each document. In order to satisfy such demand, there have been developed various techniques; e.g., techniques disclosed in Japanese Patent Application Laid-Open No. 11-15843 (SGML Document Search Apparatus and SGML Document Search Method), Japanese Patent Application Laid-Open No. 11-53400 (Structured-Document Search apparatus and Machine-Readable Recording Medium Storing Program), and Japanese Patent Application Laid-Open No. 11-242676 (Structured-Document Registration Method, Search method, and Transportable Medium used Therefor).
Japanese Patent Application Laid-Open No. 11-15843 discloses a technique such that structured documents are registered into a relational data base; and a user is allowed to input a search request by use of SQL, which is a conventional query language for data base query. When such a technique is used, a schema must be defined in advance, and document parts which do not conform to the schema cannot be registered. Further, when a large volume of documents is registered in the data base, the search speed decreases. Therefore, in order to search the contents of documents at high speed, a full-text search engine must be provided separately from the data base.
Japanese Patent Application Laid-Open No. 11-53400 discloses a technique such that a certain region of each document is divided into a plurality of zones; and searching is performed by use of a Boolean expression on the basis of combination of a zone and a keyword. Although this technique can search at high speed the contents of text data included in a certain document part, it does not allow a user to include in search conditions a hierarchical relationship between document parts.
Japanese Patent Application Laid-Open No. 11-242676 discloses a technique which utilizes a structure index obtained through superposition of document parts of the document registered in a data base and a character index in relation to contents of each document. Although this technique requires an index for holding the structures of documents in addition to an index of an ordinary full-text search engine, it can perform searching at high speed under search conditions which include a hierarchical relationship between document parts.
Japanese Patent Application Laid-Open No. 7-56908 (Document Processing Apparatus) and Japanese Patent Application Laid-Open No. 7-319918 (Apparatus for Designating Object to be Subjected to Document Searching) disclose techniques for searching structured documents. Although these publications disclose a method for searching a single structured document, the publications do not disclose a technique adapted to search a specific document from a large volume of structured documents.
The above-described Japanese Patent Application Laid-Open No. 11-242676 discloses a method for searching at high speed under search conditions which include the hierarchical relationship of document parts. However, a hierarchical relationship which can be included in search conditions is limited to a parent-child relationship and a child-grandchild relationship, and the patent publication does not disclose a method which enables a user to include a sibling relationship in search conditions.
A problem which would arise when a sibling relationship between document parts cannot be included in search conditions will be described below.
When searching conditions “SUZUKI HANAKO in SYSTEM DEVELOPMENT DEPT.” are set for searching of such structured documents, the searching conditions are described more specifically such that in a certain document part of <Employee> element, the text data of <Section> element represent “SYSTEM DEVELOPMENT DEPT.” and the text data of <Name> element represent “SUZUKI HANAKO”. In this case, if a sibling relationship between the document parts cannot be included in the search conditions, a user has no choice but to set the search conditions such that the text data of <Section> element represent “SYSTEM DEVELOPMENT DEPT.” and the text data of <Name> element represent “SUZUKI HANAKO”. Therefore, there is a possibility that a search result different from a desired one is obtained.
The above-described Japanese Patent Application Laid-Open No. 11-242676 further discloses a technique for creating a structure index obtained through superposition of structures of structured documents which are to be subjected to searching. In the technique, when the structures of structured documents are superposed, two nodes are regarded to correspond to each other, if the respective upper nodes of the two nodes correspond to each other, the two nodes are of the same element name, and the two nodes are the same in terms of order of appearance in a row of sibling nodes as determined from the head of the row of the sibling nodes with respect to the forward direction. Therefore, the following Document 1 and Document 2 are treated as having completely the same structure and text data.
In other words, although the row of sibling nodes of the same element name is reserved, the order of sibling nodes of different element name is ignored.
Further, in the technique described in Japanese Patent Application Laid-Open No. 11-242676, search conditions are always set to include search keys and structure designation in combination; and this patent publication does not disclose a method in which only structure designation is used as a search condition.
Moreover, in general, when a hierarchical relationship between document parts is retrieved from structured documents accumulated in a large volume, the time required for such retrieval increases with the degree of complexity of the structures of registered documents.
An object of the present invention is to provide means to enable a sibling relationship between document parts to be included in search conditions, in a system which retrieves a necessary document at high speed from structured documents accumulated in a large volume.
Another object of the present invention is to provide means which solves the above-described problems through use of a hierarchical index which reserves an order of sibling nodes of different element name.
Still another object of the present invention is to provide means which enables a user, at the time of searching, to select whether the order of sibling nodes is to be included in search conditions.
Still another object of the present invention is to provide means which enables designation of only a structure as a search condition.
Still another object of the present invention is to provide means to increase the speed of matching of a hierarchical relationship between document parts to thereby increase the search speed of a search system.
Still another object of the present invention is to provide means to easily effect complete-match search and partial-match search in matching of text data.
The present invention solves the above-described problems involved in the conventional methods through provision of a hierarchical index which expresses the structure of each structured document such that the hierarchical relationship among document parts is expressed in a tree structure in which a “meta part” is treated as a single node; a text index in which a character string contained in text data of each “document part” is registered; and search means which receives a user's query in a tree structure and converts it to a Boolean expression. The term “document part” refers to each element part of each structured document, and the term “meta part” refers to each element part which is common among several structured documents.
Specifically, the present invention provides a structured-document search apparatus comprising: a hierarchical index which expresses the structure of each structured document such that the hierarchical relationship among document parts is expressed in a tree structure in which a meta part is treated as a single node; a text index in which is registered correspondence between each search key and a document identifier (document-ID) of a document which includes the search key, the search key including a character string in text data and a part identifier (part-ID) of a meta part; and search means which receives or inputs a user's query in a tree structure (hereinafter referred to as an “query tree”) and refers to the hierarchical index and the text index in order to obtain a document corresponding to the query tree.
The structured-document search apparatus of the present invention enables a user to express in a tree structure a relationship among document parts which serves as a search condition in a certain query, and further enables the user to designate in the tree structure a sibling relationship among document parts. Moreover, when such a sibling relationship is incorporated into the search conditions, the user can designates a sibling relationship with order or a sibling relationship without order. This enables the search means to compare the query tree and the hierarchical index on the basis of a designated one between the sibling relationship with order and the sibling relationship without order.
The search means has processing means to perform processing in the following two steps:
Further, nodes representing meta parts are characterized in that the nodes satisfy requirements in relation to meta parts such that the nodes share a common upper node meta part, the nodes have the same element name and the same occurrence position (we call it “offset”) in a row of sibling parts in the document, and each node has an offset in a row of sibling parts in the document, a link extending from a child node to a parent node, and a link extending to another node having the same element name. This feature enables high-speed searching even when documents have a complicated structure.
Moreover, the structured-document search apparatus of the present invention includes means which is used for creation of the text index and which operates, when a set having a character string in text data and a part-ID is created, in order to create, for each document part, a special search key composed of a special character string and a part-ID and to register the special search key in the text index. This enables easy search of a structure only (search performed under a single search condition of a certain document part being contained).
Furthermore, the structured-document search apparatus of the present invention includes means which is used for creation of the text index and which operates, when a set of a character string in text data and a part-ID is created, in order to create, for each text data set, pseudo text data which are obtained through addition of special character strings at the start and end position of the text data, and to create a search key from the pseudo text data to register the search key in the text index. This enables both of complete-match search and partial-match search.
The present invention greatly differs from conventional structured-document search apparatus in that provision of the search means—which receives a user's query in a tree structure and searches a document with reference to the hierarchical index and the text index—enables a user to include a sibling relationship among document parts into search conditions. Further, use of a hierarchical index which stores an order of sibling nodes of different element name enables a user to include a sibling order relationship in search conditions.
These, together with other objects and advantages which will be subsequently apparent, reside in the details of construction and operation as more fully described and claimed hereinafter, with reference to the accompanying drawings, wherein like numerals refer to like parts throughout.
An embodiment of the present invention will now be described.
At the time of document registration, a document-structure analyzing unit 11 analyzes the structures of structured documents 20; and a hierarchical-index registering unit 12 creates a hierarchical index 13. At the same time, a text-index registering unit 14 creates a text index 15.
At the time of document search, the following processing is performed. That is, a query-accepting unit 16 accepts a query from a user via an interface 21 and creates a query tree on the basis of the accepted query. Further, while referring to the hierarchical index 13 via a hierarchical-index referring unit 18, a query-converting unit 17 collates the structure of the query tree with those registered in the hierarchical index 13, and adds temporary nodes to matched portions. By use of the temporary nodes, the query-converting unit 17 creates a Boolean logic tree, and then creates a Boolean expression (character string) from the logic tree. On the basis of the created Boolean expression, a text-index referring unit 19 obtains the document-ID of a document corresponding to the query and returns a search result to the user.
If necessary, a document management engine 22 may be disposed near the structured-document search apparatus 1. At the time of document registration, the structured documents 20 themselves are stored in a structured-document storage data base (DB) 23, and at the time of document search, a necessary portion of a searched document is extracted and returned to the user. At this time, the document management engine 22 receives a query of the user from the query-accepting unit 16, receives a search result from the text-index referring unit 19, and extracts the necessary portion of the searched document and returns it to the user. Hereinbelow, the respective means will be described in more detail.
Hierarchical Index
In the present invention, in order to increase the speed of matching of hierarchical relationships of document parts, in addition to an index which has been used in conventional full-text search engines, the hierarchical index 13 is provided. In the hierarchical index 13, the structures of the structured documents 20 subjected to searching are expressed in a tree structure.
In the hierarchical index 13, documents parts of each structured document 20 having a common structure are expressed in the form of a single node as a meta part, and a hierarchical relationship between meta parts is represented in the form of a tree structure.
As shown in
At the time of registration of structured documents, a root node of the hierarchical index 13 is first created (step S11), and the steps described below are repeated until the processing has been performed for all the documents (step S12).
First, the structure of a presently-selected (i.e., presently-considered) document to be registered is analyzed (step S13), and when the processing in step S15 to S19 has been performed for all the document parts of the selected document, the processing returns to step 12, and the same processing is repeated for a document to be registered next (step S14).
In step S15, processing for judging whether a presently-selected document part of the selected document is a meta part is performed. When the selected document part is identified as a new meta part (step S16), a new part-ID is issued (step S17) and is registered in the hierarchical index 13 (step S18). Subsequently, registration into the text index 15 is performed (step S19). When the selected document part is identified as not being a new meta part, without registration in the hierarchical index 13, registration into the text index 15 is performed (step S19).
The order in which the documents parts undergo the processing in step S14 corresponds to the order in which corresponding start tags appear in the selected document. This means that when the structure of a structured document is expressed in the form of a tree structure, respective document parts are processed while the priority is given their depth.
When the selected document part has not yet been registered in the hierarchical index 13 (step S21), the selected document part is regarded to be a new meta part (step S25) and is registered in the hierarchical index 13 (steps S17 and S18 of FIG. 3). When the selected document part has already been registered in the hierarchical index 13, it is judged that the same meta part has already been registered (step S26), and, without registration in the hierarchical index 13, the registration of text data in the text index 15 is performed (step S19 of FIG. 3).
The processing of extending an element link described in the above stage 4) is performed by means of the following steps.
The text index 15 is used to obtain a document-ID of a document including a search key, on the basis of the search key which includes a character string to be searched (hereinafter referred to as a “search character string”) and a part-ID in combination. The text index 15 assumes the same form as that of indices which are used in conventional full-text search engines, except that a search character string and a part-ID form a single search key set. That is, any of index forms, such as an inverted file, a signature file, and a bit-map file, which are used in conventional full-text search engines, may be used as a form of the index for obtaining a document-ID from a search key.
In the present embodiment, a set (special search key) which includes a special character string and a part-ID is registered for each document part, which enables searching of a structure only (search which is performed under a single search condition that a certain document part is present).
Further, pseudo text data are prepared from text data through addition of a special character string at the start and end positions of the text data; and characteristic elements serving as minimal units for searching are extracted from the pseudo text data. At the time of search, a set of “(special character string at the start position)+(search character string)+(special character string at the end position)” and a part-ID is used as a search key. Thus, a complete-match search of text data is realized. The term “characteristic element” refers to the smallest piece of information that can be searched for. Morphological analysis, N-gram division, or any other suitable method may be used to divide text data into characteristic elements.
This registration processing includes the following four stages.
Each of the special character string for representing the presence of a structure and the special character strings which are inserted at the start and end positions of text data may be a character string (a row of codes other than character codes) which does not appear in text data.
Processing for Conversion of a Query Tree to Boolean Expression at the Time of Search
The user's query may be created by use of an object-oriented query language, a query language for XML documents, which is currently standardized by W3C (World Wide Web Consortium), or any other suitable language.
Since the steps of creating a query tree vary depending on a query language to be used, here, only an example structure of a query tree is shown in FIGS. 8A. and 8B.
When the order of parts serving as children is not considered, the conditions that a “Part1” is present and that a “Part2” follows the “Part1” is replaced with the condition that the “Part1” and “Part2” are present.
Such a query tree is used for matching of a document in which nodes representing document parts form a tree structure, and which has, as a partial tree, a tree structure similar to the query tree. As shown in
Within the full-text search engine, a Boolean expression, which is a character string, is analyzed in order to create a Boolean logic tree. Therefore, when the form of the Boolean logic tree created in step S62 shown in
Each temporary node serves as a node of a Boolean logic tree. Such a temporary node is created to have a node type (AND, OR, the type of a search key), a text-data matching condition, and a part-ID. A node whose type is AND or OR is used as an intermediate node, and ultimately, a single logic tree is assembled in step S62 of FIG. 9.
The processing performed for the temporary nodes described in the above stage 2) includes the following steps.
Through the above-described processing, a Boolean logic tree is assembled. The thus-assembled logic tree is converted into a character string by means of the processing in step S63 shown in FIG. 9. In this conversion processing, the Boolean logic tree is gone around recursively while priority is given in the hierarchical direction; and child nodes of an AND node are output in the form of an AND expression, and child nodes of an OR node are output in the form of an OR expression. Further, the output of child nodes of each node is enclosed in parentheses. An example of a recursive pseudo program for performing the above conversion is shown below.
The steps of the processing for referencing the text index 15 by use of a Boolean expression obtained after the conversion to the Boolean expression are the same as those of processing performed in the full-text search engine, except that a set having a text-data matching condition and a part-ID is used as a search key.
The system shown in
The system comprises a hierarchical index expressing a structure of each structured document; a text index used for referring to a document-ID of a document including a search key, on the basis of the search key composed of a part-ID and a character string in text data; means to receive or input a user's query in the form of a tree structure and for converting the query to a Boolean expression; and means to refer to the text index by use of the Boolean expression and for obtaining a document-ID corresponding to the query tree. Thus, the system can search structured documents while maintaining the high speed of conventional full-text search engines.
Since the processing steps shown in
Since nodes having the same element name are obtained from the hierarchical index by use of an element link in step S73 shown in
The system has means to register, for each document part, a special search key composed of a special character string and a part-ID in the text index. Thus, it becomes possible to perform search under search conditions including only the presence of a document part or parts, without changing the basic configuration of the conventional search engine.
The system includes means to create, for the text data of each document part, pseudo text data which are obtained through addition of special character strings at the start and end positions of the text data and for creating search keys from the pseudo text data in order to register the search keys in the text index. Thus, it becomes possible to select one of two matching means to match the text data; i.e., means to perfect-match the text data and means to partial-match the text data.
Next, an example in relation to documents having simple structures will be described. Documents 1 and 2, which were like those used in the “Description of the Related Art” section, are used here.
The example shows documents 1 and 2 which include the Chinese (or Kanji) character, which is ideogram, more concretely shows Japanese documents 1 and 2. In the example, the word or the characters “” corresponds to the word “structuring”, the word or the characters “” to the word “document”, and the word or the characters “” to the word “retrieval”. That is, the above example is a Japanese one used in the “Description of the Related Art” section. Since the Chinese character is a ideogram, two continuous Chinese characters “” has the same meaning with the word “document” composed of eight continuous characters, for example. That is, even bigram (two continuous characters) is effective as a characteristic element in such as the documents 1 and 2.
(1) Document Registration
Example processing for registration of documents having simple structures will be described below. First, registration into the hierarchical index 13 will be described with reference to
Subsequently, “Part2,” which is the first child of “Document,” is judged to be a different meta part and is registered newly, because although a node which is identical in terms of upper meta part and element name has been registered, “Part2” differs from the registered node in terms of offset in a row of sibling parts. Further, in step S38 shown in
The processing for registering the above-described two documents in the text index 15 will be described with reference to FIG. 14. The example registration of documents in the text index shown in
The inverted file form includes search keys 51 and document identifier lists 52, each of which is a list of the document-IDs of documents including the corresponding search key. A characteristic element and a part-ID of a document part in which the characteristic element is present are registered in the search key 51. In the document identifier list 52, the document-IDs of documents including each search key are registered in a row. If necessary, offsets and the number of times of appearance in each document are registered.
A dummy value (−1 in the present embodiment) is set for the offset corresponding to the special key “??” for expressing the presence of a structure. A characteristic element “” “^” and “$” is for the word “” which has the meaning of “document”. A characteristic element “”, “^” and “$” is for the word “” which has the meaning of “retrieval”. A characteristic element “” “” and “^” and “$” is for the word “” which has the meaning of “structuring”.
The characteristic element is extracted in the form other than bigram (two continuous characters) which is shown in FIG. 14. The number of the character which comprises the characteristic element may be more than 3, or, for example, 3 to 6. The number of the character depends on the language of the document which is registered.
When the document which is registered is written by Japanese, the character number of the characteristic element is usually 1 or 2. This is owing to that Japanese character is ideogram which has a meaning with single character. Chinese which using Kanji character and other languages using ideogram are in a similar situation.
When the document which is registered is written by English, single word is usually used as single characteristic element. This is owing to that words are written in a divided form with blanks in English, and that English character is phonogram which does not have a meaning with single character, so that the N-gram division is not used in most cases. The languages belonging to the Indo-European family of language such as English using alphabet which is ideogram are in a similar situation.
(2) Search
Example processing for search of documents having simple structures will be described below. The following five search requests; i.e., Search Request 1 to Search Request #5, are taken as examples.
Search Request 1
In Search Request 1, a document including “Part1” as a document part is searched. An internal operation for converting Search Request 1 to a Boolean expression will be described with reference to
The text index shown in
Search Request 2
In Search Request 2, a document whose “Part3” includes the Chinese character “” which has the meaning of “retrieval”, as a portion of its text data is searched (partial match). An internal operation for converting Search Request 2 to a Boolean expression will be described with reference to
The text index shown in
Search Request 3
In Search Request 3, a document whose “Part3” includes the Chinese character “” which has the meaning of “retrieval”, as its text data is searched (complete match). Searching performed under the condition of complete match of text data is realized through processing which is identical with that for the above-described Search Request 2, except that the text-data matching condition of the query tree becomes “^$”, and the following Boolean expression is obtained ultimately.
(^$@3)
When the search character string is divided into a plurality of characteristic elements as described above, a mere reference of the text index shown in
In the above-described Boolean expression,
In Search Request 4, a document which includes “Part1” and “Part2” as children of “Document” and in which the text data of “Part1” includes the Chinese character “” which has the meaning of “structure”, is searched (search of a sibling relationship between document parts without conditions in relation to the order of siblings).
An internal operation for converting Search Request 4 to a Boolean expression will be described with reference to
For these nodes, the processing shown in
The text index shown in
Search Request 5
In Search Request 5, a document which includes “Part1” as a child of “Document” and in which “Part2” follows and the text data of “Part1” includes the Chinese character “” which has the meaning of “structure”, is searched (search of a sibling relationship between document parts with conditions in relation to the order of siblings).
Search performed under search conditions which include not only a sibling relationship between document parts but also the order of siblings is realized through addition of the processing in step S87 shown in
The text index shown in
Next, processing for conversion into a Boolean expression of a query tree including designation of a multi-level sibling relationship will be described. Even when documents having more complex structures are processed, the fundamental processing is the same as in the above-described case in which documents having simple structures are processed. Only the structure of the hierarchical index and the internal operation for conversion of a query tree to a Boolean expression become complex. An embodiment for documents having complex structures will be described with reference to an example case in which the following three documents (Document 1, Document 2, and Document 3) are processed.
Among the above example documents, Document 2 includes repeated elements, and Document 3 includes a deficiency of an element.
An internal operation for converting each of the following search requests to a Boolean expression will be described with reference to
Search Request 6
In Search Request 6, a document which is written by “Author” who is defined such that “Name” is “YAMADA TARO” and “Section” is “SHIBUYA RESEARCH CENTER” and in which “Title” includes “STRUCTURED DOCUMENT” is searched. The sibling order relationship between document parts is ignored.
AND (“STRUCTURED DOCUMENT”@1 OR “STRUCTURED DOCUMENT”@10))
The text index shown in
Search Request 7
In Search Request 7, the search conditions used in Search Request 6 are modified to include the sibling order relationship among document parts. That is, a document which is written by “Author” who is defined such that “Name” is ““YAMADA TARO”” and “Section” is ““SHIBUYA RESEARCH CENTER”” and in which “Title” includes ““STRUCTURED DOCUMENT””, “Name” precedes “Section,” and “Author” precedes “Title” is searched.
The text index shown in
As described above, in the present invention, there are provided a hierarchical index which expresses the structure of each structured document, and a text index which is used for searching a document-ID on the basis of a search key composed of a part-ID and a character string in text data; a user's query received in the form of a tree structure is converted to a Boolean expression; and the text index is referred to in order to obtain a document-ID corresponding to the query tree. Thus, it becomes possible to search structured documents, while maintaining the high speed of conventional full-text search engines.
Further, since the hierarchical index stores a sibling relationship between document parts, not only a parent-child relationship but also a sibling relationship can be specified in search conditions. This enables obtainment of search results of higher accuracy as compared with the case of a conventional system.
The many features and advantages of the invention are apparent from the detailed specification and, thus, the appended claims are intended to cover all such features and advantages of the invention which fall within the true spirit and scope of the invention. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not intended that the invention be limited to the exact construction and operation illustrated and described, and accordingly all suitable modifications and equivalents may be resorted to, so long as they fall within the scope of the invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 11-353857 | Dec 1999 | JP | national |
The contents of Japanese Patent Application No. 353857/1999, filed Dec. 14, 1999 in Japan, is incorporated herein by reference.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5706496 | Noguchi et al. | Jan 1998 | A |
| 5727195 | Nakatsuyama | Mar 1998 | A |
| 5870739 | Davis et al. | Feb 1999 | A |
| 5893109 | DeRose et al. | Apr 1999 | A |
| 6377946 | Okamoto et al. | Apr 2002 | B1 |
| 6457017 | Watkins et al. | Sep 2002 | B2 |
| 6510425 | Okamoto et al. | Jan 2003 | B1 |
| Number | Date | Country |
|---|---|---|
| 7-46908 | Mar 1995 | JP |
| 07-056908 | Mar 1995 | JP |
| 7-319918 | Dec 1995 | JP |
| 8-249354 | Sep 1996 | JP |
| 10-27183 | Jan 1998 | JP |
| 10-027183 | Jan 1998 | JP |
| 11-242676 | Sep 1998 | JP |
| 11-15843 | Jan 1999 | JP |
| 11-015843 | Jan 1999 | JP |
| 11-053400 | Feb 1999 | JP |
| Number | Date | Country | |
|---|---|---|---|
| 20010007987 A1 | Jul 2001 | US |
