JPH0557624B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a document data retrieval device in a document data file created based on a relational model.

[Technical background of the invention and its problems]

In information retrieval systems, relational models are increasingly being used as models for constructing data files to be searched because of their ease of handling data.

Taking a document data file as an example, the relational model is expressed in a table format as shown in FIG. here,
"Classification", "creation date", "keyword"... are called attributes, and "budget", "830220", "calculator"... are called their values (attribute values).

One horizontal row of this table is called a tuple (or record), and this table is called a relation. In this example, this will be called the main relation.

In this table, the document with "document number 1" has a classification of budget, a creation date of 830220, and a keyword of "calculator".
It shows that... (The same applies to other documents.) Now, let's consider the attribute "keyword" in the example of FIG. 1. In general, there is often not just one keyword attached to a document, but multiple keywords.

However, in the relational model, only an attribute value can be set for one attribute for one tuple. This is insufficient as a function of the document data file.

In order to solve this problem using conventional technology, the following two methods can be considered.

a. Set attributes for the number of keywords you want to set. The relation in this case is as shown in FIG.

b. Create a keyword-only relation separate from the main relation.

In a, when using a data file, it is necessary to decide which keyword value to set in which keyword column. This is a heavy burden on the user.

Furthermore, for the user, it may be handled superficially as shown in FIG. 3, and within the data file management device, it may be handled as shown in FIG. 2.

However, in this case, when performing a search to retrieve the document number of a document whose keyword is A (keyword = A when written in the formula), for "keyword 1", "keyword 2"... "keyword n", Keyword 1=A, Keyword 2=
A... Perform a search with keyword n=A,
Finally, we will take R of all these results.

At this time, it is common to create index tables corresponding to each attribute, store these on a magnetic disk, and access them. Therefore, in this case, n disk accesses occur.

In this way, in this method, We need to create n index tables.

B. The disk is accessed at least n times and it takes a long time to search.

There is a drawback.

In method b, one more thing can be done in addition to the main relation. Therefore, two relations must be handled for one data (document), which is inconvenient.

[Purpose of the invention]

The present invention provides a document data retrieval device that eliminates the drawbacks of the prior art in a data file using a relational model and can set a plurality of values (attribute values) for one attribute.

[Summary of the Invention] The present invention provides a method for inputting an attribute value of desired document data from a plurality of document data in which a document number is assigned to each document data, and outputting a document number having the attribute value. In the document data search device for searching the desired document data, a plurality of attributes of the document data corresponding to the document number and one attribute value are associated with each of these attributes and stored together with the document number. A single-value index table, a multi-value index table that is stored together with the document number in association with a plurality of attribute values, and the type of attribute value, whether single-value or multi-value, that each attribute of the document data has; The attribute information storage means stores the number, and the attribute information storage means stores the type and number of attribute values that each attribute has, and performs a search from the single-value index table or the multi-value index table. an indexing means for extracting a document number having an attribute value corresponding to an attribute value for one document data, and indexing means for extracting a document number having an attribute value corresponding to an attribute value for one document data. Document data retrieval characterized in that the document data is stored in correspondence with the document number of the data, and the document number having the attribute value corresponding to the attribute value for search is retrieved by referring to the multi-valued index table. It is a device.

[Embodiments of the invention]

FIG. 4 is a block diagram showing the configuration of one embodiment of the present invention.

Generally, an information retrieval device requires a device for storing original data (such as the document itself) and a device for inputting a search formula, but devices other than those that are the main object of the present invention are omitted.

Further, in this embodiment, the number of keywords is four.

The single-value index table 1 is an index cable for attributes that have only one value per attribute for one tuple (this is called a single-value attribute), such as attributes other than the "keyword" shown in FIG. Its configuration is shown in FIG. It is organized in order of value and document number. This table has the same configuration as the prior art. The multivalued index table 2 contains attributes that have multiple values per attribute (this is called a multivalued attribute) for one tuple, such as "keyword" in Figure 3.
It is a tow table for. Its configuration is shown in FIG. They are ordered by value, and there are no identical document numbers for the same value.

The index table R/W circuit 3 receives the search request signal, attributes, and attribute values sent from the control unit 9, and converts the single-value index table 1 or the multi-value index table 2.
The document number is taken out and written into the result memory 6, and the single value index table or multivalue index table write request signal from the control unit 9, attributes, attribute values,
This circuit receives the document number and writes it into the single-value index table 1 or the multi-value index table 2.

The attribute information memory R/W circuit 4 receives an attribute information memory read request and a write request signal from the control unit 9 and reads out information about a specified attribute from the attribute information memory 5, and vice versa. I write things down.

The attribute information memory 5 is a memory that stores information about attributes. Its composition is the seventh
As shown in the figure.

In FIG. 7, the type indicates whether it is single value or multivalue, and is set to 0 or 1, respectively. The value size indicates the maximum length (number of bytes) of the value. The number of values is a number indicating how many values can be assigned to one attribute per one tuple. In the case of a single-valued attribute, it is always 1.

The result memory 6 is the index table R/W circuit 3
is a memory that stores document numbers read from the single-value index table 1 or the multi-value index cable 2.

The management data storage section 7 is a section that stores data for managing original data (such as the document itself), and has the format shown in FIG. 8, which is slightly modified from the format shown in FIG. 3.

In FIG. 8, compared to FIG. 3, a number indicating the beginning of the unused area is entered before setting the value of the keyword. (Initial value 1) For example, document number 0001 indicates that no value has been set from the third area. The number of this area is 1
to n (4 in this case). n is the same value as the "number of values" of the attribute constant memory in FIG.

The management data R/W circuit 8 receives a signal from the control unit 9 and writes it into the management data storage unit 7 or
This is a circuit that reads data from the management data storage section 7.

The control unit 9 receives search request signals and registration request signals from the outside, issues requests to each of the circuits 1 to 8, and receives results. Then, result data and response signals are sent to the outside.

The operation of the present invention will be described using an example.

In this embodiment, a relational model is used in the management data storage section 7.

First, the case of registering data will be explained.

A registration request is sent from the outside to the control unit 9 in the following form.
is sent along with a data registration request signal.

The "classification" of document number 0001 is the budget, the "creation date" is 830220, the "keyword" is the calculator, A... the control unit 9 is the attribute information memory R/W circuit 4
, sends the "classification" together with the attribute information memory read request signal.

In response to this, the attribute information memory R/W circuit 4 reads data regarding "classification" from the attribute information memory 5 and returns it to the control unit 9 together with a read end signal.

The control unit 9 learns that "classification" is a single-valued attribute and has a size of 50 bytes, and sends this information along with the value "budget" document number 0001 to the management data along with a management data write request signal. Send to R/W circuit 8.

The management data R/W circuit 8 receives these, adjusts the value to 50 bytes, and stores it in the management data storage section 7.
Write the value "Budget" in the "Classification" area of document number 0001.

Then, a write end signal is returned to the control unit 9.

After that, the control unit 9 controls the index table R/W circuit 3.
, the attribute name (``classification''), value (``budget''), document number 0001, and value size (50) are sent along with the single value data write request signal.

In response to this, the index table R/W circuit 3
Information is written into the single-value index table 1 as shown in FIG. 5, and a write end signal is returned to the control section 9. The processing here is similar to the conventional technology. Since "creation date" is also a single-valued attribute, similar processing is performed.

Next, the control unit 9 obtains information about the "keyword" from the attribute information memory 5 through the function of the attribute information memory R/W circuit 4. The control unit 9 is
Learn that the "keyword" is a multivalued attribute, four values can be set, and the total size of the "keyword" is 128 bytes.

Then, calculate the size of one "keyword". (32 bytes in this case) After that, this information, the values "calculator" and "A", and the document number
0001 is sent to the management data R/W circuit 8 together with the management data write request signal.

In the management data R/W circuit 8, "keyword"
``Computer'' and ``A'' are written in the ``keyword'' area of document number 0001 of the management data storage unit 7 from the area indicated by the unused area according to the size of one value (32 bytes).

Then, the value of the unused area is set to 3, and the control unit 9
Returns a write end signal to. Thereafter, the control unit 9 sends the index table R/W circuit 3 a multivalued data write request signal along with the attribute name ("keyword"), value ("calculator", ("A"), document number 0001, 1 Sends the magnitude of one value (32).

In response to this, the index table R/W circuit 3
Information is written into the multi-value index table 2 in the order of values as shown in FIG.

At this time, a. If the same value already exists and the document number is also the same, the actual writing will not be performed.

B. If the same value already exists but the document numbers are different, write the same values in the order of the document numbers.

When writing is completed, a write end signal is returned to the control unit 9.

When the control unit 9 completes the above operations for all attributes indicated in the registration request, it returns a registration completion signal to the outside.

The above is the registration.

In this way, data is written to the management data storage section 7, the single-value index table 1, and the multi-value index table 2.

Next, a case will be described in which one keyword is added to a document that has two "keywords".

"Precision machinery" as the keyword for document number 0002
and ``robot'', and assume that you want to add ``sensor'' to this. An attribute value addition request signal, an attribute name (“keyword”), and a value (“sensor”) are sent from the outside to the control unit 9.

The control unit 9 performs the same processing as at the time of registration.
“Sensor” is written in the area indicated by the “keyword” unused area of the management data storage unit 7, and 4 is written in the unused area.

Furthermore, in the multi-value index table 2, "sensor" and document number 0002 are written.

In this way, a "keyword" is added.

Next, search will be explained.

First, the case of a single value attribute will be explained.

Assume that a search request for classification = budget is received from the outside to the control unit 9 (the document number of a document whose "classification" is budget is extracted and stored in the result memory 6).

The control unit 9 performs "classification" in the same manner as in the case of registration.
information about the attribute information is obtained via the attribute information R/W circuit 4. Then, knowing that "classification" is a single-value attribute, a single-value data search request is sent to the index table R/W circuit 3 along with the attribute name ("classification"), value (budget), and value size (50). send a signal.

The index table R/W circuit 3 extracts the corresponding document numbers 0001, . . . from the single value index table 1 and stores them in the result memory 6.

Thereafter, a search end signal is returned to the control unit 9. The control unit 9 returns a search end signal to the outside.

Next, the case of multivalued attributes will be explained.

A measure request in the form of keyword=A comes to the control unit 9 from the outside.

The control unit 9 obtains information about the "keyword" via the attribute information R/W circuit 4 in the same manner as in the case of registration. Then, we learn that "keyword" is a multivalued attribute, and find the size of the value of one keyword (32 bytes in this case). Then, to index table R/W circuit circuit 3, attribute name (“keyword”), value (“A”), size of one value (32)
At the same time, a multivalued data search request signal is sent.

The index table R/W circuit 3 extracts the corresponding document numbers 0001, . . . from the multivalued index table 2 and stores them in the result memory 6.

The following is the same as the single value attribute.

〔Effect of the invention〕

As described above, the present invention realizes handling of multivalued attributes in a file based on a relational model.

Assume that prior art is used to implement the equivalent of the operational description of the present invention.

For comparison, it is assumed that the system appears to the user as shown in FIG. 3, similar to the present invention.

Also, the number of relations is one and the number of keywords is four.

Under this condition, in the prior art, the search device is constructed by replacing the structure shown in FIG. 3 with the structure shown in FIG. 2.

That is, the configuration of the management data storage section 7 is as shown in FIG. (However, n is 4) Therefore, the multivalued index table 2 is eliminated, and the index table for keywords shown in FIG. 9 is changed to the one shown in FIG. 5. Keyword 2... followed by...). The attribute information memory 5 has the configuration shown in FIG. 10 instead of FIG. 7. The types shown in Fig. 7 disappear, and everything becomes single-value continuity.

The control unit 9 needs to remember that "the keyword is composed of keyword 1 to keyword 4."

Now, assume that a search request with keyword=A is given to the control unit 9 from the outside.

In the control unit 9, this request is expressed as (keyword 1=A)+(keyword 2=
A) + (Keyword 3 = A) + (Keyword 4 =
A) Replace with the request (+ indicates logical sum).

Then, the search within each bracket is executed as a single value attribute, as described in the explanation of the operation of the present invention.

Finally, the sum of the results is calculated.

Therefore, the search request to the index table R/W circuit 3 is issued four times, and the index table reference is required four times.

Since each table is typically constructed as a file on a magnetic disk, in practice disk input/output occurs at least four times. In the present invention, since the index table is referenced only once,
If the conventional method were to obtain the same results as the present method, it would take at least four times as long to access the disk.

Generally, if n values are allowed for one attribute per tuple and an attempt is made to achieve the same effect as the present invention using the conventional method, it will take at least n times as long.

As described above, by using this method, it is possible to handle multivalued attributes in a much shorter amount of time than with conventional techniques.

Being able to handle multi-valued attributes not only makes it easier for users to handle data, but also makes it easier for users to handle data, making the flexibility of relational models even more effective.

[Other embodiments of the invention]

In the explanation of the operation of the present invention, the case where the search request is keyword=A has been explained, but the effect of the present invention does not change even if the search request is changed to =>, <, ≧, or ≦.

In the explanation of the operation of the present invention, the pulling table R/W circuit 3 is common to the single-value attribute index table 1 and the multi-value index table 2, but the single-value index table R/W circuit and the multi-value index table R/W circuit corresponding to each table are It may be divided into a value index table R/W circuit.

In this case, the control unit 9 issues a read or write request signal to the single value index table R/W circuit if the attribute is single value, and to the multivalue index table R/W circuit if the attribute is multivalue. .

As described above, it is natural that any modifications are included in the present invention as long as they do not depart from the gist of the invention.

[Brief explanation of drawings]

Figure 1 is a diagram showing relations, Figure 2 is a diagram showing relationships.
FIG. 3 shows a relation in the prior art when there are n keywords. FIG. 4 shows a block diagram showing the configuration of an embodiment of the present invention. 5 shows the structure of the single-value index table 1, FIG. 6 shows the structure of the multi-value index table 2, FIG. 7 shows the structure of the attribute information memory 5, and FIG. 8 shows the structure of the attribute information memory 5. 9 shows the structure of the management data storage unit 7, FIG. 9 shows the structure of the retrieval table for keywords when using the conventional technology, and FIG. 3 is a diagram showing the configuration of an attribute information memory 5 for keywords. FIG. 1...Single index table, 2...Multi-index table, 3...Index table R/W circuit, 4...Metal information memory-R/W circuit, 5...Attribute information memory, 6...Result memory, 7 ... Management data storage section, 8 ... Management data R/W circuit, 9 ... Control section.

Claims (1)

[Scope of Claims] 1. By inputting the attribute value of desired document data from a plurality of document data in which a document number is attached to each document data, and outputting the document number having the attribute value, the desired document data can be obtained. A document data search device for searching document data, comprising: a plurality of attributes of the document data corresponding to the document number; and a single value that stores one attribute value in association with each of these attributes together with the document number; an index table and a multi-value index table that is stored together with the document number in correspondence with a plurality of attribute values; and a type and number of attribute values, whether single-value or multi-value, that each attribute of the document data has. With reference to the attribute information storage means to store, and the type and number of attribute values that each attribute has in this attribute information storage means, from the single-value index table or the multi-value index table, an indexing means for extracting a document number having an attribute value corresponding to the attribute value, and the multi-value index table is configured to retrieve a document number having an attribute value corresponding to the attribute value, and for each document data, the multi-value index table is A document data retrieval device characterized in that it is configured to be stored in correspondence with a document number, and retrieves a document number having an attribute value corresponding to an attribute value for a search by referring to this multivalued index table.