JP7098502B2 - Reporting equipment, methods, and programs - Google Patents

Description

Embodiments of the present invention relate to a report creation device, a method, and a program for creating a report based on a user's utterance.

Conventionally, in the field of customer consultation services and the like, a technique for retrieving desired information from a large amount of information by voice dialogue is known, for example, as in Patent Document 1.

In the technique disclosed in Patent Document 1, the search target information is classified into categories (hierarchical), keywords for searching are set for each classification, and the search target information is narrowed down based on the keywords included in the input voice. The desired information is searched for.

Japanese Unexamined Patent Publication No. 11-306195

However, in such conventional information retrieval based on voice dialogue, the user needs to consciously utter a phrase used as a keyword, and if the phrase used as a keyword is not included in the voice, the user needs to consciously speak the phrase. Searching becomes difficult.

The problem to be solved by the present invention is to provide a report creation device, a method, and a program for a user to search for desired information based on freely made utterances and create a report without being aware of keywords. It is to be.

The report creation device of the embodiment is a report creation device that creates a report based on voice, and includes a user information acquisition unit, a semantic class extraction unit, a slot item setting unit, and a slot item correspondence unit. It has a search unit. The user information acquisition unit acquires user information. The meaning class extraction unit extracts the keyword corresponding to the predefined meaning class from the text information and the user information converted from the voice. The slot item setting unit sets input conditions for slot items for a report based on a predetermined report template. The slot item correspondence unit associates keywords with slot items based on the set input conditions. The search unit searches for the information corresponding to the result of the mapping from the database in which the information related to the creation of the report is stored. Then, the search result is used as a candidate for the corresponding slot item.

It is a block diagram which shows the specific configuration example of the report making apparatus of 1st Embodiment. It is a flowchart which shows the operation example of the report making apparatus of 1st Embodiment. It is a data table showing an example of semantic class information. It is a schematic diagram for demonstrating the keyword extraction according to an extended rule. It is a data structure diagram which shows an example of the correspondence result. It is a data structure diagram which shows an example of a conversion table. It is a data table showing an example of the weight set for each attribute in each related information database. It is a conceptual diagram which shows the configuration example of each database table stored in each related information database. It is a block diagram which shows the detailed configuration example of the search result narrowing-down part. Among the information held in the matter information database table, it is a table that displays a list of examples of records that hit the search query. It is a table which shows the result of sorting by the score order for each record of the table shown in FIG. Among the information held in the matter information database table, it is a table that lists another example of the record that hit the search query. It is a table which shows the result of sorting by the score order for each record of the table shown in FIG. Among the information held in the matter information database table, it is a table that displays a list of yet another example of the record that hits the search query. It is a table which shows the result of sorting by the score order for each record of the table shown in FIG. Among the information held in the matter information database table, it is a table that displays a list of yet another example of the record that hits the search query. It is a table which shows the result of sorting by the score order for each record of the table shown in FIG. It is a schematic diagram which shows the display example of a slot item. It is a block diagram which shows the specific functional configuration example of the report making apparatus of 2nd Embodiment. It is a table which compares and displays the example of the weight value before and after the update. It is a table which compares and displays the example of the value before and after the update of the weight in the case of further update. It is a table which compares and displays another example of the weight value before and after the update. It is a table which compares and displays another example of the value before and after the update of the weight in the case of further update. It is a flowchart which shows the operation example of the report making apparatus of 2nd Embodiment. It is a block diagram which shows the specific functional configuration example of the report making apparatus of 3rd Embodiment. It is a flowchart which shows the operation example of the report making apparatus of 3rd Embodiment. It is a block diagram which shows the hardware configuration example of the report making apparatus of 1st to 3rd Embodiment.

Hereinafter, the report creating apparatus according to each embodiment of the present invention will be described with reference to the drawings.

(First Embodiment)
The report preparation device of the first embodiment will be described.

FIG. 1 is a block diagram showing a specific configuration example of the report creating device of the first embodiment.

The report creation device 10 includes a voice recognition unit 12, a user information acquisition unit 14, a semantic class analysis unit 16, a slot item setting unit 18, a slot item correspondence unit 20, a search unit 22, and a search result narrowing down. A unit 24, a slot item determination unit 26, and a display / selection unit 28 are provided.

The report preparation device 10 also includes a slot item database 30, a related information database 32, and a weight database 34.

The voice recognition unit 12 receives an utterance from the user (hereinafter referred to as "user utterance") a, converts the user utterance a into a voice recognition text b, and outputs the voice recognition text b to the semantic class analysis unit 16.

The user information acquisition unit 14 acquires at least one user information c among the user profile, time information, position information, and surrounding environment information, and outputs the user information c to the semantic class analysis unit 16.

The user profile can be obtained from user input. Alternatively, when the user speaks from a user terminal such as a smartphone, the user information registered in the user terminal such as a smartphone can be used as a user profile. Time information, position information, and surrounding environment information can also be acquired from a clock function, a GPS function, a thermometer function, a hygrometer function, and the like mounted on a user terminal such as a smartphone.

The semantic class analysis unit 16 extracts keywords corresponding to the predefined semantic class from the voice recognition text b output from the voice recognition unit 12 and the user information c output from the user information acquisition unit 14. The meaning class information d consisting of the meaning class and the keyword is output to the slot item corresponding unit 20.

The slot item setting unit 18 acquires a preset report template 40 from, for example, an external database, sets input conditions for slot items for the report, and sets the set input condition e to the slot item database 30. Store.

The slot item corresponding unit 20 refers to the input condition e stored in the slot item database 30 and associates the keyword d2 included in the semantic class information d output from the semantic class analysis unit 16 with the slot item. Then, the association result f is output to the search unit 22.

The search unit 22 searches for the information corresponding to the matching result f output from the slot item corresponding unit 20 from the related information database 32 in which the information related to the report creation is stored, and searches the search result g as the search result. Output to the narrowing section 24. A plurality of related information databases 32 may exist as needed, and FIG. 1 illustrates three related information databases 32 (# 1), 32 (# 2), and 32 (# 3) as an example.

The search result narrowing-down unit 24 acquires the weighting information h set for the slot item corresponding to the input condition e from the weighting database 34 in which the weighting information preset for each slot item is stored, and the weighting information h. Is used to calculate the score m. The search result g is narrowed down based on the calculated score m, and the narrowing down result i is output to the slot item determination unit 26.

The slot item determination unit 26 determines the slot item j to be displayed based on the narrowing-down result i, and outputs the slot item j to the display / selection unit 28.

The display / selection unit 28 displays the slot item j output from the slot item determination unit 26 on the display 4 of a user terminal such as a smartphone in order to allow the user to select a candidate for the slot item.

It should be noted that the slot item corresponding unit 20 uses the user utterance a and the user utterance a used for extracting the semantic class information d for the slot item that cannot be associated with the slot item or is set to perform a search as an input condition. The search query k1 and the weighted information k2 are extracted from the user information c. Then, the extracted search query k1 and the weighting information k2 are output to the search unit 22.

The search unit 22 acquires the information corresponding to the search query k1 output from the slot item corresponding unit 20 from any of the related information databases 32 (# 1), 32 (# 2), and 32 (# 3), and acquires the information. The search result g is output to the search result narrowing unit 24.

The search result narrowing-down unit 24 calculates the score m for the search result g using the weighting information k2. The search result g is narrowed down according to the calculated score m, and the narrowing down result i is output to the slot item determination unit 26.

Further, the slot item corresponding unit 20 can also convert the keyword of the semantic class information d output from the semantic class analysis unit 16 into a predefined conversion value d3.

Further, the search result narrowing unit 24 may control (control the sort key) to change the display order of the slot item candidates displayed by the display / selection unit 28 based on the semantic class information d. can.

Next, an operation example of the report creating device 10 of the present embodiment will be described with reference to the flowchart of FIG.

For example, a user (Mr. Suzuki) selected the report template 40 in advance and said, "Today, I revisited Shibahigashi in Fuchu, which I visited last time, and introduced product X" from a user terminal such as a smartphone. Suppose you did.

This user utterance a is received by the voice recognition unit 12, and is converted into the voice recognition text b "Today, I revisited Shibahigashi in Fuchu, where I visited last time, and introduced the product X" by a general voice recognition technique. .. The converted voice recognition text b is output from the voice recognition unit 12 to the semantic class analysis unit 16 (S1).

Further, the user information acquisition unit 14 acquires at least one user information c among the user profile, time information, location information, and surrounding environment information. The user profile is acquired from user information registered in a user terminal such as a smartphone, for example. The time information is acquired by the clock function of a user terminal such as a smartphone. The position information (for example, "35.6732357, 139.4603954") is acquired by the GPS function of a user terminal such as a smartphone. The user information c is output from the user information acquisition unit 14 to the semantic class analysis unit 16. The user profile, time information, and location information may be acquired not only from the user terminal but also from the user utterance a.

The semantic class analysis unit 16 extracts keywords corresponding to the predefined semantic class from the voice recognition text b output from the voice recognition unit 12 and the user information c output from the user information acquisition unit 14 ( S2).

FIG. 3 is a data table showing an example of the semantic class information d composed of the defined semantic class and the keyword.

The semantic class information d is a semantic class corresponding to a unique expression class such as "date", "company name", "address", "location", "product name", "summary", "speaker", and "location information". It is composed of d1 and the keyword d2 corresponding to each of these meaning classes d1.

The keyword d2 corresponding to the meaning class d1 is extracted by the meaning class analysis unit 16 by using the existing named entity extraction technique in natural language processing. Note that FIG. 3 shows the results of semantic analysis of user utterance a by the semantic class analysis unit 16 and keyword extraction. As shown in FIG. 3, in order to extract the keyword d2 "Fuchu Shibahigashi" in the meaning class d1 "place" and the keyword d2 "introduce product X" in the meaning class d1 "overview", the following An extended rule set as is used.

Location: = {company name} of {address}
｜ {Company name} near {Address}
| {Company name} of {station name}
・ ・ ・
Summary: = {Product name} <particle>? <Verb>
｜ {Product name} <particle>? <Verb phrase>
・ ・ ・
FIG. 4 is a schematic diagram for explaining keyword extraction according to the above-mentioned extended rule.

As shown in FIG. 4, according to the extended rule, the keyword d2 such as "Fuchu Shibahigashi" as the meaning class d1 "place" and the keyword d2 as the meaning class d1 "overview" such as "introduced product X". The keyword d2 is extracted.

The semantic class information d as shown in FIG. 3 from which the keyword d2 is extracted is output from the semantic class analysis unit 16 to the slot item corresponding unit 20.

The slot item database 30 stores the slot item input condition e for the report set by the slot item setting unit 18 based on the preset report template 40. Specifically, in each slot, the meaning class (keyword) of the item value, the search query, and the related information is set. The details of the slot item input condition e will be described later.

The slot item corresponding unit 20 refers to the input condition e stored in the slot item database 30, and the keyword d2 included in the semantic class information d extracted in S2 is associated with the slot item (S3).

FIG. 5 is a data structure diagram showing an example of the matching result.

First, the slot item input condition e set by the slot item setting unit 18 will be described. As shown in FIG. 5, the slot input condition e includes the attributes of slot f1, item name f2, item value f3, search query f4, and related information f5. Slot f1 shows the input items that the report template has. The item name f2 is an item name indicating the slot f1. The item value f3 indicates a specific report content corresponding to the item name f2. The search query f4 indicates a search query for searching for information corresponding to the item value f3. Related information f5 is information used for score calculation, and indicates an item name or an attribute of the corresponding database.

In this example, five slots are set in the report template 40, and it is shown that an item name is set for each slot. That is, the matter ID is associated with the slot 1 as the item name, and information composed of, for example, numerical values and alphabets is set. In slot 2, a visited destination is associated as an item name, and for example, a company name is set as an item value. The visit date is associated with the slot 3 as the item name, and the date is set as the item value. A visitor is associated with the slot 4 as an item name, and a speaker is set as an item value, for example. In this way, the slot item setting unit 18 sets the conditions set for each slot item.

When the user makes an utterance, the information (keyword) extracted by the above-mentioned processes S1 to S3 is registered in the corresponding item. For example, slot 4 is a slot indicating a visitor, and indicates that "Suzuki" corresponding to the speaker extracted from the user utterance a is registered as the item value f3 for slot 4.

That is, information registration from the user utterance a is realized by associating the meaning class d1 with the item value f3 according to the preset attributes for each slot. If a keyword that is not included in the user's utterance a or is ambiguous and the correspondence is not confirmed, the item value f3 is set to empty, and the search query f4 and the related information f5 are extracted.

The extraction of the search query f4 and the related information f5 is not limited to the case where the correspondence is not confirmed as described above, and the search query f4 is also applied to the slot item in which the search query f4 and the related information f5 are set in the input condition e. , Related information f5 is extracted.

Further, in the slot item correspondence unit 20, ambiguous keywords included in the user's utterance a are converted into specific information by using a conversion table described later.

FIG. 6 is a diagram showing an example of a conversion table.

The conversion table u has a configuration in which a new attribute of the conversion value d3 is added to the table shown in FIG.

The slot item corresponding unit 20 converts an ambiguous keyword included in the user's utterance a into specific information by using the conversion table u as illustrated in FIG.

For example, when the user's utterance a includes the keyword d2 "before this", the keyword d2 "before this" is converted to the conversion value d3 "2018/7/26 to 2018/8/9". .. The conversion value of this conversion table u changes dynamically and can be set for each user. That is, dates such as "today" and "yesterday" are changing values and are automatically updated every day. It also has items that the user can freely change, such as the period for "previous" and "the other day", the radius for "near", and the conversion value for "example project" (converted to project X in this example). ing.

In this way, the report is created by registering the user utterance a in the corresponding slot item value f3 of the report template 40. Here, as described above, the search unit 22 finds the item value f3 that does not include the information corresponding to the slot item set in the report template 40 in the user utterance a or cannot be associated with ambiguous information. , Search past report information and various master information and extract the relevant information. Next, the details of this search will be described.

Not limited to the item value f3 that could not be associated, the same processing is performed when the search query f4 or the related information f5 is set in the input condition e.

The slot item corresponding unit 20 outputs the correspondence result f processed in S1 to S3 to the search unit 22. The search unit 22 searches the related information database 32 based on the search query of the matching result f received from the slot item corresponding unit 20, and extracts information related to report creation (S4).

In FIG. 1, three related information databases 32 (# 1), 32 (# 2), and 32 (# 3) are illustrated as an example. These related information databases 32 (# 1), 32 (# 2), and 32 (# 3) are relational databases (RDS) using at least one attribute as a key as a prerequisite, and related information databases 32 (# 1). ), 32 (# 2), 32 (# 3), a weight is set for each attribute.

FIG. 7 is a data table showing an example of weights set for each attribute in each related information database 32 (# 1), 32 (# 2), 32 (# 3).

In addition, in FIG. 7, the related information database 32 (# 1), 32 (# 2), and 32 (# 3) are specifically the case information database 32 (# 1) related to the case information and the transaction related to the customer information, respectively. It is exemplified as the destination information database 32 (# 2) and the business report database 32 (# 3) relating to the business report. The values of these weights w1 to w5 exemplified in FIG. 7 can be updated.

FIG. 8 shows the databases 33 (# 1), 33 (# 2), 33 (# 3) stored in the related information databases 32 (# 1), 32 (# 2), and 32 (# 3). It is a figure which shows a specific example.

The matter information database 33 (# 1) of FIG. 8 has "matter ID", "matter name", "company name", "person in charge", "reporter", "registration date", and "update date" as attributes. It is equipped with.

The business partner information database 33 (# 2) has "business partner ID", "company name", "address", "nearest station", and "registration date" as attributes.

The sales report database 33 (# 3) has "report ID", "case ID", "case name", "company name", "reporter", "report date", and "summary" as attributes. ..

These related information databases 32 (# 1), 32 (# 2), and 32 (# 3) are relational databases using the matter information database 33 (# 1) as a master database. The matter information database 33 (# 1) and the business partner information database 33 (# 2) are associated with the attribute "company name" as a key. The matter information database 33 (# 1) and the sales report database 33 (# 3) are associated with the attributes "matter ID" and "reporter" as keys.

In the matter information database 33 (# 1), a very large number of "matter IDs" are registered as well as the "matter name". Further, the "case ID" having a format such as "1111-XX-1111" is an attribute that is difficult to input by voice.

For example, when it is desired to fill the item value f3 of the slot 2 based on the matching result f as shown in FIG. 5, the search unit 22 has the customer information database 33 having the attributes of “company name” and “address”. Search for (# 2). Then, when the search result g is acquired, it is output to the search result narrowing unit 24.

The search result narrowing-down unit 24 acquires the weighting information h set for the slot item corresponding to the input condition e from the weighting database 34 in which the weighting information set for each slot item is stored. Then, the score m is calculated using the weighting information h. The search result g is narrowed down based on the calculated score m, and the narrowing down result i is output to the slot item determination unit 26 (S5).

Subsequently, the details of the search narrowing down in the search result narrowing down unit 24 will be described.

FIG. 9 is a block diagram showing a detailed configuration example of the search result narrowing down portion.
The search result narrowing unit 24 includes a candidate extraction unit 24a, a score calculation unit 24b, a sort unit 24c, and a condition determination unit 24d.

The process of narrowing down the "case ID" from the search result g that the reporter "Suzuki" introduced the product X today will be described as an example.

Since this search result g is a case where only the "reporter" is confirmed and the "company name" is not confirmed, the candidate extraction unit 24a refers to the search query f4 of the matching result shown in FIG. Then, the matter information database 33 (# 1) is searched with "including Suzuki in the reporter" as a search query.

FIG. 10 is a table showing a list of records that hit the search query among the information held in the matter information database table 33 (# 1).

Next, the score calculation unit 24b weights the relevance degree H1 (case name) and the relevance degree H2 ₍ reporter) for _each record shown in the table of FIG. 10, and the score m is calculated. Will be done. For example, for "project ID" = "2222-YY-2222", a score m of "1.5" is calculated as follows.

Score (2222-YY-2222)
= H ₁ (project name = {product X, introduction}) * w ₁ (project name)
+ H ₂ (reporter = Suzuki) * w ₂ (reporter)
= 0.5 * 1.0 + 1.0 * 1.0
= 1.5
For the weights (w ₁ , w ₂ ) used above, for example, the values illustrated in FIG. 7 are used.

When the score m is calculated for each record in this way, each record is sorted in descending order of the score m by the sort unit 24c.

FIG. 11 is a table showing the results of sorting the records in the table shown in FIG. 10 in order of score.

The condition determination unit 24d identifies a record satisfying the determination condition from the sort result shown in FIG. A threshold value of the score m can be used as the determination condition. For example, when the score m is set to "1.0" or more as the determination condition, it is determined that the top five records shown in FIG. 11 satisfy the determination condition.

The record determined by the condition determination unit 24d to satisfy the determination condition is output to the slot item determination unit 26 as the narrowing down result i.
As another narrowing example, the process of narrowing down the "project ID" from the search result g that the reporter "Suzuki" visited Shibahigashi Digital Company B and introduced the product X will be described.

In this search result g, the "reporter" and the "company name" are fixed, and the "case ID" is not fixed. Therefore, the candidate extraction unit 24a "includes Suzuki in the reporter" and The matter information database table 33 (# 1) is searched with "the company name includes Shibahigashi Digital Company B" as a search query.

FIG. 12 is a table showing a list of records that hit the search query among the information held in the matter information database table 33 (# 1).

Next, the score calculation unit 24b transfers the relevance degree H ₁ (case name), the relevance degree H ₂ (reporter), and the relevance degree H ₃ (company name) to each record shown in the table shown in FIG. Is weighted, and the score m is calculated. For example, for "project ID" = "2222-YY-2222", a score m of "2.5" is calculated as follows.

Score (2222-YY-2222)
= H ₁ (project name = {product X, introduction}) * w ₁ (project name)
+ H ₂ (reporter = Suzuki) * w ₂ (reporter)
+ H ₃ (company name = Shibahigashi Digital) * w ₃ (company name)
= 0.5 * 1.0 + 1.0 * 1.0 + 1.0 * 1.0
= 2.5
For the weights (w ₁ , w ₂ , w ₃ ) used above, for example, the values illustrated in FIG. 7 are used.

When the score m is calculated for each record in this way, each record is sorted in descending order of the score m by the sort unit 24c.

FIG. 13 is a table showing the results of sorting the records in the table shown in FIG. 12 in order of score.

The condition determination unit 24d identifies a record satisfying the determination condition from the sort result shown in FIG. A threshold value of the score m can be used as the determination condition. For example, when the score m is set to "2.0" or more as the determination condition, it is determined that the top two records shown in FIG. 13 satisfy the determination condition.

The record determined by the condition determination unit 24d to satisfy the determination condition is output to the slot item determination unit 26 as the narrowing down result i.

Next, a process for narrowing down the search results based on ambiguous utterances by the search result narrowing unit 24 will be described.

This search result g is an example of narrowing down the "case ID" from the utterance that is ambiguous as "Fuchu Shibahigashi" as a candidate for the "company name" although the "reporter" is confirmed. In this case, the candidate extraction unit 24a searches the matter information database 33 (# 1) with "the reporter includes Suzuki" and "the company name includes Shibahigashi" as a search query.

FIG. 14 is a table showing a list of records that hit the search query among the information held in the matter information database 33 (# 1).

Next, the score calculation unit 24b determines the degree of relevance H1 (case name), the degree of relevance H2 ₍ reporter), the degree of relevance H3 ₍ company name), and the degree of relevance H3 (company name) for _each record shown in the table shown in FIG. And the degree of relevance H4 (address) is weighted, and the score m is calculated. For example, for "case ID" = "1111-XX-1111", a score m of "1.8" is calculated as follows.

Score (1111-XX-1111)
= H ₁ (project name = {product X, introduction}) * w ₁ (project name)
+ H ₂ (reporter = Suzuki) * w ₂ (reporter)
+ H ₃ (company name = Shibahigashi) * w ₃ (company name)
+ H ₄ (Address = Fuchu) * w ₄ (Address)
= 0 (The project name does not include product X or introduction, so it is 0) * 1.0 + 1.0 * 1.0 + 0.8 (The company name is not an exact match, so it is 0.8) * 1.0 + 0 (0 because the address of Shibahigashi Energy Company A does not include Fuchu) * 1.0
= 1.8
Further, for example, for "project ID" = "2222-YY-2222", a score of 3.3 is calculated as follows.

Score (2222-YY-2222)
= H ₁ (project name = {product X, introduction}) * w ₁ (project name)
+ H ₂ (reporter = Suzuki) * w ₂ (reporter)
+ H ₃ (company name = Shibahigashi) * w ₃ (company name)
+ H ₄ (Address = Fuchu) * w ₄ (Address)
= 0.5 (The project name includes product X, so it is 0.5) * 1.0 + 1.0 * 1.0 + 0.8 (The company name is not an exact match, so it is 0.8) ) * 1.0 + 1.0 (Since Fuchu is included in the address of Shibahigashi Digital Company B, it will be 1.0) * 1.0
= 3.3
When the score m is calculated for each record in this way, each record is sorted in descending order of the score m by the sort unit 24c.

FIG. 15 is a table showing the results of sorting the records in the table shown in FIG. 14 in order of score.

The condition determination unit 24d identifies a record satisfying the determination condition from the sort result shown in FIG. A threshold value of the score m can be used as the determination condition. For example, when the score m is set to "1.5" or more as the determination condition, it is determined that the top five records shown in FIG. 15 satisfy the determination condition.

The record determined by the condition determination unit 24d to satisfy the determination condition is output to the slot item determination unit 26 as the narrowing down result i.

Further, another process in the case of narrowing down the search results based on ambiguous utterances by the search result narrowing unit 24 will be described.

In this search result g, although the "reporter" has been confirmed, it is ambiguous that "Shibahigashi in Fuchu" is a candidate for "company name" and "visited last time" as information related to "updated date". This is an example of narrowing down the "project ID" from the utterance. In this case, the candidate extraction unit 24a searches the matter information database table 33 (# 1) with "the reporter includes Suzuki" and "the company name includes Shibahigashi" as a search query.

FIG. 16 is a table showing a list of records that hit the search query among the information held in the matter information database table 33 (# 1).

The keyword d2 "before this" included in the ambiguous utterance "visited last time" is "2018/7 / 26-2018 /" by using the conversion table shown in FIG. 6 in the slot item corresponding unit 20. It is converted to the conversion value d3 of "8/9".

Next, the score calculation unit 24b determines the degree of relevance H1 (case name), the degree of relevance H2 ₍ reporter), the degree of relevance H3 ₍ company name), and the degree of relevance H3 (company name) for _each record shown in the table shown in FIG. Weighting is performed on the relevance degree H ₄ (address) and the relevance degree H ₅ (update date), and the score m is calculated. For example, for "project ID" = "2222-YY-2222", the score m of "4.3" is calculated as follows.

Score (2222-YY-2222)
= H ₁ (project name = {product X, introduction}) * w ₁ (project name)
+ H ₂ (reporter = Suzuki) * w ₂ (reporter)
+ H ₃ (company name = Shibahigashi) * w ₃ (company name)
+ H ₄ (Address = Fuchu) * w ₄ (Address)
+ H ₅ (updated date = last time) * w ₅ (updated date)
= 0.5 (The project name includes product X, so it is 0.5) * 1.0 + 1.0 * 1.0 + 0.8 (The company name is not an exact match, so it is 0.8) ) * 1.0 + 1.0 (Because the address of Shibahigashi Digital Company B is included, it will be 1.0) * 1.0 + 1.0 ("Before" is "2018/7 / 26-2018" / 8/9 ”) * 1.0
= 4.3
When the score m is calculated for each record in this way, each record is sorted in descending order of the score m by the sort unit 24c.

FIG. 17 is a table showing the results of sorting the records in the table shown in FIG. 15 in order of score.

The condition determination unit 24d identifies a record satisfying the determination condition from the sort result shown in FIG. A threshold value of the score m can be used as the determination condition. For example, when the score m is set to "1.5" or more as the determination condition, it is determined that the top five records shown in FIG. 17 satisfy the determination condition.

The record determined by the condition determination unit 24d to satisfy the determination condition is output to the slot item determination unit 26 as the narrowing down result i.

The slot item determination unit 26 determines the slot item j based on the narrowing-down result i, and outputs the slot item j to the display / selection unit 28.

The display / selection unit 28 displays the slot item j output from the slot item determination unit 26 on the display 4 in order to allow the user to select a slot item candidate (S6).

FIG. 18 is a diagram showing a display example of slot items.

FIG. 18 shows a report 50 created based on the record of “case ID” = “2222-YY-2222” having the highest score from the narrowing-down result i shown in FIG.

The user can easily make corrections while confirming the contents of the report 50.

For example, in any of the "Matter ID / Matter name" column 50a, "Company name" column 50b, "Visit date" column 50c, "Reporter" column 50d, and "Summary" column 50e displayed in the report 50. When the cursor is placed on the cursor, all the relevant information output from the search result narrowing unit 24 to the slot item determination unit 26, for example, included in the narrowing result i as shown in FIG. 17 is displayed.

For example, as illustrated in FIG. 18, when the cursor is placed on the “company name” column 50b, the visited candidate list 50b1 is displayed. When the display / selection unit 28 creates a report based on the narrowing result i as shown in FIG. 17, when the cursor is placed on the "company name" column 50b by the user operation, the display / selection unit 28 causes the display / selection unit 28 to display / select the report. All company names ("Shibahigashi Digital Company B", "Shibahigashi Infrastructure C Company", "Shibahigashi Energy Company A") included in the "company name" of the narrowing down result i shown in FIG. 17 are candidate company names. It is displayed in the list 50b1.

When the user selects a desired company name from the company names displayed in the company name candidate list 50b1 by clicking a mouse or the like, the selected company name is selected from the narrowing results i as shown in FIG. Based on the record including, the contents of the "Matter ID / Matter name" column 50a, which is a slot item whose item value has not been determined, are also updated in conjunction with each other. In the example of FIG. 18, since "Shibahigashi Infrastructure C" was selected, "3333-ZZ-3333 / Product X defect handling" was added to "Proposal ID / Proposal name" 50a based on the information of the narrowing down result i shown in FIG. Is displayed.

Here, when "Shibahigashi Digital Company B" is selected, the "Matter ID / Matter Name" 50a is uniquely determined because there are two records of "Shibahigashi Digital Company B" in the narrowing down result i. do not have.

In this case, for example, it is preferable to display the slot items so that they can be identified by coloring the slot items so that the slot items whose information cannot be confirmed can be identified. The user can confirm the candidate (search result) of the corresponding slot item by moving the cursor to the slot item whose information has not been confirmed (S7: No).

In the example of FIG. 18, when "Shibahigashi Digital Company B" is selected, "Matter ID / Matter name" 50a is shown as a slot item whose information cannot be confirmed, and the cursor is placed on "Matter ID / Matter name" 50a. And "2222-YY-2222" and "2222-YY-2222" are displayed as candidates based on the record of FIG. When the content of the report 50 is finally confirmed, it is confirmed by an operation such as a confirmation button (not shown) (S7: Yes). It is also possible to respond by voice instructions.

The slot item determination unit 26 determines the report 52 based on the content determined by the user operation (S8).

The contents of the finalized report 52 are stored in the related information database 32. In this way, the report 52 can be easily created by using not only the user utterance a but also the related information.

According to the report creation device of the present embodiment, as described above, the user searches for desired information based on freely made utterances and creates a report (slot filtering) without being aware of the keywords. Is possible.

Further, the report creation device of the present embodiment can estimate the information necessary for report creation by the related information and the weighting even for the slot items that the user does not directly speak. This not only eliminates the need for the user to manually enter the report, but also avoids the hassle of selecting candidate information from a huge list.

Further, the report creation device of the present embodiment can interpret the meaning from ambiguous utterances such as "Shibahigashi in Fuchu" and "visited last time" and estimate the information necessary for report preparation. can. As a result, the accuracy of slot filtering by natural utterance can be improved, and a highly reliable report can be created, which makes it possible to improve the convenience of the user.

(Second embodiment)
The report preparation device of the second embodiment will be described.

The report preparation device of the second embodiment is a modification of the report preparation device of the first embodiment. Therefore, in the following description, the points different from those of the first embodiment will be described, and the same parts as those of the first embodiment will be indicated by using the same reference numerals, so that duplicate description will be omitted.

FIG. 19 is a block diagram showing a specific functional configuration example of the report creating device of the second embodiment.

The report creating device 10'shown in FIG. 19 has a configuration in which a weight updating unit 36 is added to the report creating device 10 shown in FIG.

The weight update unit 36 receives the selection result of the slot item selected from the user via the display / selection unit 28 via the slot item determination unit 26, and the weighting set for each slot item based on the selection result. Update the information. Then, the update result is reflected in the weight database 34.

The weighting information is updated, for example, by updating the weight _wi with respect to the degree of relevance _Hi according to the following update formula.

w _i = wi _i -η * g (H _i , w _i )
Here, g (Hi, _wi ) is, for example, a function depending on whether or not the _i -th attribute value of the first candidate and the i-th attribute value of the determination candidate match.

Then, when the i-th attribute value of the first candidate is equal to the _i -th attribute value of the decision candidate, g (Hi, _wi ) = | _Hi * α _positive |, and in other cases, teeth,
_g (Hi, wi) ₌ -| _Hi * α _negative |.

For example, the update parameters are η = 0.1, α _positive = 1.0, and α _negative = 1.0.

FIG. 20 is a table for comparing and displaying examples of weight values before and after updating.

In FIG. 20, when the first candidate is selected and determined in response to the user utterance a that "I visited Shibahigashi in Fuchu today and introduced the product X", the weight updated according to the above update formula is shown. Is shown.

That is, the weights of the attribute values related to the weight determination, "case name", "reporter", "company name", and "address", have been updated and increased.

The weight update based on the user utterance a different from the above will be described with reference to FIG.

FIG. 21 is updated according to the above update formula when the first candidate is selected and determined according to the user utterance a that "Today, I revisited Shibahigashi in Fuchu, which I visited last time, and introduced product X." Shows the weights that have been made.

The user utterance a considered in FIG. 21 includes the keyword "before" regarding the "updated date" which was not found in the user utterance a considered in FIG. 20. Therefore, it can be seen that the weight of the "updated date" in FIG. 21 is increased in the positive direction as compared with the result of FIG. In this way, the more keywords related to "updated date" such as "last time" or "visited last Thursday" are spoken, the more important the "updated date" is and the more weight is learned. To.

Another example will be described with reference to FIGS. 22 and 23.

FIG. 22 is a table for comparing and displaying another example of the weight value before and after the update.

FIG. 22 shows the weights updated according to the above update formula when the second candidate is selected and determined according to the user utterance a "I visited Shibahigashi today and introduced the product X". ..

It is assumed that the "company name" of the second candidate is different from that of the first candidate. In such a case, as shown in FIG. 22, learning is performed so that the weight of the "company name" is reduced.

On the other hand, FIG. 23 is a table for comparing and displaying another example of the values before and after the weight is updated when the weight is further updated.

FIG. 23 shows the weights updated according to the above update formula when the second candidate is selected and determined according to the user utterance a that "I visited Shibahigashi in Fuchu today and introduced the product X". Shows.

According to FIG. 23, although the search results are narrowed down based on the address information of "Fuchu", this user does not include "Fuchu" in the address of the company name finally selected by this user. The narrowing down based on the address information in is judged to be non-conforming, and the weight related to the "address" is reduced.

Next, an operation example of the report creating device 10'of the present embodiment will be described with reference to the flowchart of FIG. 24.

In FIG. 24, the same process as in FIG. 2 is assigned the same step number.

That is, the operation performed by the report creating device 10'of the present embodiment is performed between steps S7 (S7: Yes) and step S8 of the operation performed by the report creating device 10 of the first embodiment. As described above, the weight update process (S71) based on the selection result, which is performed by the weight update unit 36, is added.

According to the report preparation device of the present embodiment, it is possible to appropriately update the weights as described above.

It should be noted that the update of the weighting information may be learned so that the weights of the attribute values related to the weight determination are relatively increased, and the update formula is not limited to the above-mentioned example.

(Third embodiment)
The report preparation device of the third embodiment will be described.

The report preparation device of the third embodiment is a modification of the report preparation device to which the second embodiment is applied. Therefore, in the following description, the differences from the first and second embodiments will be described, and the same parts as those of the first and second embodiments will be indicated by using the same reference numerals, thereby omitting duplicate description.

FIG. 25 is a block diagram showing a specific functional configuration example of the report creating device of the third embodiment.

The report creating device 10 ″ shown in FIG. 25 has a configuration in which the dialogue unit 38 is added to the report creating device 10 ′ illustrated in FIG. 19.

The dialogue unit 38 provides dialogue information r for supplementing the missing information for the slot item specified by the slot item determination unit 26 based on the search result g by the search unit 22 and the narrowing down result i by the search result narrowing unit 24. Generate.

By displaying the table dialogue information r generated by the dialogue unit 38 on the display / selection unit 28, it is possible to urge the user to supplement the missing information.

When the user inputs supplementary information, the slot item determination unit 26 determines an appropriate slot item.

Next, an operation example of the report creating device 10 ″ of the present embodiment will be described according to the flowchart of FIG.

In FIG. 26, the same processes as those in FIGS. 2 and 24 are assigned the same step numbers.

That is, the operation performed by the report creating device 10'' of the present embodiment is described above between step S5 and step S6 of the operation performed by the report creating device 10'of the second embodiment. The determination of the presence or absence of missing information (S51) made by the slot item determination unit 26 and the creation of dialogue information for the missing information (S52) made by the dialogue unit 36 are added to the above.

As described above, according to the report creation device 10'' of the present embodiment, even if the information required for the slot item is insufficient, the user is requested to supplement it. It becomes possible to appropriately determine the slot item.

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other embodiments, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and variations thereof are included in the scope of the invention described in the claims and the equivalent scope thereof, as are included in the scope and gist of the invention.

The report creating apparatus according to the first to third embodiments described above can be realized, for example, by the cooperation of the hardware constituting a general computer and the program (software) executed by the computer. For example, when a computer executes a predetermined program, the above-mentioned voice recognition unit 12, user information acquisition unit 14, semantic class analysis unit 16, slot item setting unit 18, slot item correspondence unit 20, search unit 22, and search result. It is possible to realize functional components constituting the report creation device 10, such as the narrowing-down unit 24, the slot item determination unit 26, the display / selection unit 28, the weight update unit 36, and the dialogue unit 38. Further, the slot item database, the related information database, the weight database, and the like described above can be realized by using the large-capacity storage device provided in the computer.

FIG. 27 is a block diagram showing a hardware configuration example of the report creating device 10 of the first to third embodiments described above.

As shown in FIG. 25, it includes a processor (CPU) 1, a random access memory (RAM) 2, a read-only memory (ROM) 3, a display 4, and a storage device 5.

The CPU 1 executes a program for controlling the report creating device 10 in response to an operation by a user or an instruction from an external device. Further, the CPU 1 manages the memory space of the storage device 5. The RAM 2 is a memory used as a work area of the CPU 1. As the RAM 2, a semiconductor memory such as a DRAM is used. The ROM 3 is a non-volatile memory in which a program for controlling the report creation device 10, control data, and the like are stored in advance. The ROM 3 holds, for example, a BIOS (Basic Input / Output System).

The storage device 5 stores intermediate data and final data generated by the report creation device 10. The storage device 5 also holds various programs used by the report creation device 10, data related to the programs, and various databases. The program stored in the storage device 5 is read out when executed by the report creation device 10 and expanded in the RAM 2. As the storage device 5, for example, an SSD (Solid State Drive), a hard disk drive, or the like is used. The storage device 5 may be used as long as it can store data, and another storage medium may be used as the storage device 5. The above program may be configured to be pre-installed on the computer, or may be distributed over the network and configured to be appropriately installed on the computer.

The display 4 displays a GUI (Graphical User Interface) or the like corresponding to various programs by the CPU 1. For example, the display 4 is used to display the result of the execution process by the program.

The report creating device 10 according to the first to third embodiments described above is configured as a network system in which a plurality of computers are communicably connected, and each of the above-described parts is distributed to the plurality of computers. It may be configured to be used. Further, the report creating device 10 according to the first to third embodiments described above may be a virtual machine operating on a cloud system.

1 ... Processor (CPU), 2 ... Random access memory (RAM), 3 ... Read-only memory (ROM), 4 ... Display, 5 ... Storage device, 10 ... Report creation device, 12 ... Voice recognition unit, 14 ... User information acquisition unit, 16 ... Semantic class analysis unit, 18 ... Slot item setting unit, 20 ... Slot item correspondence unit, 22 ... Search unit, 24 ... Search result narrowing unit , 24a ... Candidate extraction unit, 24b ... Score calculation unit, 24c ... Sort unit, 24d ... Condition judgment unit, 26 ... Slot item determination unit, 28 ... Display / selection unit, 30 ... Slot item database , 32 ... Related information database, 33 ... Database table, 34 ... Weight database, 36 ... Weight update section, 38 ... Dialogue section, 40 ... Report template, 50, 52 ... Report.

Claims (10)

It is a report creation device that creates a report based on voice.
The user information acquisition unit that acquires user information and
A meaning class extractor that extracts keywords corresponding to a predefined meaning class from text information converted from voice and the user information, and a meaning class extraction unit.
A slot item setting unit that sets input conditions for slot items for the report based on a predetermined report template, and a slot item setting unit.
Based on the set input conditions, the slot item corresponding unit that associates the keyword with the slot item, and
It is provided with a search unit for searching the information corresponding to the result of the association from the database in which the information related to the creation of the report is stored.
A report creation device that uses the search result as a candidate for a corresponding slot item. A search result narrowing unit that narrows down the search results by the search unit based on the weighting information preset for each slot item, and the search result narrowing unit.
The report creation device according to claim 1, further comprising a slot item determination unit that determines a slot item to be output based on the result of the narrowing down. The report creating device according to claim 1 or 2, further comprising a voice recognition unit that receives utterances from a user and converts the utterances into the voice text information. The slot item corresponding unit extracts the search query and the weighting information from the utterance and the user information when the slot item cannot be associated or when the search query is set as the input condition of the slot item. ,
The search unit acquires the information corresponding to the search query from the database and obtains it.
The search result narrowing unit calculates a score for the acquired search result using the extracted weighting information.
The report creation device according to claim 2 or 3, wherein the slot item determination unit determines a candidate for the slot item according to the score. The report creating device according to any one of claims 2 to 4, further comprising a weight updating unit that updates the weighting information set for each slot item based on a selection by the user. The report creation device according to any one of claims 1 to 5, wherein the slot item corresponding unit converts the keyword extracted by the meaning class extraction unit into a predefined conversion value. The report creating device according to any one of claims 2 to 6, wherein the search result narrowing-down unit controls to change the output order of slot item candidates based on the meaning class. The slot item determination unit further includes a dialogue unit that determines missing information in determining the slot item based on the search result or the narrowing down result, and generates dialogue information for requesting the missing information. , The report creating device according to any one of claims 1 to 7. This is a report creation method applied to a voice-based report creation device.
Steps to get user information and
A step of extracting a keyword corresponding to a predefined meaning class from voice text information and the user information, and
A step of setting input conditions for slot items for the report based on a predetermined report template, and
Based on the set input conditions, the step of associating the keyword with the slot item and
It comprises a step of searching the database in which the information related to the creation of the report is searched for the information corresponding to the result of the association.
A report creation method in which the search result is used as a candidate for a corresponding slot item. A voice-based reporting program that runs on a computer
The function to acquire user information and
A function to extract keywords corresponding to a predefined meaning class from voice text information and the user information, and
A function to set input conditions for slot items for the report based on a predetermined report template, and
A function of associating the keyword with the slot item based on the set input conditions, and
The function of searching the database containing the information related to the creation of the report for the information corresponding to the result of the mapping is executed.
A report creation program that uses the search results as candidates for corresponding slot items.

Images