JP2018197924A - Information processing apparatus, interactive processing method, and interactive processing program - Google Patents

Abstract

To allow interaction without preparing text data of a large paraphrase.SOLUTION: An information processing apparatus interacts with a user by outputting an answer to a question from the user. The information processing apparatus comprises: an input unit that accepts the question; and an interaction control unit that selects a candidate for the answer to the question, accepted by the input unit, out of candidate data, performs confirmation processing for confirming whether a paraphrase of the question is appropriate or not to the user, and registers the candidate as the paraphrase of the question on the basis of a result of the propriety in an interaction.SELECTED DRAWING: Figure 3

Description

  The present invention relates to an information processing apparatus, a dialogue processing method, and a dialogue processing program.

  In recent years, the digital divide has expanded due to the emergence of advanced services using information technology (IT), and communication has become more complex due to globalization and lifestyle diversification. For this reason, it is difficult for existing staff alone to provide various interpersonal services such as facility guidance at airports and stations that require support in various languages, and bank and municipal offices that handle multiple complex services. It is coming. In order to support such sophisticated service operations, practical use of interactive systems such as robots and voice assistants is being promoted.

  The dialogue system is required to cope with various utterances of the user. For example, when you want to know the opening hours of a certain store, there are various ways to ask, such as "Tell me about opening hours", "How long have you been doing?", "I want to know open hours." In order to deal with such various ways of speaking, it is necessary to prepare various paraphrases in advance. However, the cost of manually creating paraphrasing is high. In addition, because the way of speaking differs depending on the person, it is difficult to prepare various paraphrasing with a small number of people, and it is necessary to have many people create it.

  So far, various techniques for preparing various paraphrasing have been considered. One of them is described in Patent Document 1. This document shows a method for increasing the paraphrase vocabulary by strategically associating unknown words with registered words through dialogue. A method of automatically generating a paraphrase sentence using such a paraphrase vocabulary or a synonym dictionary has also been proposed.

JP 2007-213055 A

  The paraphrase sentence automatically generated as in the prior art may include inappropriate paraphrase. For example, if “partner” and “spouse” are defined as synonyms, the sentence “customer is an important partner” can be rephrased as an inappropriate sentence “customer is an important spouse.” It can be considered. Therefore, there is a need for a method for determining whether or not an automatically generated paraphrase text is appropriate. In order to determine whether or not the paraphrase text is appropriate, large-scale paraphrase text data (corpus) is required. Need to be prepared in advance. Therefore, in a field where a large-scale paraphrase corpus cannot be prepared, a corpus of another field can be used, but field-dependent paraphrase cannot be handled.

  An object of the present invention is to provide an information processing apparatus, a dialogue processing method, and a dialogue processing program capable of dialogue without preparing large-scale paraphrase text data.

  An information processing apparatus according to the present invention is an information processing apparatus that outputs an answer to a question from a user and interacts with the user, the input unit accepting the question, and the input unit accepted in the dialogue A candidate for answering a question is selected from candidate data, a confirmation process for confirming whether or not the question is paraphrased is confirmed with the user, and the candidate is registered as a paraphrase for the question based on the result of the suitability An information processing apparatus including a dialogue control unit.

  The present invention is also grasped as a dialogue processing method and dialogue processing program performed in the information processing apparatus.

  According to the present invention, it is possible to interact without preparing large-scale paraphrase text data.

The hardware block diagram of a dialogue system. The flowchart which shows the flow of a process of the dialogue system including a paraphrase confirmation process. The flowchart which shows the flow of a dialog control process. The conceptual diagram which shows a confirmation process. The flowchart which shows the flow of a confirmation process. Dialogue data before and after paraphrase confirmation and user utterance paraphrase candidate data. Dialog data before and after update and user utterance paraphrase candidate data. The conceptual diagram which shows an answer process. The conceptual diagram which shows a hearing process. The processing flow figure which selects the data to confirm from several candidate data. The figure which shows the example of paraphrase confirmation result data. The figure which shows the example in case a confirmation frequency optimization module adjusts a frequency automatically in consideration of a user's discomfort degree.

Embodiments will be described in detail with reference to the drawings. However, the present invention is not construed as being limited to the description of the embodiments below.
Moreover, a component expressed in the singular form in this specification includes a plural form unless the context clearly indicates otherwise.

  As will be described below, in this embodiment, the appropriateness of paraphrase is determined by a method of making a confirmation question for clarifying whether the automatically generated paraphrase is appropriate or inappropriate. Examples include a paraphrase confirmation question technique and a dialogue system for asking a paraphrase confirmation question. In the embodiment described below, even if a large-scale paraphrase corpus cannot be prepared in advance, a paraphrase confirmation question is asked.

<1. Overall system configuration of embodiment>
FIG. 1 is a block diagram showing the configuration of the dialogue system in the present embodiment.
The dialogue system 100 is configured by an information processing apparatus such as a computer as a specific example. Similar to a normal information processing apparatus, the interactive system 100 includes a central processing unit (CPU) 101, an input / output device 103 such as a keyboard and an image monitor, and a memory 107 including a magnetic disk device and a semiconductor memory device. Further, the data communication unit 102 may be provided as an interface for exchanging data with the outside. The data communication unit 102 is connected to an external network 110, for example. Note that the term “input / output device” does not mean only a device having both input / output functions, but also means a device having only an input function, a device having only an output function, and a device having both input / output functions. It shall be.

  In this embodiment, functions such as calculation and control are realized by a program stored in the memory 107 being executed by the CPU 101 in cooperation with other hardware. A program executed by the CPU 101, its function, or means for realizing the function may be referred to as “function”, “means”, “unit”, “unit”, “module”, or the like.

  In FIG. 1, functions that the CPU 101 executes conceptually based on software are illustrated as an intention analysis unit 104, a dialogue control unit 105, a maintenance unit 106, an input processing unit 108, and an output processing unit 109. A program for realizing these functions is stored in the memory 107. The memory 107 stores dialogue data 1071, candidate data (user utterance paraphrase candidate data) 1072, paraphrase confirmation result data 1073, dialogue log data 1074, and the like as data.

  The above-mentioned program is provided by being incorporated in advance in a ROM or the like, or can be read by a computer such as a CD-ROM, CD-R, DVD (Digital Versatile Disk) in an installable or executable format file. It may be provided by being recorded on a recording medium or distributed. Furthermore, the program may be provided or distributed by storing it on a computer connected to a network such as the Internet and downloading it via the network.

  Dialog data 1071 and dialog log data 1074 stored in the memory 107 can be input via the data communication unit 102 or the input / output device 103. The candidate data 1072 and the paraphrase confirmation result data 1073 can be output via the data communication unit 102 or the input / output device 103. In this example, the explanation is based on the assumption that the candidate data 1072 and the paraphrase confirmation result data 1073 are output from the interactive system 100, but for example, an output device such as a display, a screen, or a speaker provided outside the interactive system 100 The same applies to the case where the user who has output these data and confirms the contents inputs further questions to the dialog system 100.

  The intention analysis unit 104 is a processing unit that analyzes the intention of a question sentence input from a user. The dialogue control unit 105 is a processing unit that controls dialogue between a question sentence input from the user and an answer sentence or confirmation sentence from the dialogue system 100.

  The maintenance unit 106 inputs and outputs dialogue data 1071, candidate data 1072, paraphrase confirmation result data 1073, and dialogue log data 1074 so that the system administrator manages data stored in the memory 107.

  The input processing unit 108 is a processing unit that performs input processing necessary for the present system, such as converting a question sentence input by the user into text. The output processing unit 109 is a processing unit that performs an output process necessary for this system, such as outputting an answer sentence or a confirmation sentence to the question sentence.

  The above configuration may be configured as a single computer as shown in FIG. 1, or another computer in which any part of the input device, output device, processing device, and storage device is connected via a network. It may be constituted by. In the present embodiment, functions equivalent to those configured by software can also be realized by hardware such as FPGA (Field Programmable Gate Array) and ASIC (Application Specific Integrated Circuit).

<2. Explanation of dialogue system processing>
Processing of the dialogue system 100 will be described.

<2-1. Outline of Dialog System Processing>
The dialogue control unit 105 uses the dialogue data 1071, the candidate data 1072, the paraphrase confirmation result data 1073, and the dialogue log data 1074 to perform processing for generating an answer sentence and a confirmation sentence for the question sentence in the dialogue with the user.

FIG. 2 shows a process flow of the interactive system 100 including the paraphrase confirmation process.
In step S201, the input processing unit 108 performs input processing for converting a question sentence input by the user into text. For example, when a question sentence from the user is input by voice, the voice is converted into text and normalized. In normalization, you may unify notation such as unification of uppercase and lowercase letters and numerical values. The notation may also be unified when a question text from the user is input as text.

  In step S202, the intention analysis unit 104 performs intention analysis of the user input normalized in step S201. For example, the intention analysis unit 104 finds a question sentence close to the question sentence input by the user from the pair of question sentences and answer sentences registered in the dialogue data 1071, and the question sentence input from the user by the question sentence The intention of the sentence. Moreover, you may calculate the similarity between the question sentence input from the user, and the registered question sentence as a certainty factor which shows the probability which represents the intention of the question sentence input from the user. For example, when the intention analysis unit 104 matches the character string constituting the question sentence input from the user with the character string constituting the question sentence registered in the dialogue data 1071, the intention analysis unit 104 The certainty factor is calculated as 80%.

  In step S203, the dialogue control unit 105 performs dialogue based on the question sentence and dialogue log data 1074 (confirmation status of paraphrasing of the user utterance, the number of continuous hearings, the number of dialogues, etc.) obtained as a result of step S202 An answer sentence to be output is generated from the answer sentences registered in the data 1071. The confirmation status of paraphrasing user utterances means that it means “tell me open hours?” In response to the user's “tell me business hours?” Question? Is a confirmation sentence for rephrasing the question and confirming its intention. In addition, when the speech recognition accuracy is low for some reason or when a question sentence close to the question sentence input by the user is not registered in the dialogue data 1071, a reply sentence such as “Please say again” is used. It is to output and answer. Furthermore, the continuous number of replays is the number of times of continuous replays.

  In process S204, the output processing unit 109 performs a process necessary for outputting the answer sentence generated in process S203. For example, when outputting the answer sentence by voice, the output processing unit 109 performs processing for converting the text of the answer sentence to voice by an arbitrary voice synthesis technique. If the answer sentence is to be output as text, processing normalized in step S201 may be returned to the original notation (notation used by the user). When the next dialogue continues, the confirmation status of the above paraphrasing, the ID for identifying the question being confirmed, the number of consecutive rehearsals, and the number of dialogues may be stored in the dialogue log data 1074 and handed over.

<2-2. Dialogue control processing>
FIG. 3 is a flowchart showing details of the dialogue control process S203.
In FIG. 3, the dialogue control unit 105 has three types of information based on the question text that is the result obtained by the processing S202 and the three pieces of information obtained from the dialogue log 1074 (the above-described similarity, confirmation status, and number of dialogues). An example is shown in which one is selected from the processes (confirmation process, answer process, and answer process).

  In process S301, the dialogue control unit 105 performs a branch process based on the certainty factor. Specifically, the dialogue control unit 105 determines whether or not the certainty level determined by the intention analysis unit 104 in step S202 is greater than or equal to a predetermined threshold (for example, 80%), and the certainty level is the threshold value. When it determines with it being above (S301; Yes), it progresses to process S302. When the dialogue control unit 105 determines that the certainty factor is less than the threshold (S301; No), the dialogue control unit 105 proceeds to step S307. The threshold may be set arbitrarily by the system administrator, or may be automatically set by any method.

  In process S302, the dialogue control unit 105 performs a branch process depending on the confirmation status. Specifically, the dialogue control unit 105 determines whether or not the confirmation process in S305 is being executed, that is, whether or not the confirmation sentence is being output, and determines that the confirmation process is being executed. (S302; Yes), it progresses to process S305. When it is determined that the confirmation process is not executed (S302; No), the dialogue control unit 105 proceeds to the process S303.

  In the process S303, the dialogue control unit 105 performs a branch process depending on the presence or absence of candidate data 1072 that needs to be confirmed. Specifically, the dialogue control unit 105 determines whether or not the candidate data 1072 has a question text paraphrase candidate for which the certainty factor is determined to be greater than or equal to the threshold value from the dialogue data 1071. The determination of the question sentence candidate is, for example, determining whether there is a candidate ID having the same value as the conversation ID stored in the conversation data 1071 shown in FIG. 4 and determining that there is a candidate ID having the same value. It is determined that the paraphrase corresponding to the candidate ID is registered as candidate data.

  When it is determined that there is candidate data (S303; Yes), the dialogue control unit 105 proceeds to processing S304, and when it is determined that there is no candidate data (S303; No), the processing proceeds to processing S306.

  In process S304, the dialogue control unit 105 performs branch processing according to the number of dialogues. Specifically, the dialogue control unit 105 determines whether or not the number of dialogues with a certain user is a dialogue with a predetermined threshold number (for example, N = 3). If the dialogue control unit 105 determines that the number of dialogues is the above threshold number (S304; Yes), the process proceeds to step S305. Otherwise (S304; No), the process proceeds to step S306. The threshold number N is an arbitrary number. For example, the system administrator may set an arbitrary value. In this example, the confirmation process of the next S305 is executed at a frequency of once for three dialogs. By providing such a threshold value, user discomfort due to frequent execution of confirmation processing can be suppressed.

  In process S305, the dialogue control unit 105 performs a confirmation process of selecting one candidate data that needs to be confirmed and confirming whether the candidate data is rephrased for the question sentence spoken by the user. The confirmation process will be described later.

  In process S306, the dialogue control unit 105 performs an answer process for answering a question sentence uttered by the user. Specifically, the dialogue control unit 105 reads an answer sentence corresponding to the question sentence for which the certainty factor is determined to be equal to or greater than the threshold value in S301 from the dialogue data 1071, and outputs the read answer sentence.

  In process S307, the dialogue control unit 105 performs a reply process for outputting a reply sentence such as “Please say again”. What is necessary is just to memorize | store the reply sentence in the memory 107 beforehand.

<2-3. Confirmation process in dialog control process>
FIG. 4 shows an example of the confirmation process S305 in the dialogue control process S203. In the confirmation process S305, dialogue data 1071 and candidate data 1072 are used. As shown in FIG. 4, the dialog data 1071 includes a dialog ID for identifying a dialog with the user, a question text (Q) in the dialog identified by the dialog ID, and a response text indicating an answer to the question text. (A) is stored in association with each other. For example, in the dialogue data 1071, an answer sentence “business hours are from 10:00 to 18:00” is registered in response to a question sentence “What is business hours?” Identified by the dialogue ID “1”. It shows that. It also indicates that a similar answer sentence is registered for the question sentence “When are you open?” Identified by the same dialogue ID. Since the dialogue ID is given for each intention (for each answer) of the question sentence, an arbitrary question sentence and its paraphrase are the same conversation ID.

  As shown in FIG. 4, the candidate data 1072 includes a candidate ID for identifying a question sentence candidate, a question sentence (Q) as a paraphrase candidate identified by the candidate ID, and a question from the user. An appropriate flag indicating the number of times that a sentence can be confirmed by paraphrase that the sentence is the intention of the question sentence, and an inappropriate flag indicating the number of times that the question sentence from the user cannot be confirmed by paraphrase that the sentence is the intention of the question sentence And a registration flag indicating whether or not the paraphrase candidate has already been registered as a question text of dialogue data. The appropriate flag is the number of times that the paraphrase is replied when confirmed to the user. Similarly, the inappropriate flag is the number of times that the paraphrase is answered as inappropriate. In this example, a question sentence with a candidate ID of “Tell me open time.” Is registered together with the confirmation result (number of times determined to be appropriate, number of times determined to be inappropriate) and registration status. ing. In addition to the appropriate flag value and inappropriate flag value, the confirmation result may include an unknown flag value indicating the number of times it has been determined that appropriateness cannot be determined.

  As shown in FIG. 4, when a question text “What is business hours?” Is input from the user, in process S303, the dialog control unit 105 first determines that the question text is a question text (Q ) Is registered. When the dialog control unit 105 determines that the question text is registered in the dialog data 1071, the dialog control unit 105 acquires a dialog ID corresponding to the question text.

  Next, the dialogue control unit 105 confirms whether a question sentence having a candidate ID having the same value as the acquired dialogue ID exists in the candidate data 1072. In process S305, the dialogue control unit 105 selects one question (Q) from the candidate data, and makes a question for confirming the appropriateness of paraphrasing the user utterance. How to select when there are a plurality of candidate data will be described later. In this example, the question “Tell me when you are open.” Is selected, and “Do you mean“ Tell me when you are open? ” ”Is output and the user is asked. On the other hand, the user replies “Yes” indicating that the paraphrase based on the confirmation sentence is appropriate, and the dialogue system 100 receives the reply in the confirmation process S305.

  Further, as shown in FIG. 4, a history of dialogues included in a sequence of dialogues from the beginning to the end is stored in the dialogue log data 1074. The dialogue log data 1074 is recorded by the dialogue control unit 105 every time one dialogue is performed between the user and the present system. In FIG. 4, the dialogue log data 1074 stores a log ID for identifying a question sentence, a confirmation sentence, or an answer sentence and the contents of the question sentence or the answer sentence identified by the log ID in association with each other. I understand that.

FIG. 5 is a flowchart showing details of the confirmation processing S305.
FIG. 5 shows an example of generating an answer sentence related to the confirmation process for confirming the appropriateness of paraphrasing of the user utterance.

  In process S501, the dialogue control unit 105 performs a branch process according to the paraphrase confirmation status. Specifically, the dialogue control unit 105 determines whether the paraphrase of the user utterance is being confirmed (for example, whether the confirmation process is being executed based on the confirmation sentence of the log ID 2 illustrated in FIG. 4). If it is determined that it is being confirmed (S501; Yes), the process proceeds to step S502. If it is determined that it is not being confirmed (S510; No), the process proceeds to step S512.

  In process S502, the dialogue control unit 105 performs a branch process according to the appropriateness of paraphrasing. Specifically, the dialogue control unit 105 determines whether or not the user's reply to the confirmation sentence output by the system in the confirmation process is a content indicating that the paraphrase such as “Yes” is correct. It is determined whether the paraphrase was appropriate. When the dialogue control unit 105 receives the reply indicating the correctness from the user and determines that the paraphrase is appropriate (S502; Yes), the process proceeds to step S503, and determines that it is inappropriate (S502; No). Then, the process proceeds to step S504.

In process S503, the dialogue control unit 105 adds 1 to the value of the appropriate flag of the candidate data 1072.
In process S504, the dialogue control unit 105 adds 1 to the value of the inappropriate flag of the candidate data 1072.

  FIG. 6 shows an example of the dialogue data 1071 and candidate data 1072 before and after the paraphrase confirmation in the processing S503 and the processing S504. (1) shows the state before the confirmation process, and (2) shows the state after the confirmation process. As in (2-1), when the dialog control unit 105 determines that the confirmation result is appropriate, 1 is added to the value of the appropriate flag in the candidate data 1072. On the other hand, as shown in (2-2), when the dialogue control unit 105 determines that the confirmation result is inappropriate, 1 is added to the value of the inappropriate flag in the candidate data 1072. In FIG. 6, it is determined that the paraphrase “when is the open time” is determined to be appropriate, while the paraphrase “when is the break time” is determined to be inappropriate.

  In the processes S505 to S511, when a plurality of candidate data 1072 are registered for one candidate ID, in each data (in this example, “when is the open time?” And “when is the break time?”) It repeats with respect to it, and it progresses to process S306 after that.

  In step S506, the dialogue control unit 105 performs branch processing based on the appropriate flag value and the inappropriate flag value. Specifically, the dialogue control unit 105 determines whether or not the relationship between the appropriate flag value and the inappropriate flag value satisfies a predetermined condition. For example, when the dialogue control unit 105 determines that each of the data stored in the candidate data 1072 is larger than the appropriate flag value by a certain number greater than the threshold value, the data Is determined to be inappropriate as candidate data 1072, but if it is determined that the value of the inappropriate flag is not greater than a certain number of threshold values than the appropriate flag value, the data is appropriate as candidate data 1072 Judge.

  When the dialogue control unit 105 determines that each piece of data of the candidate data 1072 is appropriate as the candidate data 1072 (S506; Yes), the process proceeds to the process S507. If the dialogue control unit 105 determines that each piece of candidate data 1072 is inappropriate as candidate data 1072 (S506; No), the process proceeds to step S508. In addition to the above, any method can be used as a method for determining whether or not it is appropriate.

  For example, instead of the relative relationship between the value of the appropriate flag and the value of the inappropriate flag as described above, the value of the appropriate flag is equal to or greater than a certain threshold A, and the value of the inappropriate flag is equal to or less than another threshold B. In such a case, it is also possible to use predetermined values for the appropriate flag value and the inappropriate flag value, such as determining that the paraphrase is appropriate and determining that the other is inappropriate. Further, it may be determined that an appropriate value / (appropriate value + inappropriate value) is not less than a certain threshold value C, and that other than that is inappropriate. In this case, the threshold value A and the threshold value B may be determined by the system administrator, or may be automatically adjusted based on a paraphrase confirmation log.

  In the process S507, the dialog control unit 105 performs a branch process depending on whether or not the dialog data 1071 includes the data of the candidate data 1072 for which the relationship is determined. Specifically, the dialog control unit 105 determines whether the data of the candidate data 1072 is included in the dialog data 1071, and determines that the data of the candidate data 1072 is included in the dialog data 1071. In the case (S507; Yes), the process proceeds to step S511. On the other hand, when the dialogue control unit 105 determines that the data of the candidate data 1072 is not included in the dialogue data 1071 (S507; No), the dialogue control unit 105 proceeds to processing S509.

  In step S508, similarly to step S507, branch processing is performed depending on whether or not the dialogue data 1071 includes data of candidate data 1072 for which the above relationship has been determined. If the dialog control unit 105 determines that the data of the candidate data 1072 whose relationship has been determined is included in the dialog data 1071 (S508; Yes), the process proceeds to step S510, and the dialog data 1071 determines the relationship as described above. If it is determined that the candidate data 1072 data is not included (S508; No), the process proceeds to step S511.

  In step S509, the dialogue control unit 105 adds data of candidate data 1072 for which the above relationship has been determined (data for which paraphrase is determined to be appropriate) to the dialogue data 1071.

  In process S510, the dialogue control unit 105 deletes the data of the candidate data 1072 for which the above relationship has been determined (data that has been determined to be inappropriate) from the dialogue data 1071.

  FIG. 7 shows an example of dialogue data 1071 and candidate data 1072 before and after the update of steps S505 to S511. (1) shows the state before update, and (2) shows the state after update. When the dialogue control unit 105 determines that the paraphrase is appropriate as in (2-1), the value of the registration flag of the candidate data 1072 becomes “completed” and is added to the dialogue data 1071. On the other hand, as shown in (2-2), when the dialogue control unit 105 determines that the paraphrase is inappropriate, the value of the registration flag of the candidate data 1072 becomes “not yet” and is not added to the dialogue data 1071.

  In FIG. 7 (2-1), since it was determined that the paraphrase “when is open?” Is appropriate in all ten dialogs, “completed” is set in the registration flag, and “open” This indicates that the paraphrase “when is the time?” Is registered in the dialogue data 1071. On the other hand, since the paraphrase “when is the break time” is determined to be inappropriate 9 times out of 10 dialogues, “not yet” is set in the registration flag, and “when is the open time?” If the paraphrase is not registered in the dialog data 1071 and is already registered, it can be understood that the paraphrase is not registered in the dialog data 1071.

  The maintenance unit 106 uses the screen data including the dialogue data 1071 and the candidate data 1072 shown in FIGS. 6 and 7 as paraphrase confirmation result data 1073 for the system administrator to manage the data stored in the memory 107. An interface for outputting may be provided. For example, the input / output device 103 of the information processing device that implements the interactive system 100 is provided on the back side of the information processing device, and the paraphrase confirmation result data 1073 operated by the administrator is output to the device as an administrator screen. May be.

  FIG. 11 is a diagram illustrating an example of the paraphrase confirmation result data 1073. As shown in FIG. 11, the paraphrase confirmation result data 1073 is registered with a result ID, a question sentence (Q), an answer sentence (A) similar to the dialog data 1071, and an appropriate flag and inappropriate flag similar to the candidate data 1072. A flag and a manual registration flag indicating whether or not to manually register the determined confirmation sentence in the question sentence are stored in association with each other. In FIG. 11, for example, since the system result registration flags of confirmation sentences “when is open time?” And “where can I get out?” Are “completed”, they are appropriate as paraphrases for each question. This indicates that the candidate data 1072 is registered in the dialogue data 1071.

  On the other hand, since the system result registration flag of the confirmation sentence “What is the break time?” Is “not yet”, it is determined that the question is paraphrased and is not registered in the dialogue data 1071. In addition, the manual registration flag is checked for the confirmation sentence “When is the open time?”. This indicates that the administrator instructs the dialog data 1071 to register. On the other hand, the manual registration flag is not checked for the confirmation sentences “What is the break time?” And “Where can I get out?”. This indicates that the administrator has not instructed the dialog data 1071 to register. Based on the manual registration flag input from these managers, the maintenance unit 106 updates the dialog data 1071.

  That is, in the dialogue system 100, as described above, when it is determined that the paraphrase is appropriate, it is possible to automatically register the candidate data 1072 in the dialogue data 1071, but the paraphrase confirmation result is the paraphrase confirmation. The result data 1073 may be output to the screen, and the administrator may visually confirm the result, select the desired candidate data 1072, and select candidate data 1072 to be registered in the dialogue data 1071. Even if the paraphrase is determined to be appropriate for the user, there are cases where various paraphrasing expressions are registered in the dialogue data 1071 because the tolerance of the paraphrase is different for each user. Therefore, after the administrator confirms the paraphrase confirmation result data 1073 as described above and determines whether or not to register, the candidate data 1072 is registered in the conversation data 1071, whereby the amount of data held in the conversation data 1071 Can be suppressed.

  In the process S512, the dialog control unit 105 performs a branch process based on the user utterance, the dialog data 1701, and the candidate data 1702. Specifically, the dialogue control unit 105 determines whether or not the paraphrase corresponding to the user utterance has a registration flag not registered in the candidate data 1702, that is, whether or not there is data stored as unregistered in the dialogue data 1071. judge. If the dialog control unit 105 determines that the candidate is stored as unregistered in the dialog data 1701 (S512; Yes), the process proceeds to step S513, and the candidate is not stored as unregistered in the dialog data 1701. When it determines (S512; No), it progresses to reply process S306.

  In step S513, the dialogue control unit 105 performs a branch process according to the number of paraphrase candidates. Specifically, the dialogue control unit 105 determines whether or not there are a plurality of candidates stored as unregistered in the dialogue data 1701, and there are a plurality of candidates stored as unregistered in the dialogue data 1701. (S513; Yes), the process proceeds to step S514. If it is determined that there are not a plurality of candidates stored as unregistered in the dialogue data 1701 (S513; No), the process proceeds to step S515.

  In process S514, the dialogue control unit 105 selects one piece of data to be confirmed to the user from the plurality of candidates. For example, the selection may be made randomly or in the order in which the paraphrasing is registered as the data of the candidate data 1072.

  In step S515, the dialogue control unit 105 generates a confirmation sentence from the data of the candidate data 1072, and sets it as a reply sentence to the user.

  In the above example, when a question sentence “What is business hours?” Is input from the user, first, in the process S305, a question sentence close to the question sentence is searched from the dialogue data 1071, and based on the result. The candidate data is used to perform a process of confirming the intention of the question to the user, and an answer sentence is output. However, when importance is attached to the relevance ratio of the answer sentence with respect to the question sentence, a threshold value may be set so that the answer is not made unless the score of the search result is equal to or greater than the threshold value.

<2-4. Answer processing in dialog control processing>
FIG. 8 shows an example of the answer process S306 in the dialog control process S203. The dialogue control unit 105 uses the dialogue data 1071 in the answer process S306. In this example, the dialogue data 1071 stores a dialogue ID, a question sentence, and an answer sentence in association with each other as shown in FIG. Accordingly, in S306, for example, the dialogue control unit 105 outputs an answer sentence “business hours are from 10:00 to 18:00” in response to a question sentence “What is business hours?” With a dialogue ID of 1. To do. In the answer process S306, as described above, in addition to the dialogue data 1071, the dialogue log data 1074 may be used to output an answer sentence based on the dialogue so far.

<2-5. Listening process in dialog control process>
FIG. 9 shows an example of the listening process S307 in the dialogue control process S203. For example, in response to a question sentence from the user “What is business hours?”, The dialogue control unit 105 outputs an answer sentence “I did not hear well. Please say again.” This process is executed when the confidence of the recognition result when converting speech to text due to the influence of surrounding noise is low, or when the intention analysis process fails (when the confidence of the intention analysis result is below the threshold) Is done. The answer text output in S307 may be stored in the memory 107 in advance.

  In this way, according to this system, even when a large-scale paraphrase corpus cannot be prepared in advance, it is possible to conduct a dialogue while confirming whether paraphrasing is appropriate in a series of dialogues with the user. And the accuracy of the automatically generated paraphrase text can be improved.

  In the first embodiment, in the process S304, the frequency N of the confirmation process (the confirmation process is performed once every N times) may be set by the system administrator. On the other hand, since this method requires an extra effort of confirming the appropriateness of paraphrasing to the user, it is desirable to set the frequency so that the user does not feel uncomfortable. Therefore, the frequency N can be automatically adjusted in consideration of the degree of user discomfort.

  FIG. 12 is a diagram illustrating an example in which the confirmation frequency optimization module 1201 included in the dialogue control unit 105 automatically adjusts the frequency N in consideration of the degree of user discomfort. In FIG. 12, for example, the confirmation log data 1202 is stored in the memory 107 of the interactive system 100, and the confirmation frequency optimization module 1201 reads the confirmation log data 1202 to determine the confirmation frequency in the dialogue with the user. The procedure for executing the confirmation process is shown.

  The confirmation log data 1202 is data representing the relationship between the frequency of confirmation processing executed at the time of interaction and the user's discomfort for each user. As shown in FIG. 12, in the confirmation log data 1202, for each user, a session ID for identifying a series of conversations, a confirmation frequency indicating the execution frequency of confirmation processing, and a user's discomfort are associated with each other. It is remembered. In FIG. 12, for example, among the conversations between the user A and this system, a series of conversations whose session ID is identified by “1” (for example, the log ID of the conversation log data 1074 shown in FIG. ”To“ 4 ”in a series of conversations from the beginning to the end), the confirmation process is executed once for 5 conversations, and the discomfort level of the user A at that time is“ 2: Somewhat It shows that it was "comfortable". The degree of discomfort depends on the interaction efficiency indicating whether the dialogue has been completed smoothly, the suppression of the voice of the user, the emotional analysis results from the user's face, gesture, and action images captured by an imaging device such as a camera What is necessary is just to calculate. In this example, the value of the discomfort level is expressed as a value from 1 to 5 (1: comfortable, 2: slightly comfortable, 3: normal, 4: slightly uncomfortable, 5: uncomfortable).

  The confirmation frequency optimization module 1201 is a processing unit that receives the confirmation frequency and the discomfort degree log at the time of dialogue as inputs and optimizes the confirmation frequency using a predetermined algorithm. For example, when the confirmation frequency is set so that at least the average discomfort is 3 or less, when the confirmation frequency is “5”, the average discomfort is “1.5” and the confirmation frequency is “4”. ", The average discomfort level is 3. Therefore, the confirmation frequency optimization module 1201 sets the confirmation frequency “4” satisfying the condition that at least the average discomfort is 3 or less among the two values as the confirmation frequency of the interactive system 100. In addition, although the case where the average of a discomfort degree was calculated as a statistical value was illustrated above, you may calculate a discomfort degree as other statistical values, such as a median value and a mode value. Thus, by setting the output frequency of the confirmation text according to the user's discomfort level, various users can interact with the dialogue system 100 comfortably.

  In the first embodiment, when there are a plurality of paraphrases in the process S514, data to be output is selected at random from the candidate data 1072. On the other hand, it is possible to select a paraphrase to be confirmed with priority over a plurality of paraphrases.

  For example, FIG. 10 shows an example of a candidate data selection processing flow when similar paraphrasing is summarized and it is desired to preferentially check a paraphrase different from the data registered in the dialogue data 1071.

  In process S1001, the dialogue control unit 105 clusters the candidate data 1072 by an arbitrary method. As one technique, the dialogue control unit 105 groups each data of the candidate data 1072 so that data having different endings are in the same group by character-analyzing each data stored in the candidate data 1072. A unique class is determined for each group. For example, in correspondence with a candidate ID in the candidate data 1072, “Tell me your name?”, “Tell me your name.”, “Tell me your name.”, “What are you calling?” When data is stored, the first three data are classified into one class, and the last one data is classified into a class different from the class.

  In step S1002, the dialogue control unit 105 selects a data class having a low similarity to the data registered in the dialogue data 1071 from the group obtained as a result of the step S1001. For example, the dialogue control unit 105 calculates the average value of the similarity between each data included in the grouped class and the registered data, and sets the average value as the similarity between the class and the registered data. For the calculation of the similarity, a conventionally known method such as Levenshtein distance may be used.

  In process S1003, the dialogue control unit 105 rephrases data that has a low similarity with the data registered in the dialogue data 1071 and does not satisfy the predetermined threshold from the data included in the class selected in step S1002. Select as a candidate. For example, the dialogue control unit 105 calculates the similarity between each data included in the selected class and the data registered in the dialogue data 1071 using the same method as described above, and the degree of similarity is the lowest among the predetermined values. The data that does not satisfy the threshold is selected.

  By clustering as described above, similar paraphrases can be collected, and similar confirmation sentences are not output continuously. In addition, by confirming the appropriateness from those with low similarity, it becomes possible to preferentially output and evaluate phrases using different vocabularies and grammars. It can be expected that construction will be possible earlier.

100 Dialog system 101 CPU
102 Data Communication Unit 103 Input / Output Device 104 Intention Analysis Unit 105 Dialogue Control Unit 106 Maintenance Unit 107 Memory 1071 Dialogue Data 1072 Candidate Data 1073 Paraphrase Confirmation Result Data 1074 Dialogue Log Data 108 Input Processing Unit 109 Output Processing Unit 1201 Confirmation Frequency Optimization Module 1202 Confirmation log data.

Claims (9)

An information processing apparatus that outputs an answer to a question from a user and interacts with the user,
An input unit for receiving the question;
In the dialogue, a candidate for answering the question received by the input unit is selected from candidate data, and a confirmation process is performed to confirm the suitability of the paraphrase of the question to the user, and based on the result of suitability A dialogue control unit for registering the candidate as a paraphrase for the question;
An information processing apparatus comprising: The dialogue control unit determines that the suitability is appropriate when the number of times the question paraphrase is determined to be appropriate and the number of times the question paraphrase is determined to be inappropriate satisfy a predetermined relationship, Registering the candidate as a paraphrase for the question;
The information processing apparatus according to claim 1. The dialogue control unit determines that the suitability is inappropriate when the number of times the question paraphrase is determined to be appropriate and the number of times the question paraphrase is determined to be inappropriate do not satisfy a predetermined relationship. And deleting the candidate registered as the paraphrase,
The information processing apparatus according to claim 1. The dialogue control unit executes the confirmation process in the dialogue at a predetermined frequency.
The information processing apparatus according to claim 1. The dialogue control unit calculates a statistical value of the degree of discomfort for each dialogue of the user based on confirmation log data in which the frequency of the confirmation processing and the degree of discomfort of the user are stored for each dialogue. Performing the confirmation process at the frequency corresponding to the statistical value of the discomfort level
The information processing apparatus according to claim 4. The dialogue control unit clusters a plurality of candidates into one or more classes, and paraphrases the candidates classified as a class whose similarity with a registered question does not satisfy a predetermined threshold. As a preferential confirmation process,
The information processing apparatus according to claim 1. The paraphrase confirmation result data including the question, the answer to the question and the result of the suitability are output to the display unit,
A maintenance unit that accepts selection of the candidate to be registered as a paraphrase for the question from the administrator via the display unit;
The information processing apparatus according to claim 1, further comprising: Accept questions from users,
In an interaction with the user that outputs an answer to the question,
Select a paraphrase candidate for the question from candidate data,
Execute a confirmation process to confirm to the user whether or not the paraphrasing of the question is appropriate;
Registering the candidate as a paraphrase for the question based on the adequacy result;
An interactive processing method characterized by the above. On the computer,
Process to accept questions from users,
In an interaction with the user that outputs an answer to the question,
A process of selecting a paraphrase candidate for the question from candidate data;
A process of executing a confirmation process for confirming to the user whether or not the paraphrasing of the question is appropriate;
A process of registering the candidate as a paraphrase for the question based on the result of suitability;
A dialog processing program characterized by causing

Images