JP2020170426A - Table structure estimation system and table structure estimation method - Google Patents

Abstract

To allow contents of a table to be efficiently converted into a format suitable as a processing target in similar sentence search by AI.SOLUTION: A table structure estimation system 10 includes: a storage device 11 that holds item knowledge 105 that prescribes affinity between each item and a sentence component in each table format, and item value knowledge 106 that prescribes values that each item can contain; and an arithmetic device 14 that performs a process of collating items included in a table to be processed with the item knowledge 105 and identifying potential sentence components among the items based on a degree of the affinity, and a process of arranging the identified items in combination according to types of the sentence components, setting a value prescribed in the item value knowledge 106 for each of the arranged items, and generating a sentence.SELECTED DRAWING: Figure 1

Description

The present invention relates to a table structure estimation system and a table structure estimation method.

As a process for Japanese sentences, similar sentence search by AI (Artificial Intelligence) is actively performed.

Such a method of searching for similar sentences by AI is different from the conventional method in which only those that exactly match the search sentence or the search keyword are targeted. For example, surrounding information whose meaning and appearance position are close to those of a search sentence or search keyword is also targeted, and sentences and vocabularies similar to this surrounding information are searched as similar search results.

However, this AI-based similar sentence search method is not performed after understanding the meaning of Japanese sentences. Therefore, structurally unifying the surrounding information of the search text and the search keyword is a condition for maintaining good search accuracy.

An example in which the above conditions are particularly likely to be a problem is the existence of tables. The table structure forms a matrix structure in which vocabulary and numerical values associated with each other are arranged. Therefore, such a table does not form a sentence as a whole, and it is difficult to obtain suitable search results by the conventional method.

Therefore, there are the following as conventional techniques related to the search method for such a table structure. For example, the memory is a text processing device that processes text including a plurality of character strings, has a processor, and has memory, and stores enumeration expression information relating to an enumeration expression expressed for enumerating hyponyms corresponding to hypernyms. The content related to one theme is repeated from the enumeration expression extraction unit that determines whether or not the enumeration expression is extracted from the text based on the enumeration expression information and the text including the enumeration expression. The repeated expression extraction unit that extracts at least one repeated expression to be expressed, and the hyponym corresponding to the hypernym indicated by the enumerated expression included in the text are determined from the character strings included in the extracted repeated expression. A text processing device (see Patent Document 1), which is characterized by having a hyponym determination unit, has been proposed.

WO2014 / 188555

In the table structure, by arranging the corresponding vocabularies with the table ruled line as a boundary, a predetermined meaning can be expressed by the vocabulary set.

However, such expressions are different from the colloquial expressions used by humans. For example, in the table (old and new comparison table) 300 shown in FIG. 4, if the corresponding vocabulary set is simply extracted and listed by ignoring the ruled line 301 that cannot be expressed in Japanese, an example sentence that does not make sense as Japanese. An expression such as 302 can be obtained. This means that, for example, the vocabulary combination of "500GB" and "none", which are adjacent in the context of contrast, is recognized as a co-occurrence related word in the similar sentence search by AI, and causes an erroneous search result. Show that it can be.
Therefore, an object of the present invention is to provide a technique capable of efficiently converting the contents of a table into a format suitable as a processing target in a similar sentence search by AI.

The table structure estimation system of the present invention that solves the above problems provides item knowledge that defines the affinity between each item and sentence components in each table format, and item value knowledge that defines the values that each item can contain. The storage device to be held and the items included in the table structure to be processed are collated with the item knowledge, and the table structure estimation process for specifying the items that can be sentence components based on the degree of affinity is specified. An operation in which items are arranged in combination according to the type of the sentence component, a value specified in the item value knowledge is set for each of the arranged items, and a sentence generation process for generating a sentence is executed. It is characterized by including an apparatus.

Further, in the table structure estimation method of the present invention, the information processing system defines the item knowledge that defines the affinity between each item and the sentence component in each table format, and the item value knowledge that defines the values that each item can include. A table structure estimation process that is provided with a storage device that holds, collates the items included in the table structure to be processed with the item knowledge, and identifies those items that can be sentence components based on the degree of affinity. , The specified items are arranged in combination according to the type of the sentence component, the value specified in the item value knowledge is set for each of the arranged items, and the sentence generation process for generating a sentence is performed. It is characterized by executing.

According to the present invention, the contents of the table structure can be efficiently converted into a format suitable as a processing target in a similar sentence search by AI.

It is a figure which shows the hardware configuration example of the table structure estimation system in this embodiment. It is a figure which shows the example of the functional block in the table structure estimation system of this embodiment. It is a flow chart which shows the table structure estimation method of this embodiment. It is explanatory drawing which shows the table composition example (old and new comparison table) in this embodiment. It is explanatory drawing which shows the table composition example (code list) in this embodiment. It is a figure which shows the structural example of the tabular knowledge table of this embodiment. It is a figure which shows the structure example 1 of the item value knowledge table in this embodiment. It is a figure which shows the structure example 2 of the item value knowledge table in this embodiment. It is a figure which shows the structural example 3 of the item value knowledge table in this embodiment. It is a figure which shows the structural example of the item value knowledge table in this embodiment. It is a figure which shows the structural example of the table structure setting table in this embodiment.

--- System configuration ---
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing a hardware configuration example of the table structure estimation system 10 of the present embodiment. Further, FIG. 2 is a diagram showing an example of functional blocks in the table structure estimation system of the present embodiment.

The table structure estimation system 10 shown in FIG. 1 is a computer system capable of efficiently converting the contents of the table into a format suitable as a processing target in a similar sentence search by AI.
Such a table structure estimation system 10 includes a storage device 11, a memory 13, an arithmetic device 14, an input device 15, an output device 16, and a communication device 17.
Of these, the storage device 11 is a device composed of an appropriate non-volatile storage element such as an SSD (Solid State Drive) or a hard disk drive.
Further, the memory 13 is a device composed of a volatile storage element such as a RAM.

Further, the arithmetic unit 14 is a CPU that executes the program 12 held in the storage device 101 by reading it into the memory 13 to perform overall control of the apparatus itself, and also performs various determinations, arithmetic operations, and control processes.
Further, the input device 15 is a keyboard, mouse, and microphone that accept key input and voice input from the user.
Further, the output device 16 is a device such as a display or a speaker that displays processed data.

Further, the communication device 17 is a network interface card or the like that connects to a predetermined network and is responsible for communication processing with other devices. However, this communication device 17 is not an indispensable configuration when the table structure estimation system 10 operates as a stand-alone machine.

In the storage device 11, as shown in the block diagram of FIG. 2, the functions required for the table structure estimation system 10 of the present embodiment, that is, the table format estimation unit 101, the table structure estimation unit 102, and the sentence generation A program 12 for mounting the unit 103 is held. These functions will be described later.

Further, in addition to the above-mentioned program 12, the storage device 11 stores at least a tabular knowledge repository 104, an item knowledge repository 105, and an item value knowledge repository 106.

Although details will be described later, the tabular knowledge repository 104 includes a tabular knowledge table 1041. The item knowledge repository 105 also includes an item knowledge table 1051. In addition, the item value knowledge repository 106 includes the item value knowledge table 1061.

--- Table structure estimation flow ---
Hereinafter, the actual procedure of the table structure estimation method in the present embodiment will be described with reference to the drawings. Various operations corresponding to the table structure estimation method described below are realized by a program read by the table structure estimation system 10 into a memory or the like and executed. The program is composed of code for performing various operations described below.

FIG. 3 is a diagram showing a flow example of the table structure estimation method in the present embodiment. In this case, the tabular format estimation unit 101 determines whether or not the value of the table name exists immediately above the tabular structure with respect to the tabular structure file received from the user by the input device 15 (step 200).

If the spreadsheet file is, for example, a spreadsheet, and a value is set in the cell directly above the matrix corresponding to the spreadsheet, the existence of the value is determined as the table name.

As a result of the above determination, when a table name exists in the table structure (step 200: Yes), the table format estimation unit 101 sets the value of the table name to the table of the table format knowledge table 1041 (see FIG. 6). A table format name whose matching degree is equal to or higher than a predetermined threshold value is collated with each value of the model name 501 (example: code list, product list, function comparison, performance comparison, change history, PCL, CRUD). It is presumed to be a tabular format (step 201).

The above-mentioned tabular knowledge table 1041 is information that collects the definitions of the table format. For example, in the design document, a list such as a list of database records, a comparison table showing comparisons between items, and a comparison table are used. Examples include a change history table showing the history of a document, and a condition table for realizing a program checklist (PCL: Program Check List) and a CRUD table (Create, Read, Update, Delete table).

These tabular knowledge tables 1041 differ depending on the characteristics of the business, and may be accumulated according to the business execution, or may be a database or repository prepared in advance by a person involved in the business.

In addition, the table format estimation unit 101 includes attribute values (values indicating business, such as “finance”, “healthcare”, and “electric power”, which are indicated by the received table structure, in the table structure. It is preferable to selectively use the tabular knowledge table 1041 for each industry based on (which can be assumed).

On the other hand, as a result of the above determination, when the table name does not exist in the table structure (step 200: No), the table format estimation unit 101 sets the item in the table structure to the item name group 502 (or item) of the table format knowledge table 1041. By collating with each value of the item name 601) in the value knowledge table 1061, a tabular name whose matching degree exceeds the threshold value defined by the item name matching degree 504 is estimated as the tabular name of the table structure (step 202).

There are two types of item names in the above-mentioned table structure, one is a column name at the top of the table and the other is a row name on the front side. Therefore, it is sufficient to carry out the two trials in the estimation in step 202 described above. If there is a match in either of them, the expansion direction of the table structure can be estimated as it is, and it is possible to fix whether the item name is on the front side or the front side (corresponding to the item name position 503 of the explicit knowledge table 1041). In addition, when the layers are layered on both the front side and the front side, the degree of matching of the combinations can be measured by normalizing and expanding the layered structure as it is.

Further, the tabular classification process related to the above-mentioned table structure is not limited to the estimation according to the above-mentioned steps 200 to 202, and may be specified by external information or user designation.

Subsequently, the table structure estimation unit 102 performs table structure estimation processing in order to compose a Japanese sentence for the table structure whose tabular form is estimated as described above (steps 203 to 206).

In this case, the table structure estimation unit 102 collates each item included in the above-mentioned table structure with each item name 601 of the item knowledge table 1051, and fixes the item whose conditional clause susceptibility 602 is "○". It is estimated as an item that can correspond to the conditional clause in a Japanese sentence (step 203).

The item knowledge table 1051 described above is information that collects definitions of values that can be items as a business handled in each table format. For example, in the item knowledge table 1051 related to the "code list" illustrated in FIG. 7A, item names such as "code", "code name", "processing function name", and "storage table" are defined (item name 601). ).

In the item knowledge table 1051 related to the tabular form "code list" of FIG. 7A, it is stipulated that each value of "code name", "processing function name", and "storage table name" tends to be a conditional clause. Yes (easiness to become a conditional clause 602). Therefore, the table structure estimation unit 102 acquires the value of the conditional clause susceptibility 602 from the item knowledge table 1051 and collates it with the table structure to estimate, for example, "code name" among the items in the table structure as the conditional clause. Will be done.

In addition, in the item knowledge table 1051 relating to the tabular form "performance comparison" of FIG. 7B, as shown in the conditional clause easiness 612, it is defined that all the item names are likely to be conditional clauses.

Further, in the item knowledge table 1051 regarding the tabular form "change history" of FIG. 7C, as shown in the conditional clause easiness 622, it is defined that each item name other than the "change number" is likely to be a conditional clause. ..

If there are a plurality of item names that are likely to be conditional clauses, the order of ease of being conditional clauses shall be defined in advance in each item name in the item knowledge table 1051. In that case, the table structure estimation unit 102 can select the items to be the conditional clauses in order from the one with the highest rank indicated by the rank information.

In step 203 described above, when the item of the condition section cannot be estimated from the information of the item knowledge table 1051 (step 2031: No), the table structure estimation unit 102 determines the condition from the information of the item value knowledge table 1061. Estimate the items in the section (step 204).

For example, as shown in the classification (item name) 701 of the item value knowledge table 1061 in FIG. 8, the item generally described as "code name" is described as "ID name" (711) in the table 710. There are cases where it has been done.

In that case, since the item names do not match in the estimation based on the item knowledge table 1051, there is a high possibility that the items in the conditional clause cannot be estimated. Therefore, the table structure estimation unit 102 in this case reads the item values (instances) 712 to 714 for each item in the table structure 710, collates this with each value of the item value 702 in the item value knowledge table 1061, and matches. When exceeds a predetermined threshold value (item value matching degree 703), it can be determined that the "code name" indicated by the classification (item name) 701 corresponds to the "ID name" in the table structure 710.

Then, the table structure estimation unit 102 refers to the item knowledge table 1051 in which the conditional clause easiness is defined with respect to the "code name" determined in this way, and the value of the conditional clause easiness 602 is "○". For example, the item is estimated as a conditional clause.

As described above, values such as "customer code" (712) and "product code" (713) are described in the "ID name" item of the table 710. Therefore, if it is accumulated in the "code name" classification (item name) in the item value knowledge table 1061, it is estimated that the item "ID name" in the table 710 can be regarded as the "code name".
The item value knowledge table 1061 may be prepared in advance in the same manner as the tabular knowledge table 1041 and the item knowledge table 1051.
In addition, the calculation of the degree of agreement may be the same as the calculation in the tabular estimation using the item knowledge of a certain number.

In order to compose a natural Japanese sentence, it is necessary to further set the subject and predicate. Therefore, the table structure estimation unit 102 estimates the subject based on the item knowledge table 1051 as in the above-mentioned conditional clause (step 205).

The processing itself here is the same as the processing in which the conditional clause is set (step 203), and the items other than the items set in the conditional clause regarding the table structure are set to each value of the subject easiness 603 of the item knowledge table 1051. Collate sequentially and identify the one whose susceptibility is "○" as the subject.

For example, in the item knowledge table 1051 relating to the "code list", it is set that the "code name", the "processing function name", and the "storage table name" are likely to be the subjects. Therefore, the table structure estimation unit 102 identifies these items that are likely to be the subject from the item knowledge table 1051, and estimates the corresponding items in the table as the subject.

In step 205 described above, when the subject item cannot be estimated from the information in the item knowledge table 1051 (step 206: No), the table structure estimation unit 102 uses the information in the item value knowledge table 1061. Estimate the subject item (step 207).

In this case, by the same processing as in step 204, the item in the table structure 710 is read as the classification (item name) in the item value knowledge table 1061, and the value of the subject susceptibility 603 of the item knowledge table 1051 is "○" with the item. In the case of ", the item is estimated as the subject.

If the item value knowledge table 1061 is configured to add information on whether or not the item value can be the subject, the item value of the row of interest becomes the subject as the subject by using the information of the item value. You can also directly decide whether to get it.

For example, since the item value "1" representing the item number is unlikely to be the subject of a Japanese sentence, the ease of becoming the subject can be set low in the item value knowledge table 1061 and erroneous conversion to a Japanese sentence can be made. Can be reduced.

Subsequently, the table structure estimation unit 102 estimates items for the predicate as well as the above-mentioned conditional clause and subject (step 208). Since the processing contents are the same, the description will be omitted.

As for the predicate, the item value (instance value) of the row or column of interest of the item set as the subject may be adopted as it is. However, as in the case of subject setting, if information on whether or not the item value can be a predicate is added to the item value knowledge table 1061, the item value of the line of interest can be changed by using the information of this item value. It is also possible to directly judge whether it can be a predicate as a context.
In the process of the table structure estimation unit 102 shown above, the information of the contents shown in the table structure setting table 801 shown in FIG. 9 is finally stored in the memory 13 or the like.
As described above, when the estimation of sentence components such as conditional clauses, subjects, and predicates is completed, the sentence generation unit 103 executes steps 209 to 212.

That is, the items estimated as the above-mentioned sentence components in the steps up to the above exist as a framework for pointing to each word, and even if they are arranged as they are, the example sentence 302 shown with respect to the table 300 in FIG. As such, the result is similar to a meaningless array.

Therefore, the sentence generation unit 103 assigns the value of the corresponding item in the table structure, that is, a specific phrase, to each item as a sentence composition sentence, and performs a process of adding particles and the like that are lacking as a Japanese sentence. ..

The sentence generation unit 103 first generates the item corresponding to the subject by adding the particle "ha" to the item name (802) of the item set as the subject in steps 205 to 207 (step 209). For example, when the "processing function name" item is set as the subject, the subject is "processing function name is".

On the other hand, in order to generate a conditional clause and a predicate, not only the item name but also the item value (instance value) of the row or column of interest is required. For example, in the table structure 400 of FIG. 5 showing a specific table structure, the first row (401) of the table contents can be designated as the row of interest. The row of interest and the column of interest may be sequentially applied to the expansion direction of the table structure estimated in the tabular estimation step.

When the first line (401) of the table 400 is specified as the line of interest, the sentence generation unit 103 sets the item name (803) of the item set in the condition clause in step 203 as the content of the specific condition clause. Add the particle "ga".

Further, the sentence generation unit 103 completes by concatenating the item value (instance) of the line of interest of the item, "customer code (804)", and then adding the conditional word "case" and the comma ",". (Step 210). That is, the completed form in this specific example is "when the code name is a customer code".

Further, the sentence generation unit 103 sets the item value (instance) of the line of interest in the item set as the subject in steps 205 to 207 as the content of the specific predicate (step 211).

That is, specifically, the "customer code update process (805)", which is the item value (instance) of the line of interest in the "processing function name" item, is set. Further, in order to make a Japanese sentence, the auxiliary verb "is" and the punctuation mark "." That form the verb form as a predicate are added to complete the sentence (step 211). The completed form in this specific example is "customer code update process."

In this way, the sentence generation unit 103 completes each part as a Japanese sentence, then synthesizes the Japanese sentence through a process of connecting them (step 212), and ends the process. That is, in this specific example, "when the code name is a customer code, the processing function name is the customer code update process" (410).

When the specific line of interest is moved to 402, the same procedure is taken, and the final Japanese sentence is "If the code name is a product code, the processing function name is product shipping processing." ..

The word to be added to each part is not limited to the word shown above as long as it has a meaning such as a subject or a conditional clause. For example, the phrase "case" added in the generation of the conditional clause may be the phrase "time" that can represent the same conditional clause.

In the explanation so far, the business knowledge that can be classified into the code system of the design document is applied as an example, but the business knowledge that can be classified into another system can be similarly applied.

For example, even if the design document is the same, when a table structure that can be classified into the change history table is input, the information of the business knowledge repository related to the change history table of the design document prepared in advance is used as in step 201. Since the degree of matching is high, it is estimated in the change history table as a tabular format, and the table structure is determined from the item knowledge information and item value (instance) knowledge information related to the change history table in the same manner as in steps 202 to 209. It can be estimated and reconstructed into a Japanese sentence.

The tabular business knowledge related to the change history table is, for example, a table name of "change history table", and the item knowledge is, for example, an item name of "changer", "change date", or "change content". Knowledge of item values is, for example, item values such as "Suzuki", "Sato" (in some cases, position and staff number) and "2018/12/24".

When the instance is specified in column units, that is, when it is estimated that the extension direction of the table structure is the horizontal direction by tabular estimation, the composition of the table described as the item name above is at the top of the table. It is not the row but the column that is on the front side and pays attention to the process of specifying the instance. The processing step itself simply replaces the rows and columns.

The original table data reconstructed into Japanese sentences in this way has a form similar to the search sentence of the similar error search set by the person, and even when the similar sentence search by AI is performed, it is similar to the person's intention. It makes it easier to select sentences.

Although the best mode for carrying out the present invention has been specifically described above, the present invention is not limited to this, and various modifications can be made without departing from the gist thereof.

According to the present embodiment, even in a document including a table structure, the format and structure of the table structure can be estimated and reconstructed into a natural sentence. As a result, the content indicated by the table can be converted into a natural sentence according to the search sentence set by the person, and the searchability by the person can be facilitated.
That is, the contents of the table can be efficiently converted into a format suitable as a processing target in the similar sentence search by AI.

The description herein reveals at least the following: That is, in the table structure estimation system of the present embodiment, the storage device further holds the table format knowledge in which each table format is classified according to the items included in the table format, and the arithmetic unit further holds the table format knowledge to be processed. The items included in are collated with the tabular knowledge, the tabular estimation process for estimating the tabular form of the tabular form is further executed, and the tabular structure estimation process and the sentence generation process are performed as the tabular form of the estimated tabular form. It may be the one that executes.

According to this, it is possible to efficiently estimate the table format for a table structure of unknown format. As a result, the contents of the table structure can be efficiently converted in a format suitable as a processing target in the similar sentence search by AI.

Further, in the table structure estimation system of the present embodiment, the storage device holds the table explicit knowledge for each business, and the arithmetic unit performs the target business indicated by the table structure in the table format estimation process. It may be assumed that the tabular form of the table is estimated by using the tabular form knowledge corresponding to.

According to this, even when a table structure for a wide range of business is to be processed, the table format can be estimated efficiently. As a result, the contents of the table structure can be efficiently converted in a format suitable as a processing target in the similar sentence search by AI.

Further, in the table structure estimation system of the present embodiment, the arithmetic unit performs the table structure estimation process by selecting one of the items that can be any of the subject, the conditional clause, and the predicate of the sentence component. Specified based on the degree of sex, and in the sentence generation process, each item specified for each of the subject, the conditional clause, and the predicate is arranged in the order of the subject, the conditional clause, and the predicate, and each of the arranged items It may be said that the value specified in the item value knowledge is set and a sentence is generated.

According to this, it becomes possible to efficiently generate a sentence in a form that makes sense as a Japanese sentence. As a result, the contents of the table structure can be efficiently converted in a format suitable as a processing target in the similar sentence search by AI.

Further, in the table structure estimation system of the present embodiment, when the arithmetic unit cannot identify a potential sentence component by collating the item with the item knowledge in the table structure estimation process, the table structure estimates. The value of the item is collated with the item value knowledge, the item associated with the value is specified in the item value knowledge, and the item is collated with the item value to specify what can be the sentence component. , May be.

According to this, even if the items included in the table are different from those specified in the item knowledge, the items that are considered to be synonymous with the item are identified from the item knowledge, and the subsequent processing is smoothed. It will be possible to proceed. As a result, the contents of the table structure can be efficiently converted in a format suitable as a processing target in the similar sentence search by AI.

Further, in the table structure estimation method of the present embodiment, the information processing system further holds the table format knowledge in which each table format is classified according to the items included in the table format in the storage device, and the table to be processed. The items included in the set are collated with the tabular knowledge, the tabular estimation process for estimating the tabular form of the tabular form is further executed, and the tabular structure estimation process and the sentence generation are performed as the table form of the estimated tabular form. The process may be executed.

Further, in the table structure estimation method of the present embodiment, the information processing system retains the table explicit knowledge for each business in the storage device, and when the table format estimation process is performed, the target business indicated by the table structure is used. The table format of the table structure may be estimated by using the corresponding table format knowledge.

Further, in the table structure estimation method of the present embodiment, the information processing system may be one of the subject, the conditional clause, and the predicate of the sentence component among the items in the table structure estimation process. Specified based on the degree of affinity, and in the sentence generation process, each item specified for each of the subject, the conditional clause, and the predicate is arranged in the order of the subject, the conditional clause, and the predicate, and the arranged items are arranged. For each, the value specified in the item value knowledge may be set to generate a sentence.

Further, in the table structure estimation method of the present embodiment, when the information processing system cannot identify a potential sentence component by collating the item with the item knowledge in the table structure estimation process, the table structure is described. The value of the item is collated with the item value knowledge, the item associated with the value is specified in the item value knowledge, and the item is collated with the item value to specify what can be the sentence component. May be good.

10 Table structure estimation system 11 Storage device 12 Program 13 Memory 14 Computing device 15 Input device 16 Output device 17 Communication device 101 Table format estimation unit 102 Table structure estimation unit 103 Statement generation unit 104 Table format knowledge repository 1041 Table format knowledge table 105 Item Knowledge Repository 1051 Item Knowledge Table 106 Item Value Knowledge Repository 1061 Item Value Knowledge Table 300 Table Structure Example 301 Ruled Line in Table Structure Example 302 Simple Conversion Statement in Table Structure Example 400 Code List of Design Document Table 401 First Code List Note line 402 Second note line 501 in the code list Example of description contents of tabular knowledge repository (table format name)
502 Example of description contents of tabular knowledge repository (item name group)
503 Example of description contents of tabular knowledge repository (item name position)
504 Example of description contents of tabular knowledge repository (item name matching degree)
601 Item Knowledge repository description content example (item name)
602 Item Knowledge repository description content example (ease of conditional clause)
603 Item Example of contents described in the knowledge repository (subject ease)
701 Item Value Knowledge Repository Description Example (Classification)
702 Knowledge of item values Example of contents described in repository (item values)
801 Example of table structure setting table (in the case of code list)
802 Subject setting in table structure setting table 803 Conditional clause subject setting in table structure setting table 804 Conditional clause subject setting in table structure setting table 805 Predicate setting in table structure setting table

Claims (10)

A storage device that holds item knowledge that defines the affinity between each item and sentence components in each table format, and item value knowledge that defines the values that each item can contain.
Table structure estimation processing that collates the items included in the table structure to be processed with the item knowledge and specifies the items that can be sentence components based on the degree of affinity, and the specified items are the sentence components. An arithmetic unit that executes a statement generation process that generates a statement by setting a value specified in the item value knowledge for each of the arranged items and arranging them in combination according to the type of
A table structure estimation system characterized by including. The storage device is
Further retains explicit knowledge of each table format by classifying each table format according to the items included in the table format.
The arithmetic unit
The items included in the table structure to be processed are collated with the table format knowledge, the table format estimation process for estimating the table format of the table structure is further executed, and the table structure estimation process is performed as the table structure of the estimated table format. And the statement generation process is executed.
The table structure estimation system according to claim 1, wherein the system is characterized by the above. The storage device is
The above explicit knowledge is held for each business,
The arithmetic unit
In the table format estimation process, the table format of the table structure is estimated by using the table format knowledge corresponding to the target business indicated by the table structure.
The table structure estimation system according to claim 2, wherein the system is characterized by this. The arithmetic unit
In the table structure estimation process, among the items, those that can be any of the subject, conditional clause, and predicate of the sentence component are specified based on the degree of affinity.
In the sentence generation process, each item specified for each of the subject, the conditional clause, and the predicate is arranged in the order of the subject, the conditional clause, and the predicate, and each of the arranged items is defined by the item value knowledge. It sets the value that is set and generates a statement,
The table structure estimation system according to claim 1, wherein the system is characterized by the above. The arithmetic unit
In the table structure estimation process, when it is not possible to identify a potential sentence component by collating the item with the item knowledge, the value of the item in the table structure is collated with the item value knowledge, and the item value knowledge The item associated with the value is specified, and the item is collated with the item value to specify what can be the sentence component.
The table structure estimation system according to claim 1, wherein the system is characterized by the above. Information processing system
It is provided with a storage device that holds item knowledge that defines the affinity between each item and sentence components in each table format, and item value knowledge that defines the values that each item can contain.
The table structure estimation process that collates the items included in the table structure to be processed with the item knowledge and specifies the items that can be sentence components based on the degree of affinity, and the specified items are the sentence components. The sentence generation process of generating a sentence by setting the value specified in the item value knowledge for each of the arranged items and arranging them in combination according to the type of
A table structure estimation method characterized by executing. The information processing system
In the storage device, the tabular knowledge that classifies each tabular form according to the items included in the tabular form is further retained.
The items included in the table structure to be processed are collated with the table format knowledge, the table format estimation process for estimating the table format of the table structure is further executed, and the table structure estimation process is performed as the table structure of the estimated table format. And execute the sentence generation process,
The table structure estimation method according to claim 6, wherein the table structure is estimated. The information processing system
In the storage device, the tabular knowledge is retained for each business,
In the table format estimation process, the table format of the table structure is estimated by using the table format knowledge corresponding to the target business indicated by the table structure.
The table structure estimation method according to claim 7, wherein the table structure is estimated. The information processing system
In the table structure estimation process, among the items, those that can be any of the subject, conditional clause, and predicate of the sentence component are specified based on the degree of affinity.
In the sentence generation process, each item specified for each of the subject, the conditional clause, and the predicate is arranged in the order of the subject, the conditional clause, and the predicate, and each of the arranged items is defined by the item value knowledge. Set the value to be set and generate a statement,
The table structure estimation method according to claim 6, wherein the table structure is estimated. The information processing system
In the table structure estimation process, when it is not possible to identify a potential sentence component by collating the item with the item knowledge, the value of the item in the table structure is collated with the item value knowledge, and the item value knowledge Identify the item associated with the value, collate the item with the item value, and identify what can be the sentence component.
The table structure estimation method according to claim 6, wherein the table structure is estimated.

Images