JPH04123264A - Device for preparing relational word table and device for retrieving document - Google Patents

Abstract

PURPOSE:To efficiently collect relational word data by preparing a means for forming/updating the contents of a relational word table stored in a storage device based upon relational word data extracted from a meaning expression table. CONSTITUTION:Respective sentences of a document inputted by an input device 1 are successively analyzed to obtain meaning expression. The obtained meaning expression is registered as a meaning expression table, respective relational word extraction rules are successively inputted and whether structure coincident with the structure of the condition part of each rule exists in the meaning expression or not is checked. When a coincident part exists, whether key words of an operation part in a relational word extraction rule table have been already registered in the key words 91 of the relational word table or not is checked. When the key words have been registered, no operation is executed, but in the case of no registration, relational word data 92 are added to the relational word table and the processing is repeated until the processing of all read sentences is ended. Thus, relational word extraction appropriate for each user can be efficiently executed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a device for creating a related word table for keywords or key phrases, and a document search device using keywords, key phrases, or the like.

[Prior Art] Conventionally, attempts have been made to improve document retrieval accuracy and document analysis accuracy using related words such as synonyms, generic terms, and same-field words.

However, as seen in Japanese Patent Application Laid-Open No. 2-41565, even though there are tools for maintaining related word data, the content of the related word data itself has generally been collected manually.

Furthermore, as seen in the Information Processing Society of Japan's 34th National Conference IX-6, a method has been proposed for automatically constructing a thesaurus from the meanings and sentences of commercially available dictionaries, but the thesaurus is a general-purpose one. , it was not possible to create related word data according to a specific field or define related words according to the user.

[Problems to be Solved by the Invention] The method of manually inputting all related word data is inefficient and difficult to collect sufficient data. Furthermore, the related word data may differ depending on the user's field of use and purpose of use, so the effect of creating a general-purpose data is not great.

The purpose of the present invention is to automatically extract related word data from a document and improve the efficiency of related word data collection. The purpose is to make it possible to collect data.

[Means for solving the problem] In order to achieve the above purpose, a means is provided to pick up related words from syntactic information such as parallel structure in a document or syntax including specific words and phrases, and register them in a related word table. Furthermore, in order to achieve the above purpose, a means for registering related word extraction rules is provided. Furthermore, the document search device is combined with means for registering in the related word table.

[Look for a specific predetermined syntax from the syntactic analysis results of the action document, identify related words,
Register in the related word table. By using sentence analysis processing technology, a user can also specify a specific syntax using a natural language sentence. Furthermore, even during document searching, related word data is automatically extracted from the searched document and related word data is collected.

[Example] Hereinafter, an example in which the present invention is applied to a Japanese document will be described in detail with reference to the drawings.

FIG. 1 is a flowchart of related word table creation processing.

FIG. 2 is a block diagram of the related word table creation device. l
is an input device, which performs a process of reading a document from which related words are to be extracted. 2 is a central processing unit that extracts related words and creates a related word table. 3 is an output device, which outputs the processing results if necessary. 4 is a storage device that stores various tables.

FIG. 3 is an example of a meaning expression. A semantic expression is represented by a directed graph consisting of nodes and arcs.

FIG. 4 is an example of a semantic expression table consisting of a plurality of records storing the semantic expressions shown in FIG. A record is composed of a node number 41, a word 42, a child node number 43, and a relation 44.

FIG. 5 is an example of a related word table (already created) that stores related word data, and is made up of a plurality of records. The record contains key words 51. Related words 52,
It is composed of appearance frequency 53.

FIG. 6 is a related word table that is an updated version of the related word table shown in FIG.

Next, the related word table creation process of the present invention will be explained according to the flowchart shown in FIG. First, a document is read from an input device (100). Each read sentence is analyzed in turn (101) to obtain a semantic representation. Note that since this is already a known matter, the explanation will be omitted. For semantic representation, a directed graph consisting of nodes representing concepts and relations representing relationships is used. The obtained semantic expression is registered as a semantic expression table (102). From the meaning expression,
A search is made for a relation that means parallel (103), and all words that are parallel are extracted (104). For each parallel word, it is checked whether the word has already been registered in the key word field 51 of the related word table shown in FIG. 5 (105). If it is not registered, a record having the parallel word as a key word is added to the related word table (106). If it is registered, do nothing. Next, it is checked whether other parallel elements in the same row as the parallel word are registered in the related word tank 52 of the related word table (107). If it is registered, nothing is done, but if it is not registered, the parallel element is registered in the related word tank of the related word table (108). Furthermore, 1 is added to the value in the related word appearance frequency column 53 of the related word table (109). The processing from 105 to 1.09 is performed for all parallel words (110). After processing all parallel words, 1 for all sentences read.
Check whether processing from 01 to 110 has been completed,
The processes from 101 to 110 are repeated until all the sentences read have been processed (111).

As an example, [Operations such as search, transcription, aggregation, and analysis can be performed easily. Consider the case of reading the sentence J. The state of the related word table before the document is read is as shown in FIG. Moreover, the analysis result of this sentence is as shown in FIG. From this, the parallel elements ``search, transcription, aggregation, analysis'' connected by the relation ``parallel'' are extracted. As shown in the related word table in Figure 5, there is already a record for "Search", so the related word for "Search" among "Transfer J", "Aggregation", and "Analysis" ``1 posting not registered as `` and ``totalization'' are added to the key word column 41. Since there are no records for "transcription,""aggregation," and "analysis," records are added to the related word table, and three words for each are written in the related word column 52. In addition, 1 is added to the 8-current frequency of all related words. The related word table after the processing path T is as shown in FIG.

When extracting related words, in addition to processing using parallel elements, processing that refers to specific syntax that includes specific words can be considered.

FIG. 8 shows the contents of a table of related word extraction rules, which consists of a plurality of records. The table of related word extraction rules is composed of a condition part and an operation part. Furthermore, the condition part is composed of a node number 81, a word 82, a child node number 83, and a relation 84. The operation unit also includes key words 85, related words 86, and related types 87 (
A plurality of tables are prepared for this related word extraction rule, which are composed of synonyms, L rank, and same field).

FIG. 9 is a related word table of a different format from the related word tables of FIGS. 5 and 6 that store related word data, and is composed of a plurality of records. The record contains key word 91. It consists of related words 92 and related types 93.

This process will be explained according to the flowchart shown in FIG. First, a text is read from an input device (700).

Each read sentence is analyzed in turn (701) to obtain a semantic representation. The obtained semantic expression is registered as a semantic expression table (702). Load related word extraction rules in order (7
03), it is checked whether there is a structure in the semantic expression that matches the structure of the conditional part (704). If there is a matching part, it is checked whether the key word of the operation unit in the table of related word extraction rules has already been registered in the key word 91 of the related word table of FIG. 9 (705). If not registered, a record having the key word of the operation unit as a key word is added to the related word table (706). If it is registered, do nothing.

Next, it is checked whether the contents of the related word data shown in the operation section in the table of related word extraction rules have already been registered in the related word table of FIG. 9 (707). If it is registered, nothing is done, but if it is not registered, the related word data 92 is added to the related word table (708).
. Furthermore, it is checked whether the processing up to now has been completed for all the sentences read, and the processing is repeated until the processing is finished for all the sentences read (709).

The related word extraction rules described above can also be registered manually.

FIG. 11 is an example of a rule input screen. The screen has a condition section input field and an operation section manual field. The operation unit input field further includes a key word input field, a related field, and a related type input field. FIG. 12 is an example of a record that stores manually-written rules.

The record includes a condition part 1201 and an operation part key word 1202
, an operation unit related term 1203, and an operation unit related type 1204.

The registration process will be explained according to flowchart 1 shown in FIG. First, the content of the related word extraction rule is input manually. The phrase or sentence that forms the basis of the conditional part is input in natural language. In the operation section, enter the key word and related words in text, and enter the number for the related type (
1001). The input phrase or sentence in the conditional part is parsed (1002), and the result of the parsing is written in the conditional part in the related word extraction rule table (1003). The input operation section is written as is into the operation section of the related word extraction rule (1004).

For example, when the syntactic analysis result of the phrase "Structure of A1 including A2" is input to the conditional part, the operation part determines that the key word is A1.
, the related word is A2, and the related type is specified as [Top J]. The result of reading the rules is the same as the state shown in FIG.

By combining the related word extraction process described above with a document search system, highly accurate searches are possible.

FIG. 14 is an example of a word decomposition result table that stores word decomposition results, and is composed of a plurality of records. The record consists of a word column 1401. FIG. 15 is an example of a search key related word table that stores search key related words, and is made up of a plurality of records.

The record consists of a related word tank 1501.

The search process will be explained according to the flowchart shown in FIG. First, a user inputs a search key from an input device (1301). The search key may be a compound word. The input key is divided into words (1302), and the division results are sequentially registered in the word column 1401 of the word decomposition result table. For each divided word, the related word table is searched to see if there is a record with this word as a key (1303).

If there is a record, read the word in the related word tank (1
304), the word of the search key is changed into a related word and is sequentially registered in the related word tank 1501 of the search key related word table that stores the related words of the search key (1305). Next, the documents to be searched are read one by one (1306), and it is checked whether the words in the key word and related word tank 1501 exist (1307). If the document exists, the document is output (1.308). Furthermore, in exactly the same way as the process described in FIG. 1, related word data is extracted from the search target document and the related word table is updated (1309).

As a result, the updated related word table can be used the next time a document is searched, and a more accurate document search can be performed. The above process is repeated for all documents (1310).

For example, when you enter "data search" as a search key, using the related word table shown in Figure 6, "data transcription", "data aggregation", and "data analysis" are recognized as related words, and these words are You can retrieve all documents containing .

[Effects of the Invention] According to the present invention, related word data can be automatically extracted from natural language sentences. Furthermore, by automatically creating a table of related word extraction rules based on the user's input of phrases, sentences, etc., and using this created table, related word extraction suitable for the user can be efficiently performed.

Furthermore, by combining this related word extraction process with a document search system, highly accurate searches become possible.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing one embodiment of related word table creation processing according to the present invention, FIG. 2 is a block diagram showing the basic configuration of the device of the present invention, and FIG. 3 is a diagram showing an example of semantic expression.
FIG. 4 is an example of a semantic expression table that stores semantic expressions, FIGS. 5 and 6 are examples of related word tables that store related word data, and FIG. 7 is a related word table creation process according to the present invention. FIG. 8 is an example of a related word extraction rule table, FIG. 9 is an example of a related word table that stores related word data, and FIG. 10 is a related word extraction rule according to the present invention. A flowchart showing an example of table creation processing, FIG. 11 is a related word extraction rule input screen,
FIG. 12 is an example of an input record for related word extraction rules. FIG. 13 is a flowchart showing an embodiment of search processing according to the present invention, FIG. 14 is an example of a word division result table, and FIG. 15 is an example of a search key related word table. 1...Input device, 2...Processing device, 3
...Output device, 4...Storage device Patent applicant Hitachi, Ltd.

Claims (1)

[Scope of Claims] 1. A device for extracting related word data from a natural language sentence to create a related word table, comprising an input device, a processing device, a storage device, and an output device, the processing device means for analyzing a natural language sentence input from an input device to obtain a semantic expression and creating a semantic expression table on the storage device; extracting related word data from the semantic expression table; and extracting related word data from the semantic expression table; A related word table creation device comprising a related word table creation and updating means for creating and updating the contents of the related word table on the storage device. 2. The related word table creation device according to claim 1, wherein the related word table creation and updating means extracts each parallel element in a parallel syntax of a natural language sentence as related word data. 3. The related word table creation and updating means associates the union of two related word data when related word data extracted from a natural language sentence and related word data extracted from another natural language sentence contain a common word. 2. The related word table creation device according to claim 1, wherein the related word table creation device registers the related word table in a word table. 4. The related word table creation device according to claim 1, wherein the related word table is made up of a plurality of records, and each record is made up of a key word, related word data, and appearance frequency. 5. A device for extracting related word data from a natural language sentence to create a related word table, comprising an input device, a processing device, a storage device, and an output device, and the processing device receives input from the input device. means for analyzing a natural language sentence to obtain a semantic expression and creating a semantic expression table on the storage device; and a structure registered in a condition section in a table of related word extraction rules stored in the storage device. is present in the meaning expression, and if so, extracts the key word and related word data registered in the operation section of the related word extraction rule table, and extracts the key word and related word data registered in the operation section of the related word extraction rule table A related word table creation device comprising related word table creation and updating means for creating and updating the contents of the related word table on the storage device based on related word data. 6. The related word table creation device according to claim 4, wherein a plurality of tables of the related word extraction rules are registered in advance in the storage device. 7. The related word table creation and updating means is characterized in that when the extracted key word is not registered in the related word table, a record having the key word as a key word is additionally registered in the related word table. A related term table creation device according to claim 4. 8. A patent characterized in that the related word table creation and updating means additionally registers the extracted related word data in the related word table when the extracted related word data is not registered in the related word table. A related term table creation device according to claim 4. 9. A device for creating a table of related word extraction rules consisting of a condition part and an operation part from a natural language sentence, comprising an input device, a processing device, a storage device, and an output device, and the processing device is configured to extract information from the input device. The input natural language sentence is analyzed to obtain a semantic expression, and the semantic expression is written into the condition section of the table on the storage device, and the data input from the input device is transferred to the operation section of the table. 1. A table creation device for related word extraction rules, characterized in that the device includes means for writing in a related word extraction rule. 10. A document search device for searching documents related to an input keyword, comprising an input device, a processing device, a storage device, and an output device, and the processing device searches a related word table on the storage device using the input keyword. means for searching and registering the search results as a search key related word table on the storage device; and means for sequentially reading search target documents such that the input keyword or related word data on the search key related word table is present in the document. means for inspecting whether or not the document exists and outputting the document when it exists; means for analyzing the sequentially read search target documents to obtain a semantic expression and creating a semantic expression table on the storage device; extracting related word data from the meaning expression table and creating the contents of the related word table on the storage device using the extracted related word data;
A document retrieval device comprising a means for creating and updating a related word table to be updated. 11. The processing device includes means for dividing the compound word into words when the input keyword is a compound word and registering the compound word into words on the storage device as a search key related word table; 12. The document retrieval device according to claim 11, further comprising means for searching the above related word table and registering the search results as a search key related word table on the storage device.