Dialogue knowledge creation device and dialogue knowledge creation program

The dialogue knowledge creation device efficiently constructs dialogue knowledge through slot recognition and relationship extraction, addressing high preparation costs and direct user response challenges in conventional systems.

JP7581844B2Active Publication Date: 2024-11-13OKI ELECTRIC INDUSTRY CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2020212784
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-12-22
Publication Date
2024-11-13
Estimated Expiration
2040-12-22

AI Technical Summary

Technical Problem

Conventional systems for creating dialogue knowledge require high costs to prepare all interaction branches and often only output similar FAQ candidates without directly answering user questions, necessitating repetitive questioning based on multiple conditions.

Method used

A dialogue knowledge creation device and program that includes document input, slot recognition, inter-slot relationship extraction, slot relationship tree construction, and integration to efficiently create dialogue knowledge for direct user responses.

Benefits of technology

Enables efficient creation of dialogue knowledge for interactive systems, allowing direct answers to user inquiries without repetitive questioning, even in the absence of response history or FAQs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007581844000001
    Figure 0007581844000001
  • Figure 0007581844000002
    Figure 0007581844000002
  • Figure 0007581844000003
    Figure 0007581844000003
Patent Text Reader

Abstract

To provide a dialog knowledge creation device and a dialog knowledge creation program capable of efficiently creating dialog knowledge.SOLUTION: A dialog knowledge creation device 1 includes document input means 11 for receiving an input of document information, slot recognition means 12 for acquiring slots defined corresponding to the document information, inter-slot relation extraction means 13 for extracting relations among the plurality of acquired slots on the basis of dependence relations between the slots, slot relation tree construction means 14 for constructing a slot relation tree of a tree structure from the extracted relations among the slots, each slot table creation means 15 for creating a list of user inquiry items from the relations among the slots and the slot relation tree, and creating a slot table of each value of a user inquiry item in the list, slot table integration means 17 for integrating slot tables created for each value of the user inquiry item, and dialog knowledge output means 18 for outputting the integrated slot table as dialog knowledge.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a dialogue knowledge creating device and a dialogue knowledge creating program, and can be applied to a dialogue knowledge creating device that creates dialogue knowledge for realizing a dialogue that answers a user's inquiry, for example. [Background technology]

[0002] Conventionally, there are systems in which operators at contact centers and the like use FAQs (Frequently Asked Questions), which are combinations of answers to questions created from past response history, etc., to answer inquiries from users. The operator, who is talking to the user, searches the FAQs and answers the user's inquiry according to the FAQ that is found.

[0003] As a technology related to FAQs, for example, Patent Document 1 discloses a system that extracts question sentences that are candidates for FAQs from data based on information related to the semantic content obtained as a result of analyzing the structure of the data, and extracts question sentences that are candidates for FAQs by integrating them by keyword without defining a framework of sentence structure. Also, Patent Document 2 discloses a system that generates FAQ candidates by generating a dependency structure tree as a result of dependency relationships between independent words and syntactic analysis, and extracting and aggregating representative sentence candidates for inquiries and answers. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2012-3704 A [Patent Document 2] JP 2013-50896 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the conventional system described above, it is necessary to create FAQs to answer questions from users. in In addition, the system only outputs candidates similar to the user's question from the FAQ, and does not directly present an answer to the user's question.

[0006] Incidentally, when the answer to be presented to the user varies depending on multiple conditions, a dialogue-style system that repeatedly asks the user questions until the system has enough information to determine an answer is more effective than a question-and-answer format such as an FAQ.

[0007] However, when implementing such a dialogue format, it is necessary to prepare all possible branches of interaction with the user, which poses the problem of high cost in creating knowledge (ontology) for the dialogue.

[0008] Therefore, there is a demand for a dialogue knowledge creating device and a dialogue knowledge creating program that can efficiently create dialogue knowledge for realizing a dialogue that answers a user's inquiry. [Means for solving the problem]

[0009] The dialogue knowledge generating device of the first invention comprises: (1) a document input means for receiving input of document information; calligraphy (3) slot recognition means for acquiring slots defined corresponding to information; (4) slot relationship extraction means for extracting relationships between the acquired multiple slots based on the dependency relationships between the slots; (5) slot relationship tree construction means for constructing a tree-structured slot relationship tree from the extracted relationships between the slots; (6) slot table creation means for creating a list of user inquiry items from the relationships between the slots and the slot relationship tree, and for creating a slot table for each value of the user inquiry item in the list; (7) slot table integration means for integrating the slot tables created for each value of the user inquiry item; and (8) dialogue knowledge output means for outputting the integrated slot table as dialogue knowledge.

[0010] The dialogue knowledge creation program of the second aspect of the present invention includes a computer including: (1) a document input unit for receiving input of document information; calligraphy The system is characterized by having the system function as: (3) a slot recognition means for acquiring slots defined corresponding to information; (4) an inter-slot relationship extraction means for extracting relationships between the acquired multiple slots based on the dependency relationships between the slots; (5) a slot relationship tree construction means for constructing a tree-structured slot relationship tree from the extracted inter-slot relationships; (6) a slot table creation means for creating a list of user inquiry items from the inter-slot relationships and the slot relationship tree, and for creating a slot table for each value of the user inquiry item in the list; (7) a slot table integration means for integrating the slot tables created for each value of the user inquiry item; and (8) a dialogue knowledge output means for outputting the integrated slot table as dialogue knowledge. Effect of the Invention

[0011] According to the present invention, dialogue knowledge for realizing a dialogue for answering a user's inquiry can be efficiently created. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing a functional configuration of a dialogue knowledge creating device according to an embodiment. [Diagram 2] 4 is a flowchart showing the overall operation of the dialogue knowledge creating device according to the embodiment. [Diagram 3] 10 is a flowchart showing a detailed operation of a slot recognition unit according to the embodiment. [Figure 4] 13 is a flowchart showing a detailed operation of an inter-slot relation extraction unit according to the embodiment. [Diagram 5] 13 is a flowchart showing a detailed operation of a slot relation tree construction unit according to the embodiment. [Figure 6] 13 is a flowchart showing a detailed operation of a slot relation tree construction unit according to the embodiment. [Figure 7]13 is a flowchart illustrating a detailed operation of an inter-slot relationship addition and modification unit according to the embodiment. [Figure 8] 13 is a flowchart showing a detailed operation of a slot table integration unit according to the embodiment. [Figure 9] FIG. 2 is an explanatory diagram showing an example of a document input to the dialogue knowledge creating device according to the embodiment; [Figure 10] 11 is an explanatory diagram showing an example of slots and their values ​​defined by a slot recognition unit according to the embodiment; FIG. [Figure 11] 11 is an explanatory diagram showing an example of a relationship between slots extracted by an inter-slot relationship extraction unit according to the embodiment; FIG. [Figure 12] 1 is an explanatory diagram showing an example of a slot relationship tree constructed by a slot relationship tree construction unit according to an embodiment; [Figure 13] FIG. 11 is an explanatory diagram illustrating an example of a user inquiry item list according to the embodiment. [Figure 14] FIG. 2 is an explanatory diagram showing the details of each slot in a slot table according to the embodiment; [Figure 15] FIG. 11 is an explanatory diagram (part 1) showing an example of a slot table (a slot table created with a user inquiry item set to "day trip business trip category") according to the embodiment. [Figure 16] 11 is an explanatory diagram showing an example in which the inter-slot relation addition and correction unit according to the embodiment presents information that is lacking for completing the slot table to an expert or the like. FIG. [Figure 17] 11 is an explanatory diagram showing an example of a relationship between slots in which missing information is repeatedly added and corrected by an inter-slot relationship addition and correction unit according to the embodiment; FIG. [Figure 18] FIG. 2 is an explanatory diagram (part 2) showing an example of a slot table (a slot table created with a user inquiry item set to "day trip category") according to the embodiment; [Figure 19] 11 is an explanatory diagram showing an example of a slot table obtained by integrating each slot table by a slot table integrating unit according to the embodiment; FIG. [Figure 20] FIG. 2 is an explanatory diagram showing an example of a dialogue using dialogue knowledge created by the dialogue knowledge creation device according to the embodiment; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] (A) Main embodiment DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a dialogue knowledge creating device and a dialogue knowledge creating program according to an embodiment of the present invention will be described in detail with reference to the drawings.

[0014] (A-1) Configuration of the embodiment FIG. 1 is a block diagram showing a functional configuration of a dialogue knowledge creating device according to an embodiment.

[0015] In FIG. 1, the dialogue knowledge creation device 1 has a document input unit 11, a slot recognition unit 12, an inter-slot relationship extraction unit 13, a slot relationship tree construction unit 14, an individual slot table creation unit 15, an inter-slot relationship addition and correction unit 16, a slot table integration unit 17, and a dialogue knowledge output unit 18.

[0016] The dialogue knowledge creation device according to the embodiment may be configured as hardware, such as a dedicated IC chip equipped with each component shown in FIG. 1, or may be configured as software, primarily consisting of a CPU and a program executed by the CPU, but functionally it can be represented as shown in FIG. 1.

[0017] The document input unit 11 receives input of documents such as manuals, specification collections, and instruction manuals.

[0018] The slot recognition unit 12 recognizes slots from an input document. Here, a slot is a variable for a question that is prepared by a dialogue system that performs a slot-filling dialogue to ask a user a question and store the answer value. For example, an answer or a value resulting from an inference is input to a slot (variable). Hereinafter, a slot and a candidate value to be input to the slot are referred to as a slot candidate.

[0019] The inter-slot relationship extraction unit 13 extracts the relationship between the recognized slots. The relationship between slots is a dependency in which the value of one slot is determined when the value of another slot is determined. The relationship between slots can also be extracted using, for example, notation and natural language processing techniques, in the same manner as above.

[0020] Furthermore, when a method for adding and correcting the inter-slot relationships is passed to the inter-slot relationship extraction unit 13 from an inter-slot relationship addition and correction unit 16 (to be described later), the inter-slot relationship extraction unit 13 corrects the inter-slot relationships based on the added correction.

[0021] The slot relation tree construction unit 14 constructs a slot relation tree from the inter-slot relations extracted by the above-mentioned inter-slot relation extraction unit 13. The slot relation tree is tree-structured data in which relations between a plurality of slots are constructed according to dependency relationships.

[0022] The slot table creation unit 15 creates a list of user inquiry items (slots expected to be inquired from the user) from the slot relation tree constructed by the above-mentioned slot relation tree construction unit 14, and creates a slot table (each slot table) by assigning a dialogue order to each slot for each user inquiry item. A slot table is a table created to determine the slots required to store a value in a certain slot from the slot relation tree. Creating a slot table determines the slots that the dialogue system must ask in order to present a final answer to the user.

[0023] Furthermore, when each slot table creation unit 15 finds information that is insufficient for creating (completing) a slot table, it passes the insufficient information to an inter-slot relationship addition and correction unit 16, which will be described later.

[0024] The inter-slot relation adding and correcting unit 16 adds and corrects information that is lacking for creating a slot table. For example, the inter-slot relation adding and correcting unit 16 generates information for adding and correcting the inter-slot relations extracted by the inter-slot relation extracting unit 13 based on information input from outside (input from a document expert, etc.). The inter-slot relation adding and correcting unit 16 passes the generated information for adding and correcting the inter-slot relations to the inter-slot relation extracting unit 13.

[0025] The slot table integration unit 17 integrates the slot tables created by the slot table creation units 15. The integrated slot table is used as dialogue knowledge (ontology) in a dialogue system (for example, an AI automated voice dialogue system, etc.).

[0026] The dialogue knowledge output unit 18 outputs dialogue knowledge to be used in dealing with inquiries.

[0027] (A-2) Operation of the embodiment Next, the operation of the dialogue knowledge creating device 1 according to the embodiment having the above-mentioned configuration will be described.

[0028] (A-2-1) Overall operation of the dialogue knowledge creation device FIG. 2 is a flowchart showing the overall operation of the dialogue knowledge creating device according to the embodiment.

[0029] The dialogue knowledge creation device 1 accepts input of a document such as a rule set via the document input unit 11 (S11). Next, the slot recognition unit 12 recognizes slots from the input document and defines slots and value candidates (S12). Next, the slot relationship extraction unit 13 extracts relationships between the recognized slots (S13). Next, the slot relationship tree construction unit 14 constructs a slot relationship tree from the relationships between the slots (S14). Next, each slot table creation unit 15 creates a list of user inquiry items and creates a slot table for each user inquiry item (S15).

[0030] If there is information that is insufficient for each slot table creation unit 15 to create a slot table (S16), the inter-slot relationship addition and correction unit 16 performs processing to add and correct the information that is insufficient for creating (completing) the slot table (S17). Based on the addition and correction information passed from the inter-slot relationship addition and correction unit 16, the inter-slot relationship extraction unit 13 adds and corrects the relationships between the slots, the slot relationship tree construction unit 14 reconstructs the slot relationship tree, and each slot table creation unit 15 completes the slot table that was temporarily suspended.

[0031] When each slot table creation unit 15 has completed creating the slot table because there is no information that is missing for creating the slot table, the slot table integration unit 17 integrates the slot tables (S18). The dialogue knowledge output unit 18 outputs the integrated slot table as dialogue knowledge (S19).

[0032] (A-2-2) Detailed Operation of Slot Recognition Unit 12 (Details of Step S12) FIG. 3 is a flowchart showing a detailed operation of the slot recognition unit according to the embodiment.

[0033] The slot recognition unit 12 recognizes slot candidates in a parent-child relationship from definitions and case distinctions in the input document (S12-1). The parent-child relationship here means a relationship in which the slot is the "parent" and the value of the slot is the "child." When recognizing a slot or a parent and a child, if it is necessary to analyze the division of a phrase or a dependency relationship due to a compound word, a noun phrase, a verb phrase, or a dependent modifier, the slot recognition unit 12 can use ordinary natural language processing techniques such as morphological analysis and syntactic analysis.

[0034] The slot recognition unit 12 judges whether or not the recognition of the slot candidates is complete (S12-2). For example, if the child slot candidates still have definitions or case distinctions, the process returns to the above-mentioned step S12-1, and the child slot candidates are made into parents, and the process of recognizing the child slot candidates into parent-child slot candidates is repeated. When the slot recognition unit 12 has recognized all the slot candidates in the text, it performs the following process.

[0035] The slot recognition unit 12 groups the recognized identical slot candidates together and defines the slot and the value (S12-3).

[0036] 9 is an explanatory diagram showing an example of a document input to the dialogue knowledge creating device according to the embodiment. A specific example of recognizing slot candidates from the input document 100 in FIG.

[0037] For example, from the part of the input document 100 that states the definition, "A day trip is ... divided into trip A, trip B, and trip C," slot candidates such as "day trip," "category," "trip A," "trip B," and "trip C" are recognized, and the slot candidates are grouped together to define the slot "day trip category" and the values ​​"trip A," "trip B," and "trip C," with "day trip" and "category" as the parents and "trip A," "trip B," and "trip C" as the children.

[0038] In addition, from the part that describes the case distinction, for example, "...a day trip allowance is paid" and "...a day trip allowance is not paid", slot candidates such as "day trip allowance", "paid", "yes", and "no", are recognized, with "day trip allowance" and "paid" as the parents and "yes" and "no" as the children, and by replacing "yes" with "yes" and "no" with "no" for naming purposes, the slot candidates "day trip allowance", "paid", "yes", and "no" are combined to define the slot "day trip allowance paid or not" and the values ​​"yes" and "no".

[0039] Fig. 10 is an explanatory diagram showing an example of slots and their values ​​defined by the slot recognition unit according to the embodiment. Fig. 10 shows a slot definition 200 defined based on the input document 100 of Fig. 9. The name of the slot may be determined from the notation in the document, or if there is a hierarchy based on chapters or the like, the name may be determined using the notation of a higher hierarchy. In any case, the method of naming the slot is not particularly limited.

[0040] (A-2-3) Detailed Operation of the Slot Relation Extraction Unit 13 (Details of Step S13) FIG. 4 is a flowchart showing a detailed operation of the inter-slot relation extraction unit according to the embodiment.

[0041] If there are slots recognized by the slot recognition unit 12, the inter-slot relationship extraction unit 13 performs the processing from the next step S13-2 onwards (processing to extract relationships between slots).On the other hand, if the processing to extract relationships between slots has been performed for all recognized slots, the series of processes is terminated (S13-1).

[0042] The inter-slot relation extraction unit 13 judges whether or not there is a dependency relationship, in which the value of one slot is determined when the value of another slot is determined, for the slots recognized in the input document (S13-2). If the above-mentioned relationship exists, the inter-slot relation extraction unit 13 performs the following process, whereas if the above-mentioned relationship does not exist, the process returns to the process of step S13-1 described above.

[0043] If the inter-slot relationship extraction unit 13 finds a relationship in the input document in which the value of one slot is determined when the value of another slot is determined in the slots recognized in the input document in step S13-2, the inter-slot relationship extraction unit 13 extracts the relationship as a relationship between the slots (S13-3). After extracting the relationship between the slots, the inter-slot relationship extraction unit 13 returns to the process of step S13-1 (performs similar processing to extract the relationships between the slots for other slots).

[0044] In addition, the inter-slot relationship extraction unit 13 extracts the inter-slot relationships for all slots recognized from the input document, and then terminates the series of processes. However, if information on additionally modified relationships between slots is passed from the inter-slot relationship addition and correction unit 16 (described later), the unit performs a process of additionally correcting the relationships between the slots based on that information (S13-4).

[0045] The relationship between the slots mentioned above can be expressed, for example, by a "condition sequence" and a "result sequence." A condition sequence consists of "slot = value (AND slot = value ... AND slot = value)." A result sequence consists of "slot = value." The relationship between slots is that once the condition sequence is determined, the result sequence is also determined.

[0046] Fig. 11 is an explanatory diagram showing an example of the relationship between slots extracted by the slot relationship extraction unit according to the embodiment. For example, as shown in row 302 of the relationship between slots 300 in Fig. 11, if "whether or not express train was used" is determined to be "yes" and "total" is determined to be "Ekm or more", then "day trip business trip category" is determined to be "business trip A". Therefore, the condition column is "whether or not express train was used=yes AND total=Ekm or more" and the result column is "day trip business trip category=business trip A".

[0047] (A-2-4) Detailed Operation of Slot Relation Tree Construction Unit 14 (Details of Step S14) FIG. 5 is a flowchart showing a detailed operation of the slot relation tree constructing unit according to the embodiment.

[0048] If there is a relationship between the slots extracted by the slot relationship extraction unit 13, the slot relationship tree construction unit 14 performs the processing from the next step S14-2 onwards (processing to associate the relationships between the slots); on the other hand, if the processing to associate the relationships between the slots has been performed for all the extracted slots, the slot relationship tree construction unit 14 terminates the processing (S14-1).

[0049] The slot relationship tree construction unit 14 associates the slot in the result column of the relationship between slots (for example, the relationship between slots 300 in FIG. 11) with the slot in the condition column as the "child" and the slot in the condition column as the "parent" (S14-2).

[0050] The slot relation tree construction unit 14 judges whether or not the slot in the associated condition string is in the result string of another slot (S14-3). If the slot in the condition string is in the result string of another slot, the slot relation tree construction unit 14 returns to the above-mentioned step S14-2 and performs child-to-parent association with another slot as the parent. If there are no more slots that can be associated (if the slot in the condition string is not in the result string of another slot), the slot relation tree construction unit 14 returns to the above-mentioned step S14-1.

[0051] The slot relation tree construction unit 14 constructs a tree structure of slots by repeatedly tracing the parent-child relations as described above. Note that, since the relations between slots may include AND conditions, there may be multiple parents.

[0052] FIG. 12 is an explanatory diagram showing an example of a slot relationship tree constructed by the slot relationship tree construction unit according to the embodiment.

[0053] In the slot relationship tree 400 in Fig. 12, the association between slots is indicated by solid lines. For example, the slot (parent) "whether or not to use express train" in the condition column and the slot (child) "total" in the result column are connected by a solid line, indicating that there is a relationship between the slots.

[0054] Also, for example, the slot "Day Trip Business Trip Daily Allowance Payment" with a value of "3,000 yen" in the result column of row 311 of the relationship between slots 300 in Fig. 11 can be traced to the slot "Day Trip Business Trip Category" with a value of "Business Trip A" in the condition column. In this case, the "Day Trip Business Trip Category" which is the slot (parent) in the condition column and the "Day Trip Business Trip Daily Allowance Payment" which is the slot (child) in the result column are associated (connected with a solid line).

[0055] Also, for example, from the slot "One-day business trip category" with the value "Business trip A" in the result column of rows 302 to 304 of the relationship between slots 300 in Fig. 11, the slot "Total" with the value "E km or more" in the condition column, the slot "One-way" with the value "D km or more" in the condition column, and the slot "One-way travel time" with the value "F hours or more" in the condition column can be traced. As above, the relationship between the slots is shown by connecting the "One-day business trip category" with "Total" or "One-way" or "One-way travel time" respectively with a solid line.

[0056] (A-2-5) Detailed Operation of Each Slot Table Creation Unit 15 (Detailed Operation of Steps S15 and S16) FIG. 6 is a flowchart showing a detailed operation of the slot relation tree constructing unit according to the embodiment.

[0057] To create a slot table, each slot table creation unit 15 first creates a user inquiry item list from the relationships between slots (S15-1). A user inquiry item list is a list in the same format as a slot definition, in which the slot of the user inquiry item is the parent and the value of the user inquiry item is the child (for example, FIG. 13). A user inquiry item is a slot in the result column of the relationship between slots. For example, in the example of FIG. 12, the values ​​of the user inquiry items are "Day trip business trip type," "Whether or not daily allowance for day trip business trip is paid," and "Amount of daily allowance for day trip business trip paid."

[0058] Next, each slot table creation unit 15 performs the processing from the next step S15-2 onwards (slot table creation processing) for the values ​​in the user inquiry item list created in the above-mentioned step S15-1, but if the processing has been performed for all values ​​in the user inquiry item list, the series of processing ends (S15-2).

[0059] Each slot table creation unit 15 creates a format of a slot table (S15-3). The format of the slot table is such that for each value of the user inquiry item, the column name of the first row and first column is "user inquiry item", the value of the user inquiry item is entered in the first column of the second row, and the slots are traced from parent to child in the slot relationship tree constructed to arrive at the value of the user inquiry item, arranged from left to right, and the column names of the second column of the first row and subsequent columns are set. Values ​​are sequentially filled in this table to complete the slot table. For example, in FIG. 15, "whether or not to use express train", "total", "one-way trip", and "one-way travel time" are shown as column names in the second column of the first row and subsequent columns to arrive at the value of the user inquiry item, "one-day business trip category".

[0060] Next, each slot table creation unit 15 attaches a presentation information mark to the presentation information slot in the slot table, and attaches an answer required mark and a dialogue order to the answer required slot (S15-4).

[0061] In each slot from the second row onwards in the slot table, information is written according to the slot breakdown as shown in Fig. 14. In the slot table, the final answer to be output to the user is treated as presentation information. A slot that stores presentation information is a presentation information slot. In Fig. 14, slot 605 is a presentation information slot (a slot marked "value" and shaded to the left).

[0062] Moreover, a slot that requires an answer value is an answer-required slot. An answer-required slot is given a number indicating the order of dialogue along with a mark indicating that an answer is required (for example, an asterisk "*"). An answer-required slot with a smaller number is always processed first. In FIG. 14, slot 602 (a slot marked with "* number") is an answer-required slot. Furthermore, a slot whose value is determined by an answer is an answered slot. In FIG. 14, slot 603 is an answered slot (a slot marked with "value"). Furthermore, a slot whose value is determined based on the relationship between slots is an inference result slot. In FIG. 14, slot 604 is an inference result slot (a slot marked with "value" and lined to the right). Moreover, a slot that does not require an answer value because the value is not important is an indefinite value slot. In FIG. 14, slot 601 is an indefinite value slot (a blank slot). Note that the presentation information slot that stores the value of the inference result is slot 605 (a slot marked with "value" and lined to the left and right).

[0063] Each slot table creation unit 15 marks (for example, "*") the slots in the second row of the slot table whose column name matches the value of the user inquiry item as "presentation information slots", marks the slots on the left side of the line of the slot relationship tree as answer-required slots, and assigns a dialogue order so that the numbers increase from left to right of the slots. For example, in the second row (row 702) of the slot table 700 in Fig. 15, the slots indicating "whether to use express train", "total", "one-way trip", and "one-way travel time" are answer-required slots, and are indicated by "*1" to "*4".

[0064] Next, each slot table creation unit 15 performs processing such as filling in values ​​according to the dialogue order of the answer-required slots created in the above-mentioned step S15-4, and making the slots with filled values ​​into answered slots (S15-5). More specifically, in the third row and onwards of the slot table, each slot table creation unit 15 fills in values ​​in slots marked as answer-required, thereby making the answer-required slot an answered slot, making another slot an inference result slot, erasing the answer-required mark of another answer-required slot, or marking a new answer-required slot to a slot that was not an answer-required slot. The state in which the slot value or mark has been changed is added to the slot table in sequence. As shown below, adding rows is repeated until values ​​are filled in the presentation information slots.

[0065] Next, while creating the slot table, each slot table creation unit 15 determines whether there is information that is insufficient (missing information) to complete the slot (S15-6). If there is missing information, each slot table creation unit 15 proceeds to the process of step S15-8, which will be described later. On the other hand, if there is no missing information, each slot table creation unit 15 executes the process of the next step S15-7.

[0066] Each slot table creation unit 15 determines whether or not a value has been filled in the presentation information slot (15-7). If a value has not been filled in the presentation information slot, each slot table creation unit 15 returns to the above-mentioned step S15-5, and creates a new slot row and adds it to the slot table until a value is filled in the presentation information slot of the new slot row. On the other hand, if a value has been filled in the presentation information slot, each slot table creation unit 15 returns to the above-mentioned step S15-1, and repeats creating the slot table format and creating the slot table for the values ​​of the other user inquiry item lists in the same way (i.e., for the values ​​of all other user inquiry item lists).

[0067] Furthermore, if each slot table creation unit 15 finds information that is insufficient for creating a slot table in the above-mentioned step S15-6, it passes the found information (information on inter-slot relationships and slot relationship tree related to the parts of the slot table that could not be filled in) to the inter-slot relationship addition and correction unit 16 (S15-8).

[0068] FIG. 15 is an explanatory diagram (part 1) showing an example of a slot table (a slot table created with the user inquiry item set to "one-day business trip category") according to the embodiment.

[0069] When the user inquiry item is “day trip business trip category,” in the second row (row 702) of FIG. 15 (slot table 700), “day trip business trip category” is the presented information slot, and “whether to use express train,” “total,” “one-way,” and “one-way travel time” on the left side of the line in the slot relationship tree of FIG. 12 are the slots requiring a response.

[0070] In the third row (row 703) of the slot table 700, if the value of "whether or not express train is used" is filled in, "whether or not express train is used" becomes an answered slot, and in the relationship between the slots in FIG. 11, when the value of "whether or not express train is used" is "yes", "total" becomes a slot requiring an answer, so when the value of "whether or not express train is used" is "no", the "total" becomes an indefinite value slot with the answer required mark removed. In the fourth row (row 704) of the slot table 700, if the value of "one-way" is filled in, "one-way" becomes an answered slot. In the fifth row (row 705) and sixth row (row 706) of the slot table 700, if the value of "one-way travel time required" is filled in, "one-way travel time required" becomes an answered slot, and in the relationship between the slots in FIG. 11, the value of "daily allowance category" is filled in, so "daily allowance category" becomes a presented information slot that stores the value of the inference result, and the value of the presented information slot is filled in.

[0071] In the case of the fifth row (row 705) and sixth row (row 706) of the slot table 700, the value of "one-way" is "D km or more" and the value of "day trip daily allowance category" is "business trip A." However, when the value of "one-way" is "G km or more and less than D km" in the seventh row (row 707) of the slot table 700 and the value of "one-way travel time" is "F hours or more" in the eighth row (row 708) of the slot table 700, the value of "day trip business trip category" for "one-way" value of "G km or more and less than D km" is "business trip B" in the relationship between the slots in the sixth row (row 306) of Fig. 11, and the value of "day trip business trip category" for "one-way" value of "F hours or more" is "business trip A" in the relationship between the slots in the fourth row (row 304) of Fig. 11, so that the value of the inference result slot is not determined to be one.

[0072] Here, if the "one-way" value is "G km or more but less than D km" and the "one-way travel time" value is "F hours or more", it is found that there is insufficient information, so each slot table creation unit 15 passes the found information to the inter-slot relationship addition and correction unit 16 (creation of the slot table is temporarily suspended). If there is no insufficient information, creation of the slot table for the user inquiry item "day trip daily allowance category" is completed.

[0073] (A-2-6) Detailed Operation of the Slot Relation Addition and Modification Unit 16 (Detailed Processing of Step S17) FIG. 7 is a flowchart showing a detailed operation of the inter-slot relationship addition and modification unit according to the embodiment.

[0074] When the inter-slot relation addition and correction unit 16 receives information that is insufficient to complete the slot table from each slot table creation unit 15, it presents the insufficient information to an expert on the document or the like (S17-1).

[0075] The inter-slot relationship addition and correction unit 16 prompts the knowledgeable person to confirm the missing information and input information to make up for the missing information (S17-2).

[0076] The inter-slot relationship addition and correction unit 16 adds and corrects the relationship between the slots (S73), and passes the addition and correction information to the inter-slot relationship extraction unit 13 (S17-3).

[0077] Fig. 16 is an explanatory diagram showing an example in which the inter-slot relationship adding and correcting unit according to the embodiment presents to an expert or the like information that is lacking for completing a slot table. Fig. 16 shows a dialogue knowledge creation device 1, as well as a display unit 2 for presenting the missing information to an expert or the like of a document, and an input unit 3 for inputting information to make up for the deficiency. In addition, the dialogue knowledge creation device 1 may present the missing information to an information terminal such as an expert's PC, tablet terminal, or smartphone via a network, and accept input of information to make up for the deficiency from these information terminals.

[0078] For example, when the inter-slot relationship addition and correction unit 16 asks the expert about the "day trip business trip category" in the case of the missing information "one-way = Gkm or more but less than Dkm AND one-way travel time = F hours or more", the inter-slot relationship addition and correction unit 16 presents the content to the expert on the missing information input / output screen 4 of the display unit 2. The expert inputs the answer to the question from the system (inter-slot relationship addition and correction unit 16) on the missing information input / output screen 4 using the input unit 3 (for example, inputting "business trip A"). In addition to directly inputting "business trip A", for example, buttons for "business trip A", "business trip B", and "business trip C" may be displayed and the button may be selected (for example, if the display unit 2 is a touch panel type, the screen may be touched to select). In any case, the input method is not particularly limited.

[0079] Furthermore, when a different relationship between slots is found in the content input by the expert, the inter-slot relationship addition and correction unit 16 requests the expert to correct the relationship between slots. For example, the missing information input / output screen 4 presents that when "one-way = G km or more but less than D km", "day trip business trip category = business trip B" is a relationship with a different interpretation. The expert corrects the relationship by adding "one-way travel time = not more than F hours" to the AND condition.

[0080] Fig. 17 is an explanatory diagram showing an example of an inter-slot relationship in which missing information has been repeatedly added and corrected by an inter-slot relationship addition and correction unit according to the embodiment. Inter-slot relationship 300A in Fig. 17 shows an example in which the fifth to seventh lines (lines 305 to 307) of inter-slot relationship 300 in Fig. 11 have been corrected. That is, an example is shown in which "one-way travel time = not more than F hours" has been added as an AND condition to the condition column in the fifth to seventh lines (lines 305A to 307A) of inter-slot relationship 300A.

[0081] Fig. 18 is an explanatory diagram showing an example of a completed slot table (a slot table for one-day business trips after the eighth row in Fig. 15 has been created) based on the corrected relationships between slots (Fig. 17). In row 708 in Fig. 15, the one-day business trip category could not be determined to be either "business trip A" or "business trip B", but in row 708A in Fig. 18, by correcting the relationships between slots, it can now be determined to be "business trip A".

[0082] (A-2-7) Detailed Operation of Slot Table Integration Unit 17 (Detailed Processing of Step S18) FIG. 8 is a flowchart showing a detailed operation of the slot table integration unit according to the embodiment.

[0083] The slot table integration unit 17 arranges the column names of all the slot tables created by each slot table creation unit 15, excluding duplicates, to create column names (S18-1), and arranges the rows of all the created slots according to the column names (S18-2).

[0084] Fig. 19 is an explanatory diagram showing an example of a slot table in which each slot table is integrated by a slot table integration unit according to the embodiment. Fig. 19 shows an example in which a slot table of "Day trip business trip category", a slot table of "Whether or not daily allowance for day trip business trip" and a slot table of "Day trip daily allowance amount" are integrated.

[0085] (A-3) Effects of the embodiment According to this embodiment, the following effects are obtained.

[0086] The dialogue knowledge creation device 1 can efficiently create dialogue knowledge by recognizing slots from documents, extracting relationships between slots, constructing a slot relationship tree, and creating a slot table. By using dialogue knowledge, slots requiring answers are determined once the user inquiry items are determined, so the necessary questions and the order can be determined from the dialogue system.

[0087] In addition, if a description is omitted in the document, and the document expert can understand it, but there is insufficient information to create a slot table, the missing information can be presented to the document expert and have them enter and add the information to create a slot table, even if the document does not contain enough information.

[0088] 20 is an explanatory diagram showing an example of a dialogue using dialogue knowledge created by the dialogue knowledge creating device according to the embodiment. The dialogue knowledge here is the integrated slot table 800 shown in FIG.

[0089] In the dialogue example 900 in FIG. 20, the first line is where the system (dialogue system) first asks the user what he or she wants to inquire about. The second line is where the user answers one of the user inquiry items. Since the user inquiry item is decided, the answer-required slot is decided, and the system asks the user a question on the third line. Lines 4 to 7 are where the user's answer and the system's question are repeated, and since the user's answer on line 8 fills in the presentation information slot, the system outputs line 9 to the user, and the dialogue ends.

[0090] When the answered slot is filled with a value from the user's answer, the answered slot and the inference result slot are filled with values ​​and another slot becomes a to-be-answered slot, and the answers to other questions are also determined, so it is possible to ask the user only the questions that are necessary. Also, by extracting user inquiry items from documents and creating and integrating multiple slot tables, it is possible to create dialogue knowledge that can be used in dialogues appropriate to each of multiple inquiries.

[0091] The dialogue can be conducted by a person or a dialogue device, so either a person or a machine can use dialogue knowledge to dialogue. Therefore, even if there is no response history or FAQ, it is possible to ask the questions necessary to present the answer to the user, and to directly answer the user's inquiry.

[0092] (B) Other embodiments The present invention is not limited to the above-described embodiment, and modified embodiments such as those exemplified below can also be mentioned.

[0093] (B-1) In the above embodiment, the dialogue knowledge creation device 1 is configured to include the inter-slot relationship addition and modification unit 16, but the configuration of the inter-slot relationship addition and modification unit 16 (and the processing related to the inter-slot relationship addition and modification unit 16) may be omitted.

[0094] That is, when the dialogue knowledge creation device 1 equipped with the inter-slot relation adding and correcting unit 16 found information that was insufficient to complete the slot table, it completed the slot table based on information given from outside, but when the input document has been defined and case-specificated without causing any doubt, the configuration and processing of the inter-slot relation adding and correcting unit 16 are unnecessary. Note that when a dialogue knowledge creation device that does not have the configuration of the inter-slot relation adding and correcting unit 16 finds information that is insufficient to complete the slot table, it may, for example, interrupt the processing of creating the slot table and output an error message.

[0095] (B-2) In the above embodiment, an example was shown in which information that is missing to complete the slot table is presented to an expert on the document and input and added / corrected. However, as a variant, the system (inter-slot relationship addition / correction unit 16) may add / correct the missing information through inference to create the slot table.

[0096] (B-3) In the above embodiment, if information that is insufficient to complete the slot table is found, the creation of the slot table is temporarily suspended, but the slot table may be temporarily completed without suspension. After that, the missing information may be presented to an expert on the document and input, and the relationship between the slots may be added and corrected to complete the slot table.

[0097] (B-4) In the above embodiment, no specific use was made of the dialogue knowledge (slot table) created by the dialogue knowledge creating device 1, but it can be used, for example, in a dialogue system that realizes inquiries similar to those handled by a person at a counter. Also, since an operator can ask questions using the dialogue system, even an inexperienced operator can use the dialogue knowledge to efficiently and accurately provide an answer.

[0098] In addition, dialogue knowledge created based on input documents such as encyclopedias and legal texts can be used as learning materials that can answer questions. In any case, there is no particular limit to how the dialogue knowledge can be used. [Explanation of symbols]

[0099] 1...dialogue knowledge creation device, 2...display unit, 3...input unit, 4...missing information input / output screen, 11...document input unit, 12...slot recognition unit, 13...inter-slot relationship extraction unit, 14...slot relationship tree construction unit, 15...slot table creation unit, 16...inter-slot relationship addition / correction unit, 17...slot table integration unit, 18...dialogue knowledge output unit, 100...input document, 200...slot definition, 300...relationship between slots, 400...slot relationship tree, 500...user inquiry item list, 600...slot breakdown, 700, 700A, 800...slot table, 900...dialogue example.

Claims

1. A document input means for accepting input of document information; a slot recognition means for acquiring a slot defined corresponding to the document information; an inter-slot relationship extraction means for extracting relationships between the acquired plurality of slots based on dependency relationships between the slots; a slot relation tree constructing means for constructing a tree-structured slot relation tree from the extracted inter-slot relations; a slot table creating means for creating a list of user inquiry items from the relationships between the slots and the slot relationship tree, and for creating a slot table for each value of the user inquiry items in the list; a slot table integration means for integrating the slot tables created for each value of the user inquiry item; a dialogue knowledge output means for outputting the integrated slot table as dialogue knowledge; A dialogue knowledge creating device comprising:

2. 2. A dialogue knowledge creating device according to claim 1, wherein each of said slot table creating means detects missing information which is information regarding a portion where said slot table cannot be created based on said relationships between said slots and said slot relationship tree.

3. 3. The dialogue knowledge creating device according to claim 2, further comprising an inter-slot relationship adding and correcting means having a missing information presenting unit for presenting said missing information to the outside.

4. The dialogue knowledge creation device according to claim 3, characterized in that the inter-slot relationship addition and correction means further comprises an information input unit that accepts input of information to supplement the missing information from the outside, and generates addition and correction information to correct the inter-slot relationship based on the information input to the information input unit.

5. The inter-slot relationship extraction means modifies the inter-slot relationship based on the additional modification information, The slot relation tree construction means reconstructs the slot relation tree based on the modified inter-slot relations, Each of the slot table creation means creates the slot table based on the corrected inter-slot relationships and the reconstructed slot relationship tree.

5. A dialogue knowledge creating device according to claim 4.

6. Computer, A document input means for accepting input of document information; a slot recognition means for acquiring a slot defined corresponding to the document information; an inter-slot relationship extraction means for extracting relationships between the acquired plurality of slots based on dependency relationships between the slots; a slot relation tree constructing means for constructing a tree-structured slot relation tree from the extracted inter-slot relations; a slot table creating means for creating a list of user inquiry items from the relationships between the slots and the slot relationship tree, and for creating a slot table for each value of the user inquiry items in the list; a slot table integration means for integrating the slot tables created for each value of the user inquiry item; a dialogue knowledge output means for outputting the integrated slot table as dialogue knowledge; A dialogue knowledge creation program characterized by functioning as a dialogue knowledge creation program.

Citation Information

Patent Citations

  • Message model conversion system

    JP2001265669A

  • FAQ candidate extraction system and FAQ candidate extraction program

    JP2012003704A

  • FAQ preparation support system and program

    JP2013050896A

  • Question and answer data generation device and question and answer data generation method

    JP2020080025A

  • Contextual Search Disambiguation

    US20160117360A1