JP2013225036A - Automatic interactive scenario creation support device and automatic interactive scenario creation support program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an automatic interaction scenario creation support device and an automatic interaction scenario creation support program which allow the reduction in operation cost of an automatic interactive system, quick creation of a novel type of scenario, and improvement of scenario quality.SOLUTION: An automatic interactive scenario creation support device 1 includes: a scenario information DB 11 in which scenario information constituted by associating type specification information for specifying types of scenarios, a plurality of conversational sentences, and phase specification information for specifying phases with each other is stored; a word weight information DB 13; a phase weight information DB 14; a word extraction unit 6; a type determination unit 7 which sums up word scores calculated by multiplying word weights by appearance frequencies of words and phase scores calculated by multiplying phase weights by the appearance frequencies of words to calculate scenario scores and determines a type of a scenario on the basis of the calculated scenario scores; and a scenario creation unit 8 which creates a scenario corresponding to an objective conversational sentence on the basis of the scenario information.

Description

  The present invention relates to an automatic dialog scenario creation support apparatus and an automatic dialog scenario creation support program that support creation of a scenario for an automatic dialog system.

  In recent years, an interactive voice response system (IVR: Interactive Voice Response) for the purpose of automatically accepting orders by voice from customers has been spreading. In this automatic voice response system, voice is output in accordance with the scenario information stored in the storage means, and when a response voice from a speaker is input, the response voice is voice-recognized, and based on the voice recognition result. Then, the next voice output is performed with reference to the scenario information, and the subsequent dialogue is similarly performed to provide information to the customer or collect information from the customer (for example, see Patent Document 1).

  In order to increase the accuracy of such a voice dialogue, it is necessary to create a scenario assuming the flow of the dialogue as accurately as possible. For this reason, conventionally, when creating a new type of scenario, a skilled person manually creates a scenario in consideration of his / her experience while referring to a past scenario.

Japanese Patent Laying-Open No. 2005-062398

  However, in the past, experts have created scenarios manually, so it takes time and effort to create scenarios, raises the operating costs of automatic dialog systems, and creates new types of scenarios quickly. It was difficult to start the operation. In addition, since such a scenario creation method has not been established, there is a problem that the quality of the scenario is not stable.

  The present invention has been made in view of the above, and by supporting the automatic creation of a scenario for an automatic dialogue system, the operation cost of the automatic dialogue system can be reduced, and a new type of scenario can be quickly developed. An object of the present invention is to provide an automatic dialogue scenario creation support apparatus and an automatic dialogue scenario creation support program that can be created and started to operate, and that can improve the quality of the scenario.

  The automatic dialogue scenario creation support device according to claim 1 is an automatic dialogue scenario creation support device that supports creation of a scenario for an automatic dialogue system, the scenario information relating to a plurality of scenarios, Mutually specifying type specifying information for specifying each type, a plurality of conversation sentences constituting each of the plurality of scenarios, and phase specifying information for specifying a phase indicating a stage of conversation in each of the plurality of scenarios, Scenario information storage means for storing scenario information configured in association with each other, a word included in the plurality of conversation sentences, the type specifying information, and a type of scenario in which the word is specified by the type specifying information Word weight information configured by associating a word weight indicating the weight of the word when used in a conversation sentence with each other A word weight information storage means for storing, a word included in the plurality of conversation sentences, the type specifying information, the phase specifying information, and a phase of the type in which the word is specified by the type specifying information. Phase weight information storage means for storing phase weight information that is formed by correlating the phase weight indicating the weight of the word when used in the conversation sentence of the phase specified by the specific information, When a target conversation sentence, which is a plurality of conversation sentences to be created, is specified by a predetermined method, a word extraction unit that extracts a word from the specified target conversation sentence, and a word extracted by the word extraction unit And the word weight corresponding to the word, and the word weight for each type specifying information is the word weight information. A phase weight corresponding to the word extracted by the word extracting means, obtained from the payment means, calculating a word score for each type specifying information by multiplying the acquired word weight by the number of appearances, and The phase weight for each type specifying information and the phase specifying information is obtained from the phase weight information storage unit, and the phase for each type specifying information and the phase specifying information is obtained by multiplying the acquired phase weight by the number of appearances. A scenario that calculates a score, calculates a scenario score by adding the calculated word score and the calculated phase score for each type specifying information, and a scenario that matches the target conversation sentence based on the calculated scenario score The type determining means for determining the type of the item and the type determining means Scenario creation means for obtaining scenario information corresponding to the type specifying information for specifying the type of scenario made from the scenario information storage means, and creating a scenario corresponding to the target conversation sentence based on the acquired scenario information; Is provided.

  The automatic dialogue scenario creation support device according to claim 2 is the automatic dialogue scenario creation support device according to claim 1, wherein the importance level of the word extracted by the word extraction means is determined by a predetermined method. The scenario information corresponding to the word determined to be is not included in the scenario information acquired from the scenario information storage means by the scenario creation means, and can be acquired from the scenario information storage means In some cases, the scenario information is acquired from the scenario information storage means, and based on the acquired scenario information, the scenario creation means supplements the scenario information acquired from the scenario information storage means by the first scenario. Complementary means are provided.

  The automatic dialog scenario creation support apparatus according to claim 3 is the automatic dialog scenario creation support apparatus according to claim 1 or 2, wherein the importance level of the word extracted by the word extraction unit is determined by a predetermined method. The scenario information corresponding to the word determined to be important is not included in the scenario information acquired from the scenario information storage means by the scenario creation means, and is acquired from the scenario information storage means If not possible, the scenario information is acquired from the target conversation sentence, and the scenario information acquired from the scenario information storage means by the scenario creation means is supplemented based on the acquired scenario information. Means.

  The automatic dialogue scenario creation support program according to claim 4 is an automatic dialogue scenario creation support program that supports creation of a scenario for an automatic dialogue system, and is scenario information relating to a plurality of scenarios, Mutually specifying type specifying information for specifying each type, a plurality of conversation sentences constituting each of the plurality of scenarios, and phase specifying information for specifying a phase indicating a stage of conversation in each of the plurality of scenarios, Scenario information storage means for storing scenario information configured in association in the scenario information storage means, words included in the plurality of conversation sentences, the type specification information, and the words are specified by the type specification information. The word weights that indicate the weight of the word when used in a conversation sentence of a certain type of scenario. Word weight information storage means for storing the word weight information configured in the manner described above, the words included in the plurality of conversation sentences, the type specifying information, the phase specifying information, and the word in the type specifying information Stores phase weight information that is formed by correlating the phase weight indicating the weight of the word when used in the conversation sentence of the phase specified by the phase specifying information in the specified type of scenario. A word that extracts a word from a specified target conversation sentence when a target conversation sentence that is a plurality of conversation sentences to be created in a scenario is specified by a predetermined method on a computer comprising phase weight information storage means The number of appearances of the word extracted by the extraction means and the word extraction means is counted, and a word weight corresponding to the word is obtained. The word weight for each type specifying information is acquired from the word weight information storage unit, the word weight for each type specifying information is calculated by multiplying the acquired word weight by the number of appearances, and the word extracting unit Phase weights corresponding to the extracted words and for each type specifying information and for each phase specifying information are obtained from the phase weight information storing means, and the obtained phase weights are multiplied by the number of appearances. The phase score for each type specifying information is calculated for each type specifying information, and the scenario score is calculated by adding the calculated word score and the calculated phase score for each type specifying information. Scenario that matches the target conversation sentence based on the scenario score The scenario determination unit corresponding to the type determination unit that determines the type of the scenario and the type specification information that specifies the type of scenario determined by the type determination unit is acquired from the scenario information storage unit, and the acquired scenario information is included in the acquired scenario information. Based on this, it is operated as scenario creating means for creating a scenario corresponding to the target conversation sentence.

  According to the automatic dialogue scenario creation support device according to claim 1 or the automatic dialogue scenario creation support program according to claim 4, a scenario corresponding to the target conversation sentence is created based on the scenario information acquired from the scenario information storage means. Compared to the case where a skilled person manually creates a scenario as in the past, the operation cost of the automatic dialogue system can be reduced, and a new type of scenario can be created quickly and operated. And the quality of the scenario can be improved.

  According to the automatic dialogue scenario creation support apparatus according to claim 2, scenario information corresponding to a word determined to be important can be acquired from the scenario information storage means and can be supplemented. By adding scenario information corresponding to important words to the information, a new scenario that matches the target conversation sentence can be created.

  According to the automatic dialogue scenario creation support apparatus according to claim 3, scenario information corresponding to a word determined to be important can be acquired from the target conversation sentence and complemented. By adding scenario information corresponding to an important word, it is possible to create a new scenario that matches the target conversation sentence.

It is a block diagram which shows functionally the automatic dialog scenario creation assistance apparatus which concerns on embodiment of this invention. It is a figure which shows the structural example of scenario information. It is a figure which shows the structural example of unnecessary word information. It is a figure which shows the structural example of word weight information. It is a figure which shows the structural example of phase weight information. It is a flowchart of an automatic dialog scenario creation process. It is a flowchart of a word extraction process. It is a figure which illustrates an object conversation sentence. It is a flowchart of a type determination process. It is a figure which illustrates a score totaling table. It is a figure which illustrates a scenario score totaling table. It is a flowchart of a scenario creation process. It is a figure which illustrates the scenario information created automatically. It is a figure which illustrates scenario information in which information was partially updated.

  Exemplary embodiments of an automatic dialogue scenario creation support apparatus and an automatic dialogue scenario program according to the present invention will be explained below in detail with reference to the accompanying drawings. First, after explaining the basic concept of the present embodiment, the specific contents of the present embodiment will be described, and finally, a modification to the present embodiment will be described. However, the present invention is not limited by the present embodiment.

[Basic concept of the present embodiment]
First, the basic concept of this embodiment will be described. The automatic dialogue scenario creation support apparatus and the automatic dialogue scenario program according to the present embodiment are for supporting creation of a scenario for an automatic dialogue system. Here, the “automatic dialogue system” is a system for automatically conducting a dialogue between a plurality of dialogue persons. The “plurality of interlocutors” is typically one person and one automatic dialogue apparatus, but it is sufficient that at least one automatic dialogue apparatus is included, and a plurality of persons are included. In some cases, a plurality of automatic interactive devices may be included. “Automatic dialogue” means that at least a part of the dialogue is automatically (mechanically) performed. Typically, in the dialogue between one human and one automatic dialogue device, automatic This means that the dialog part of the dialog device is automatically performed, but includes automatically performing all parts of the dialog as in the case of a dialog between two automatic dialog devices, for example. The “dialogue” is typically a voice dialogue, but includes a display (image) dialogue in which a dialogue is performed via a text display. “Scenario for automatic dialogue system (hereinafter simply“ scenario ”)” means at least a dialogue person who performs an automatic dialogue, a conversation sentence expected to be conducted by each conversation person, and an order of the conversation sentence. It is information for specifying. “Supporting scenario creation” includes not only automatically generating the entire scenario but also automatically generating a part of the scenario.

[Specific contents of this embodiment]
Next, specific contents of the present embodiment will be described. First, the configuration of the automatic dialog scenario creation support apparatus according to the present embodiment will be described, and then the processing performed by the automatic dialog scenario creation support apparatus will be described.

(Constitution)
First, the configuration of an automatic dialogue scenario creation support apparatus (hereinafter referred to as support apparatus) according to the present embodiment will be described. FIG. 1 is a block diagram conceptually showing the support device according to the present embodiment. The support device 1 is configured by connecting an input unit 2, an output unit 3, a control unit 4, and a storage unit 5 in terms of functional concept. For example, the support device 1 is a known desktop computer, notebook computer, or tablet computer. Or, it is configured using a smartphone.

  The input unit 2 is an input unit for inputting various types of information to the support apparatus 1. The specific configuration of the input unit 2 is arbitrary, and can be configured as, for example, a mouse, a keyboard, or a touch panel.

  The output unit 3 is output means for outputting various information from the support apparatus 1. The specific configuration of the output unit 3 is arbitrary. For example, the output unit 3 can be configured as a flat panel display such as a known liquid crystal display or organic EL display.

  The control unit 4 includes a CPU (Central Processing Unit), various programs interpreted and executed on the CPU (control programs such as an OS, programs that define various processing procedures, etc.), required programs and required data. It is configured with a cache memory for storing. Various programs interpreted and executed on the CPU include an automatic dialogue scenario creation support program (hereinafter referred to as a support program), and this support program is stored in an arbitrary storage medium including a CD-ROM and a DVD, for example. After that, it is installed, stored in the storage unit 5 in a nonvolatile manner, and interpreted and executed by the CPU, thereby constituting the substantial function of the control unit 4.

  The control unit 4 includes a word extraction unit 6, a type determination unit 7, a scenario creation unit 8, a first scenario supplement unit 9, and a second scenario supplement unit 10 in terms of functional concept. The word extraction unit 6 is a word extraction unit that extracts a word from the specified target conversation sentence when the target conversation sentence, which is a plurality of conversation sentences to be a scenario creation target, is specified by a predetermined method. The number of appearances of the words extracted by the type determining unit 7 and the word extracting unit 6 is counted, and word weights corresponding to the words and word weights for each type specifying information are acquired from the word weight information DB 13 described later, A word score for each type specifying information is calculated by multiplying the obtained word weight by the number of appearances, and the phase weight corresponding to the word extracted by the word extracting unit 6 is set for each type specifying information and for each phase specifying information. Is obtained from the phase weight information DB 14 to be described later, and the phase score for each type specifying information and each phase specifying information is calculated by multiplying the acquired phase weight by the number of appearances, and the calculated word score and the calculation are calculated. The scenario score is calculated by adding the phase scores for each type-specific information, A type determining means for determining the type of scenario that matches the target sentence based on the calculated scenarios score. The scenario creating unit 8 acquires scenario information corresponding to the type specifying information for specifying the type of scenario determined by the type determining unit 7 from a scenario information DB 11 described later, and based on the acquired scenario information, the target conversation This is a scenario creation means for creating a scenario corresponding to a sentence. The first scenario complementing unit 9 determines the importance of the word extracted by the word extracting unit 6 by a predetermined method, and scenario information corresponding to the word determined to be important is If it is not included in the scenario information acquired from the information DB 11 and can be acquired from the scenario information DB 11, the scenario information is acquired from the scenario information DB 11, and based on the acquired scenario information The first scenario complementing means supplements the scenario information acquired from the scenario information DB 11 by the scenario creation unit 8. The second scenario complementing unit 10 determines the importance of the word extracted by the word extracting unit 6 by a predetermined method, and scenario information corresponding to the word determined to be important is When it is not included in the scenario information acquired from the information DB 11 and cannot be acquired from the scenario information DB 11, the scenario information is acquired from the target conversation sentence, and based on the acquired scenario information, This is a second scenario complementing means for complementing the scenario information acquired from the scenario information DB 11 by the scenario creating unit 8. The function of each unit will be described together with each process.

  The storage unit 5 is a storage unit that stores information and parameters necessary for various processes executed by the support apparatus 1 in a nonvolatile manner, and is configured by, for example, an HD (Hard Disk) or a semiconductor memory. The storage unit 5 stores a scenario information database (hereinafter referred to as DB) 11, an unnecessary word information DB 12, a word weight information DB 13, and a phase weight information DB 14.

  The scenario information DB 11 is scenario information storage means for storing scenario information regarding a plurality of scenarios. As shown in the configuration example of FIG. 2, the scenario information includes information corresponding to the item “dialogue”, information corresponding to the item “conversation sentence”, information corresponding to the item “scenario type”, and item Information corresponding to “phase” and information corresponding to the item “variable” are associated with each other. The information corresponding to the item “interactive person” is interactive person specifying information for specifying the interactive person of the conversation sentence. In the present embodiment, it is assumed that the interlocutor is one customer who is a human and one automatic dialog device. In FIG. 2, the customers are “CU” and “automatic dialog device”. Are indicated as “OP”, respectively. The information corresponding to the item “conversation sentence” is a plurality of conversation sentences constituting each of the plurality of scenarios. Information corresponding to the item “scenario type” is type specifying information for specifying each type of a plurality of scenarios. This scenario type setting method is arbitrary, but in this embodiment, it is assumed that the scenario is a scenario for accepting a product order from a customer, and the scenario type is determined depending on the type of product. It is set. For example, in FIG. 2, “Shoe Order 1” indicates that the scenario type is a scenario type for receiving an order for shoes (shoes). In addition, even if the scenario is for accepting the same or similar product order, assuming that there are cases where there are a plurality of different types of scenarios, for example, “Shoe Order 1” and “Shoe Order 2” It means different types of scenarios. The information corresponding to the item “phase” is phase specifying information for specifying a conversation stage (phase) in each of a plurality of scenarios. The method of setting the phase is arbitrary, but for example, it is set by “phase name” or “phase name (phase variable)”. The “phase name” is the name of the phase, and the “phase variable” is information to be provided or acquired by a conversation sentence in the phase. Specifically, in FIG. 2, “START”, “customer confirmation (order history)”, “customer confirmation (name)”, and the like are set as phases. In phase = “customer confirmation (order history)”, phase name = “customer confirmation” and phase variable = “order history”. Note that different phases may be set for different types of scenarios. The information corresponding to the item “variable” is information to be provided or acquired by a conversation sentence. For example, when the scenario is a scenario for receiving a product order from a customer, the ordered product is specified. Information (product name, quantity, size, etc.), information for identifying the customer who placed the order (name, phone number, etc.), information for identifying the payment method of the order (payment method, etc.) Including. As described above, the records constituting the scenario information are stored in the order of conversation. Therefore, the phases of the records are also arranged in the order of conversation. The scenario information configured as described above is manually created by an expert at the beginning of the introduction of the support device 1 and stored in the scenario information DB 11 after that, and then automatically generated by the processing described later and the scenario information DB 11. Accumulated in.

  The unnecessary word information DB 12 is scenario information storage means for storing unnecessary word information. As shown in the configuration example of FIG. 3, the unnecessary word information is configured by associating information corresponding to the item “unnecessary word” and information corresponding to the item “classification” with each other. The information corresponding to the item “unnecessary word” is information for specifying the unnecessary word. An “unnecessary word” is a word that can be included in a target conversation sentence to be described later and is considered unnecessary for a scenario. For example, as information for specifying an unnecessary word, a word itself (“center” or “phone” in the example of FIG. 3) considered to be unnecessary in a scenario, or a part of speech attribute ( The result of morphological analysis (in the example of FIG. 3, “prefix”, “impression verb”, etc.) are stored. The information corresponding to the item “classification” is information indicating the classification of unnecessary words. As information indicating such a classification, “attribute” and “noun” are shown in the example of FIG. Unnecessary word classification = “attribute” indicates that the part of speech attribute (result of morphological analysis) is stored as information corresponding to the item “unnecessary word”, not the word itself, and has the part of speech attribute. The word indicates that it is not necessary for the scenario. The classification of the unnecessary word = “noun” indicates that the word itself is stored as information corresponding to the item “unnecessary word”, and indicates that the word is unnecessary for the scenario. The unnecessary word information configured as described above is arbitrarily set by the administrator of the support apparatus 1 and stored in the unnecessary word information DB 12.

  The word weight information DB 13 is word weight information storage means for storing word weight information. As shown in the configuration example of FIG. 4, the word weight information includes information corresponding to the item “word”, information corresponding to the item “scenario type”, and information corresponding to the item “word weight”. It is configured in association with. The information corresponding to the item “word” is a word set to be included in the conversation sentence. The information corresponding to the item “scenario type” is the same as the information corresponding to the same item name in FIG. Information corresponding to the item “word weight” is a word weight. “Word weight” is word weight specifying information indicating the weight of a word when used in a conversation sentence of the type of scenario specified by the type specifying information. For example, in the top record of FIG. 4, when the word = “size” is used in the scenario of the scenario type = “shoes order 1”, the word weight of the word = “size” = It indicates “1.00”. In the present embodiment, it is assumed that the larger the word weight value, the greater the weighting of the word (the importance of the word is greater). As shown in FIG. 4, even if the word is the same, the word weight may be different depending on the type of scenario in which the word is used. Different word weights are set. The word weight information configured as described above is arbitrarily set by the administrator of the support apparatus 1 and stored in the word weight information DB 13.

  The phase weight information DB 14 is phase weight information storage means for storing phase weight information. As shown in the configuration example of FIG. 5, the phase weight information includes information corresponding to the item “word”, information corresponding to the item “scenario type”, information corresponding to the item “phase”, and an item “ Information corresponding to “phase weight” is configured to be associated with each other. Information corresponding to the item “word”, information corresponding to the item “scenario type”, and information corresponding to the item “phase” are the same as the information corresponding to the same item name in FIGS. The information corresponding to the item “phase weight” is the phase weight indicating the weight of the word when the word is used in the conversation sentence of the phase specified by the phase specifying information in the type of scenario specified by the type specifying information. Specific information. For example, in the record at the top of FIG. 5, when the word = “size” is used in the phase = “START” in the scenario of the scenario type = “shoe order 1”, the word = “size” phase. It indicates that the weight = “0.00”. In the present embodiment, it is assumed that the larger the phase weight value, the greater the weighting of the word (the importance of the word is greater). As shown in FIG. 5, even if the word is the same, the phase weight may vary depending on the type and phase of the scenario in which the word is used. The same or different phase weights are set for each phase. The phase weight information configured as described above is arbitrarily set by the administrator of the support apparatus 1 and stored in the phase weight information DB 14.

(processing)
Next, an automatic dialogue scenario creation process performed in the support device 1 configured as described above will be described. In the following description, it is assumed that the processing is performed by the control unit 4 except for the subject to be specially described. “Step” is abbreviated as “S”. FIG. 6 is a flowchart of the automatic dialogue scenario creation process. This process is activated by a predetermined method via the input unit 2 of the support device 1.

(Processing-word extraction processing)
First, a word extraction process (SA1) is executed. FIG. 7 is a flowchart of word extraction processing. In this process, the word extraction unit 6 stands by until the target conversation sentence is specified (SB1). A target conversation sentence is a plurality of conversation sentences to be created as a scenario. For example, by recording a conversation sentence by telephone between a customer and an order reception operator, Created. When such a target conversation sentence is input to the support apparatus 1 via the input unit 2 in a predetermined format, the word extraction unit 6 determines that the target conversation sentence has been specified. FIG. 8 is a diagram illustrating such a target conversation sentence. The target conversation sentence is configured by associating information corresponding to the item “conversator” and information corresponding to the item “conversation sentence” with each other. The information corresponding to the item “dialogue” and the information corresponding to the item “conversation sentence” are the same as the information corresponding to the same item name in FIG.

  When the target conversation sentence is specified (SB1, Yes), the word extraction unit 6 extracts a word from the specified target conversation sentence (SB2). Specifically, a known morphological analysis or dependency analysis is performed on the target conversation sentence, and words included in the target conversation sentence are extracted. At this time, the part-of-speech attribute of each word is also specified by a known morphological analysis or dependency analysis.

  Next, the word extraction unit 6 excludes unnecessary words from the words extracted in SB2 (SB3). Specifically, the word extraction unit 6 determines whether or not unnecessary words stored in the unnecessary word information DB 12 are included in the words extracted in SB2, and if included, deletes the unnecessary words. To do. For example, the subject conversation sentence contains the sentence “Thank you for your call. ABC Co., Ltd. Order Center. Are you ordering?”, And the word extracted in SB2 for this sentence is “O ( "Prefix)" "Telephone (noun)" "Thank you (impression verb)" "There (auxiliary verb)" "Mas (auxiliary verb)" ". (Symbol)" "ABC Co., Ltd. (proprietary noun)" "Center (noun)" "Is (auxiliary verb)" ". (Symbol)" "Go (prefix)" "Order (noun)" "Doshi (auxiliary verb)" "U (auxiliary verb)" "Ka (particle)" (Symbol) ”(note that the part of speech attribute of each word is in parentheses). In this case, referring to the unnecessary word information illustrated in FIG. 3, the unnecessary word information includes a record of unnecessary word = “prefix” and classification = “attribute”, and a word having a prefix as a part-of-speech attribute is Since it is understood that the scenario is unnecessary, “o (prefix)” is determined to be unnecessary and is deleted. For the same reason, "Thank you (impression verb)" "There (auxiliary verb)" "Masu (auxiliary verb)" ". (Symbol)" "ABC Co., Ltd. (proprietary noun)" "It (auxiliary verb)" ". (Symbol) "" (Prefix) "" deho (auxiliary verb) "" u (auxiliary verb) "" ka (particle) "". (Symbol) "is determined to be unnecessary and is deleted. Further, referring to the unnecessary word information illustrated in FIG. 3, the unnecessary word information includes a record of unnecessary word = “telephone” and classification = “noun”, and the word “telephone” is unnecessary in the scenario. Since it is understood that there is, “center (noun)” is determined to be unnecessary and is deleted. For the same reason, it is determined that the “center (noun)” is unnecessary and is deleted. As a result, only “order (noun)” and “order (noun)” remain without being deleted. This completes the word extraction process.

(Processing-type determination processing)
Next, the type determination process (SA2) in FIG. 6 is executed. FIG. 9 is a flowchart of the type determination process. In this process, the type determination unit 7 calculates the number of appearances of the word extracted in the word extraction process (SC1). Specifically, the type determination unit 7 counts the number of extracted words for each word extracted in the word extraction process, and sets the number of extractions = the number of appearances. For example, when three words = “size” are extracted from the entire target conversation sentence, the type determining unit 7 counts the number of appearances of the word = “order” = “3”.

  Next, the type determination unit 7 obtains word weights corresponding to the words extracted by the word extraction process from the word weight information DB 13 for each type specifying information (SC2). Specifically, the type determination unit 7 acquires a word weight corresponding to the word finally extracted by the word extraction process from the word weights stored in the word weight information DB 13. At this time, when a plurality of word weights are stored in the word weight information DB 13 for each type specifying information for the same word, all the word weights are acquired together with the type specifying information. For example, in the example of the word weight information in FIG. 4, for the word = “size”, the scenario type = word weight of “Shoe Order 1” = “1.00”, and the scenario type = “Shoe Order 2” Weight = “1.00”, scenario type = word weight of “apparel order 1” = “1.00”, scenario type = word weight of “apparel order 2” = “1.00”, and scenario type = Get the word weight of "Sweets Order" = "0.60".

  Next, the type determining unit 7 calculates a word score for each type specifying information by multiplying the acquired word weight by the number of appearances (SC3). Specifically, the type determination unit 7 multiplies the word weight acquired in SC2 by the number of appearances acquired in SC1 for each word extracted in the word extraction process, thereby obtaining the word score of the word. Is calculated. Here, when a plurality of the same or different word weights are set for each type specifying information with respect to the same word, the plurality of word weights are respectively multiplied by the number of appearances. For example, regarding the word = “size”, the number of appearances calculated in SC1 = “3”, and the word weight acquired in SC2 is the word weight = “1.00” of the scenario type = “shoe order 1”. , Scenario type = word weight of “shoes order 2” = “1.00”, scenario type = word weight of “apparel order 1” = “1.00”, scenario type = word of “apparel order 2” When weight = “1.00” and scenario type = word weight of “Sweets Order” = “0.60”, scenario type = word score of “Shoes Order 1” = 1.00 × 3 = “ 3.00, scenario type = word score of “shoes order 2” = “3.00”, scenario type = word score of “apparel order 1” = “3.00”, scenario type = “apparel” Order two-word score = "3.00" ", and the scenario of the type = to calculate the" word score = 0.60 × 3 = "1.80 Suites orders. '"

  Next, the type determination unit 7 obtains, from the phase weight information DB 14, phase weights corresponding to the words extracted in the word extraction process and for each type specifying information and for each phase specifying information (SC4). Specifically, the type determination unit 7 acquires a phase weight corresponding to the word finally extracted by the word extraction process from the phase weights stored in the phase weight information DB 14. At this time, if a plurality of phase weights are stored in the phase weight information DB 14 for each type specifying information and each phase for the same word, all the phase weights are acquired together with the type specifying information and the phase specifying information. To do. For example, in the example of the phase weight information in FIG. 5, for the word = “size”, the scenario type = “shoes order 1” phase = “START” phase weight = “0.00”, and the scenario type = “ Phase of “Shoe Order 1” = Phase Weight of “Customer Confirmation (Name)” = “0.00”, etc. Scenario Type = Phase of “Shoe Order 2” = Phase Weight of “START” = “0.00” Scenario type = Phase of “Shoes order 2” = Phase weight of “Customer confirmation (name)” = “0.00”, etc. are acquired.

  Next, the type determining unit 7 calculates a phase score for each type specifying information and each phase specifying information by multiplying the acquired phase weight by the number of appearances of the word extracted in the word extraction process (SC5). Specifically, the type determination unit 7 multiplies the phase weight acquired in SC4 by the number of appearances acquired in SC1 for each word extracted in the word extraction process, thereby obtaining the phase score of the word. Is calculated. Here, for the same word, when the same or different phase weights are set for each type specifying information and each phase specifying information, by multiplying the plurality of phase weights by the number of appearances, A phase score is calculated for each type specifying information and each phase specifying information. For example, regarding the word = “size”, the number of appearances calculated in SC1 = “3”, and the phase weight acquired in SC4 is the phase of scenario type = “shoe order 1” = phase weight of “START” = “0.00”, scenario type = phase of “Shoe Order 1” = phase weight of “Customer Confirmation (Name)” = “0.00”, etc. Scenario Type = “Shoe Order 1” Phase = “START” phase score = 0.00 × 3 = “0.00”, scenario type = “shoes order 1” phase = “customer confirmation (name)” phase score = “0.00” Calculate.

  Next, the type determination unit 7 calculates a scenario score for each word based on the word score calculated in SC3 and the phase score calculated in SC5 (SC6). Specifically, the type determining unit 7 aggregates the word score calculated in SC3 for each type of scenario in which the word weight used when calculating the word score is set, and the phase calculated in SC5. The score is aggregated for each scenario type for which the phase weight used when calculating the phase score is set, and the aggregated result of these word scores and the aggregated result of the phase weight are further aggregated for each scenario type. To do. For example, in SC3, the word score of “shoe order 1” is calculated as the word score of the word = “size” = “3.00”, and the scenario is set as the phase score of the word = “size”. Type = “Shoe Order 1” Phase = “Product Confirmation (Size)” Phase Score = “3.00”, Word = “Size” Scenario Type = “Shoe Order 1” Word Another scenario score = 3.00 + 3.00 = 6.00 is calculated. Actually, since there can be many combinations according to the number of words, the number of types of scenarios, and the number of phases, the scenario score for each word is calculated for an enormous number of combinations.

  Next, the type determining unit 7 extracts only words having high importance from the words (SC7). For example, the type determining unit 7 extracts, from the words extracted by the word extraction process, a word whose appearance count calculated in SC1 is equal to or greater than a predetermined threshold as a highly important word.

  And the kind determination part 7 produces a score total table for the word extracted by SC7 (SC8). Specifically, for the word extracted in SC7, the type determining unit 7 uses the number of appearances calculated in SC1, the word score calculated in SC3, the phase score calculated in SC5, and the scenario score for each word calculated in SC6. A score summary table is created (SC8). FIG. 10 is a diagram illustrating a score summary table. This score summary table includes information corresponding to the item “word”, information corresponding to the item “number of occurrences”, information corresponding to the item “word score”, information corresponding to the item “phase score”, and item “scenario type” ", Information corresponding to the item" phase ", and information corresponding to the item" scenario score by word "are associated with each other. The information corresponding to the item “word” is a word extracted in SB2 of FIG. 7 and remains without being excluded as an unnecessary word in SB3 of FIG. The information corresponding to the item “appearance count” is the appearance count calculated in SC1. The information corresponding to the item “word score” is the word score calculated in SC3. The information corresponding to the item “phase score” is the phase score calculated in SC5. The information corresponding to the item “scenario type” is the type of scenario in which the phase weight used when the phase score is calculated in SC5 is set. The information corresponding to the item “phase” is a phase in which the phase weight used when the phase score is calculated in SC5 is set. The information corresponding to the item “scenario score by word” is the scenario score by word calculated in SC6. The score summary table created in this way is displayed and output via the output unit 3, for example, and is referred to by the user.

  Thereafter, the type determining unit 7 calculates a scenario score based on the score tabulation table created in SC8 (SC9). Specifically, the type determination unit 7 totals the word-specific scenario scores (word-specific scenario scores calculated in CS6) included in the score totaling table for each scenario type. For example, the scenario score for each word calculated in CS6 includes the scenario type for word = “size” = the scenario score for each word for “shoes order 1” = “6.00”, and the scenario type for word = “order”. = Scenario score by word of “Shoe Order 1” = “5.00” Scenario Type = Scenario Score of “Shoe Order 1” = 6.00 + 5.00 = 11.00 is calculated.

  Then, the type determining unit 7 creates a scenario score summary table based on the scenario score calculated in SC9 (SC10). FIG. 11 is a diagram illustrating a scenario score tabulation table. This scenario score totaling table is configured by associating information corresponding to the item “scenario type” and information corresponding to the item “scenario score” with each other. The information corresponding to the item “scenario type” is the type of scenario in which the phase weight used when the phase score is calculated in SC5 is set. Information corresponding to the item “scenario score” is the scenario score calculated in SC9. Each record of this scenario score correspondence count table is arranged in descending order of scenario score. For example, the scenario score = 352.50 of the scenario type = “Shoe Order-1” is higher than the scenario scores of the other scenario types, and is displayed at the top position. The scenario score summary table created in this way is displayed and output via the output unit 3, for example, and is referred to by the user.

  Then, the type determining unit 7 determines the type of scenario that matches the target conversation sentence based on the scenario score tabulation table created in SC10 (SC11). Specifically, the type of scenario having the highest scenario score (the type of scenario displayed at the top position in the scenario score correspondence count table) is determined as the type of scenario that matches the target conversation sentence. For example, in the example of FIG. 11, the scenario type “shoes order 1” matching the target conversation sentence is determined. When there are a plurality of type specifying information corresponding to the highest scenario score, one type specifying information is specified by an arbitrary method from the plurality of type specifying information. This ends the type determination process.

(Processing-scenario creation processing)
Next, the scenario creation process (SA3) of FIG. 6 is executed. FIG. 12 is a flowchart of the scenario creation process. In this process, the scenario creation unit 8 acquires scenario information corresponding to the type specifying information for specifying the type of scenario determined in the type determining process from the scenario information DB 11 (SD1). For example, when the scenario type determined in the type determination process = “Shoes order-2”, the scenario information including the scenario type = “Shoes order-2” is copied from the scenario information DB 11, and The copied scenario information is used as the base of the scenario corresponding to the target conversation sentence.

  Thereafter, the first scenario complementing unit 9 is scenario information corresponding to a word having high importance, and is not included in the scenario information acquired in SD1, but the scenario information included in the scenario information DB 11 is included in the scenario information. It is determined whether or not there is. If there is such scenario information, the scenario information acquired in SD1 is supplemented based on the scenario information, assuming that the scenario information is also scenario information to be included in the scenario corresponding to the target conversation sentence. (SD2). For example, when the word “color” is extracted in SC7 as a highly important word and included in the score summary table of FIG. 10, scenario information corresponding to the word “color” is acquired in SD1. If the scenario information is not included in the scenario information, the scenario information corresponding to the word “color” is other scenario information in the scenario information DB 11 (other than the scenario information corresponding to the scenario type determined in the type determination process). (Scenario information) is determined. If the scenario information is included, the scenario information is acquired from the scenario information DB 11 and added to the scenario information acquired in SD1, thereby supplementing the scenario information acquired in SD1. This additional position is, for example, at the end of the same phase as the phase of the scenario information to be added. For example, when the phase of the scenario information corresponding to the word “color” acquired from the scenario information DB 11 = “product confirmation (color)”, the phase = “product” in each record of the scenario information acquired in SD1. The record corresponding to “confirmation (variable)” is specified, and the scenario information record acquired from the scenario information DB 11 is added to the end of the specified record. If the phase is set only with the phase name, it is added to the end of the record having the same phase name.

  FIG. 13 shows an example of scenario information supplemented in this way. In this example, the scenario type determined in the type determination process is “shoe order 1”, and the scenario information shown in the example of FIG. 2 is acquired in SD1. In the scenario information of FIG. 13, records L1 and L2 are complemented to the scenario information of FIG. Since the record L1 is a scenario information record of scenario type = “apparel order 1” and is a record of phase = “product confirmation (color)”, the last of the record of phase = “product confirmation” (phase = It is added to the record of “product confirmation (product name)” and the phase = immediately after the record of “product confirmation (size)”. Further, the record L2 is a scenario information record of scenario type = “Sweets Order 1” and is a record of phase = “customer question”, so it is added to the end of the record of phase = “customer question”. Yes.

  Next, the second scenario complementing unit 10 is scenario information corresponding to a word having high importance, and is not included in the scenario information acquired in SD1, but is also included in the scenario information DB 11. It is determined whether or not there is. If there is such scenario information, it is acquired in SD1 based on the scenario information extracted from the target conversation sentence, assuming that the scenario information is also scenario information to be included in the scenario corresponding to the target conversation sentence. Scenario information is supplemented (SD3). For example, in the case where the word “payment” is extracted as a highly important word in SC7 and included in the score summary table of FIG. 10, the scenario information corresponding to the word “payment” is the scenario information acquired in SD1. If it is not included in the scenario information DB 11, it is determined whether or not the scenario information corresponding to the word “payment” is included in the scenario information DB 11.

  If the scenario information is not included, the scenario information acquired in SD1 is corrected by extracting the scenario information from the target conversation sentence and adding it to the scenario information acquired in SD1. Thus, scenario information that is actually required is predicted. This additional position is, for example, the last position. However, when a conversation sentence in a phase to be performed before the last tail is specified, the last tail excluding the conversation sentence is set as an additional position. As the conversation sentence in the phase to be performed before the last tail in this way, for example, there is a conversation sentence in the “customer question” phase. At this time, the scenario type of the scenario information to be added is the scenario type determined in the type determination process. The phase of the scenario information to be added is set by adding “confirmation” to the word determined to be a word having high importance when determining whether or not the scenario information needs to be added. For example, when the word is “payment”, the phase of the scenario information to be added = “payment confirmation” is set. In addition, the variable of the scenario information to be added is set to the word determined to be a word having high importance when determining whether or not the scenario information needs to be added. For example, if the word is “payment”, the variable = “payment” is set.

  At this time, an identification symbol (in this case, []) is added to the newly set phase and variable so that the newly set information (predicted information) can be easily identified. . Also, a set stage identifier is assigned to the variable. The set stage identifier is information for determining which stage of the conversation should be set when the scenario execution apparatus sets a variable when the scenario execution apparatus automatically executes a scenario. In the embodiment, it is “p” or “d”. The set stage identifier = “p” indicates that the customer's response is set directly. For example, when the variable = “order history”, in the conversation “OP: first order?” “CU: yes”, the variable = “yes” is set directly by the customer's response. Stage identifier = “p” is given. The set stage identifier = “d” indicates that the response obtained interactively is set. For example, if variable = “payment method”, “OP: how will the payment method be handled?” “CU: what kind of payment method is available” “OP: cash on delivery payment, credit card settlement, convenience store payment "CU: How much is the fee for cash on delivery payment?" "OP: 315 yen." "CU: Well, convenience store payment is OK." . ”“ CU: Yes ”, the variable =“ convenience store payment ”is set by the interactive response, so the set stage identifier =“ d ”is given. However, when such a set stage can be ignored, the set stage identifier may be omitted, or all “p” may be added.

  In the scenario information of FIG. 13, the record L3 is supplemented with respect to the scenario information of FIG. This record L3 is added to the end of the scenario information excluding “customer question”, “shoe order 1” is set for the scenario type, “payment confirmation” is set for the phase, and the variable is "Payment" is set. More specifically, the phase = “payment confirmation” and the variable = “payment” are added with an identification symbol, and the phase = “[payment confirmation]” and the variable = “[payment]” are set.

  Finally, the scenario creation unit 8 stores the scenario information corrected in SD3 in the scenario information DB 11 (SD4). At this time, a scenario name is given to the scenario information corrected in SD3 and stored. Generally, when trying to automatically generate a scenario, the person in charge (the user of the support device) knows what business conversation the target conversation sentence is, so the first step of the automatic dialogue scenario creation process In this stage, the person in charge inputs a scenario name via the input unit, and the input scenario name is assigned to the scenario information. Since the target conversation sentence in FIG. 8 is a conversation sentence related to the business of “apparel (pajamas) order”, scenario name = “apparel (pajamas) order” is assigned to the scenario information in FIG. However, this scenario name may be automatically acquired by, for example, syntactic analysis of scenario information. At this time, the item “scenario type” in the scenario information corrected in SD3 is uniformly updated with the set scenario name. The scenario information after the update is illustrated in FIG. This completes the scenario creation process. Similarly, each time a new target conversation sentence is specified, scenario information is generated and stored in the scenario information DB 11, so that the scenario information in the scenario information DB 11 is automatically learned. Moreover, the scenario information created in this way is displayed and output via the output unit 3, for example, and is referred to by the user.

  Thereafter, in the interactive voice automatic response system, the scenario information is called from the scenario information DB 11 and used for the automatic voice response by a known technique.

[Modification]
Although the embodiments of the present invention have been described above, the specific configuration and means of the present invention can be arbitrarily modified and improved within the scope of the technical idea of each invention described in the claims. Can do. Hereinafter, such a modification will be described.

(About problems to be solved and effects of the invention)
First, the problems to be solved by the invention and the effects of the invention are not limited to the above-described contents, and the present invention solves the problems not described above or has the effects not described above. There are also cases where only some of the described problems are solved or only some of the described effects are achieved. For example, even if the operation cost of an automatic dialog system and the time required to create a new type of scenario are similar to the conventional case, if the scenario can be created by a method different from the conventional case, The problems of the invention have been solved.

(About distribution and integration)
Further, each of the electrical components described above is functionally conceptual and does not necessarily need to be physically configured as illustrated. That is, the specific form of distribution or integration of each part is not limited to the one shown in the figure, and all or a part thereof may be functionally or physically distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured. For example, the function of the support device 1 can be distributed to a plurality of servers. Further, the database stored in the storage unit 5 of the support apparatus 1 can be arbitrarily distributed or integrated.

(About score summary table and scenario score summary table)
In the above embodiment, the case of creating the score summary table and the scenario score summary table has been described. However, when the user does not need to use these summary tables themselves, the creation of these summary tables is omitted, Similarly, the scenario score may be calculated or the scenario type may be determined based on the scenario score.

(About the importance of words)
In the above embodiment, the degree of importance of the word is determined based on the number of appearances of the word, but it may be determined based on the word weight or the phase weight, or may be determined by combining these criteria. Also good.

(About supplementation and prediction)
Either or both of the first scenario complementing unit 9 and the scenario prediction may be omitted, or the scenario may be supplemented or predicted by another method.

DESCRIPTION OF SYMBOLS 1 Automatic dialog scenario creation assistance apparatus 2 Input part 3 Output part 4 Control part 5 Memory | storage part 6 Word extraction part 7 Type determination part 8 Scenario creation part 9 1st scenario complementing part 10 2nd scenario complementing part 11 Scenario information DB
12 Unnecessary word information DB
13 Word weight information DB
14 Phase weight information DB

Claims (4)

An automatic dialogue scenario creation support device that supports creation of a scenario for an automatic dialogue system,
Scenario information related to a plurality of scenarios, the type specifying information specifying each type of the plurality of scenarios, the plurality of conversation sentences constituting each of the plurality of scenarios, and the conversation information in each of the plurality of scenarios Scenario information storage means for storing scenario information configured to correlate with phase specifying information for specifying a phase indicating a stage;
A word included in the plurality of conversation sentences; the type specifying information; and a word weight indicating a weight of the word when the word is used in a conversation sentence of a type specified by the type specifying information; A word weight information storage means for storing word weight information configured in association with each other,
The conversation of the phase specified by the phase specifying information in the scenario of the type in which the word included in the plurality of conversation sentences, the type specifying information, the phase specifying information, and the word is specified by the type specifying information Phase weight information storage means for storing phase weight information configured to correlate phase weight indicating the weight of the word when used in a sentence;
A word extraction means for extracting a word from the specified target conversation sentence when a target conversation sentence that is a plurality of conversation sentences to be created as a scenario is specified by a predetermined method;
The number of appearances of the word extracted by the word extraction unit is counted, a word weight corresponding to the word, and a word weight for each type specifying information is acquired from the word weight information storage unit, and the acquired word A word score for each type specifying information is calculated by multiplying a weight by the number of appearances, and a phase weight corresponding to the word extracted by the word extracting unit, for each type specifying information and for each phase specifying information. A phase score for each type specifying information and each phase specifying information is calculated by multiplying the acquired phase weight by the number of appearances, and calculating the word score And the calculated phase score for each type specifying information Was calculated Riosukoa, and type determination means for determining the type of scenario that matches the target sentence based on the calculated scenarios score,
Scenario information corresponding to the type specifying information for specifying the type of scenario determined by the type determining unit is acquired from the scenario information storage unit, and the scenario corresponding to the target conversation sentence is acquired based on the acquired scenario information Scenario creation means for creating
An automatic dialogue scenario creation support device comprising: The importance of the word extracted by the word extracting means is determined by a predetermined method, and the scenario information corresponding to the word determined to be important is acquired from the scenario information storing means by the scenario creating means If it is not included in the scenario information and can be acquired from the scenario information storage means, the scenario information is acquired from the scenario information storage means, and based on the acquired scenario information, the scenario information Comprising a first scenario complementing means for complementing scenario information acquired from the scenario information storage means by a scenario creating means;
The automatic dialogue scenario creation support apparatus according to claim 1. The importance of the word extracted by the word extracting means is determined by a predetermined method, and the scenario information corresponding to the word determined to be important is acquired from the scenario information storing means by the scenario creating means If it is not included in the scenario information and cannot be acquired from the scenario information storage means, the scenario information is acquired from the target conversation sentence, and the scenario creation means is based on the acquired scenario information. A second scenario complementing means for complementing the scenario information acquired from the scenario information storage means by
The automatic dialogue scenario creation support apparatus according to claim 1 or 2. An automatic dialogue scenario creation support program for supporting creation of a scenario for an automatic dialogue system,
Scenario information related to a plurality of scenarios, the type specifying information specifying each type of the plurality of scenarios, the plurality of conversation sentences constituting each of the plurality of scenarios, and the conversation information in each of the plurality of scenarios Scenario information storage means for storing in the scenario information storage means scenario information configured to correlate the phase specifying information for specifying the phase indicating the phase;
A word included in the plurality of conversation sentences; the type specifying information; and a word weight indicating a weight of the word when the word is used in a conversation sentence of a type specified by the type specifying information; A word weight information storage means for storing word weight information configured in association with each other,
The conversation of the phase specified by the phase specifying information in the scenario of the type in which the word included in the plurality of conversation sentences, the type specifying information, the phase specifying information, and the word is specified by the type specifying information Phase weight information storage means for storing phase weight information configured to correlate phase weight indicating the weight of the word when used in a sentence;
A computer comprising
A word extraction means for extracting a word from the specified target conversation sentence when a target conversation sentence that is a plurality of conversation sentences to be created as a scenario is specified by a predetermined method;
The number of appearances of the word extracted by the word extraction unit is counted, a word weight corresponding to the word, and a word weight for each type specifying information is acquired from the word weight information storage unit, and the acquired word A word score for each type specifying information is calculated by multiplying a weight by the number of appearances, and a phase weight corresponding to the word extracted by the word extracting unit, for each type specifying information and for each phase specifying information. A phase score for each type specifying information and each phase specifying information is calculated by multiplying the acquired phase weight by the number of appearances, and calculating the word score And the calculated phase score for each type specifying information Was calculated Riosukoa, and type determination means for determining the type of scenario that matches the target sentence based on the calculated scenarios score,
Scenario information corresponding to the type specifying information for specifying the type of scenario determined by the type determining unit is acquired from the scenario information storage unit, and the scenario corresponding to the target conversation sentence is acquired based on the acquired scenario information Scenario creation means for creating
Act as a
Automatic dialogue scenario creation support program.

Images