JP6486760B2 - Dialog pattern automatic generation apparatus, method and program - Google Patents

Description

  BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dialog pattern automatic generation apparatus, method, and program, and in particular, a dialog pattern automatic generation apparatus, method, and program that can automatically generate a dialog pattern for the user based on the user's experience information and can be used effectively for memory recall support Regarding the program.

  In the past, human experiences were generally left behind in images such as pictures and videos, as well as in texts such as diaries, but in recent years, they are also usually left behind in records such as blogs. It was.

  When trying to recall the memory of past experiences from such a recorded material, the user needs to narrow down the recorded material of the experience to be recalled and look through all the sentences of the narrowed-down recorded material.

  Patent Document 1 proposes a dialogue corpus generation device having a mechanism for automatically constructing a dialogue corpus using two or more dialogue sentences such as chat.

JP 2008-299754 A

  By automatically narrowing down recorded items for memory recall from recorded items such as blogs and automatically generating dialogue patterns based on them, memory recall can be efficiently supported by reducing the burden on the user.

  The dialog corpus generation device proposed in Patent Document 1 automatically generates a dialog corpus, but cannot generate a dialog pattern for memory recall for a user. That is, the dialog corpus generation device of Patent Document 1 is premised on a dialog format. However, the memory recall is based on a recorded material in the past regarding the user's behavior and behavior (experience), and a diary-based recorded material is also an object of the memory recall, and cannot be handled in an interactive format.

  Further, when generating a dialog pattern for memory recall for a user based on a recorded material such as a diary format, it is necessary to extract information necessary for generating the dialog pattern from the recorded material. However, for this purpose, there are problems (1) and (2) that must be solved.

  (1) If the recorded material has been collected for a long period of time (for example, 10 years), the recorded material that the user wishes to recall must be narrowed down.

  (2) Even if the recorded material that the user wants to remember is narrowed down, it takes too much time to read all the sentences of the recorded material. In addition, the conversation is unilateral and does not become a conversation. In addition, conversations must be advanced little by little (for example, one sentence at a time) and effectively (in an order that is easy to remember).

  SUMMARY OF THE INVENTION An object of the present invention is to provide an interactive pattern automatic generation apparatus, method, and program that can solve the above-mentioned problems, can automatically generate an interactive pattern for the user based on the user's experience information, and can be used effectively for memory recall support. It is in.

  In order to solve the above-described problems, the present invention provides an automatic dialogue pattern generation apparatus that automatically generates a dialogue pattern for a user based on the user's experience information, and each sentence of the body of one or more accumulated experience information. Based on the experience element extraction means for extracting the experience elements from the experience elements and the experience elements extracted by the experience element extraction means, a score is calculated for each sentence, and the degree of whether each sentence is effective for memory recall support is calculated for each sentence. And a score calculation means for giving priority to each sentence according to the score, and a dialogue pattern generation means for generating a dialogue pattern based on the priority given by the score calculation means and the template for dialogue pattern generation. It is characterized by.

  Here, it is preferable that the experience element extraction means decomposes the text of the experience information for each sentence, and extracts words and idioms given part-of-speech information from each sentence as experience elements.

  The score calculation means calculates a similarity score with each sentence of other experience information based on the experience element extracted by the experience element extraction means and its superordinate concept, and uses the similarity score as the score It is also preferable.

  Moreover, it is also preferable that the score calculating means calculates a similarity score of each sentence by the following formula.

  Similar score of sentence A = average value {number of matches / number of experience elements included in sentence A}

  Here, the number of matches is the number of experience elements and their superordinate concepts that match between the sentence A and the sentences of other experience information.

  Further, the score calculating means further calculates an emotion score of each sentence based on the number of emotion words included in each sentence, and calculates a total score of each sentence by a product of the emotion score and the similarity score. It is also preferable to calculate and use the total score as the score.

  Emotion score of sentence A = 1-number of emotion words / number of experience elements included in sentence A

  Moreover, it is also preferable that the score calculation means calculates an emotion score of each sentence by the following formula based on the number of emotion words included in each sentence, and sets the emotion score as the score.

  Emotion score of sentence A = 1-number of emotion words / number of experience elements included in sentence A

  Further, the dialog pattern generation template describes sentence fitting in the dialog pattern with priority, and is prepared separately according to the type of the dialog pattern, and the dialog pattern generation means includes the dialog pattern generation means. It is also preferable to generate a dialogue pattern by applying the priority sentence given by the score calculation means according to the priority described in the generation template.

  The present invention can be realized not only as an automatic dialog pattern generation apparatus but also as an automatic dialog pattern generation method having steps for sequentially executing processing in the automatic dialog pattern generation apparatus. It can also be realized as a program for functioning as each part.

  According to the present invention, from the viewpoint of whether it is effective for memory recall support, priorities are automatically assigned to the experience sentences included in the experience information, and the experience information effective for memory recall support is efficiently used according to the priorities. It is possible to automatically generate a dialog pattern by selecting the target. Thereby, it is possible to realize an agent function that supports chat concerning user experience and user memory recall.

  For example, when there are multiple experiences as a memory recall target, it is possible to efficiently narrow down in the interactive format by creating a dialogue sentence for the user from the highest priority experience sentence included in the record of each experience And can be realized easily. Thereby, the above problem (1) can be solved.

  In addition, prioritize the use of conversations for multiple experience sentences contained in individual experience records in terms of whether they are effective for memory recall support, and construct the conversations according to these priorities. Thus, the above problem (2) can be solved.

  In addition, low-priority experience sentences that are not subject to fitting in the generation of experience patterns do not need to be stored in the experience information DB, and low-priority experience sentences are not stored in the experience information DB, so they can be stored and searched. Can reduce resources.

It is a functional block diagram which shows one Embodiment of the dialogue pattern automatic generation apparatus which concerns on this invention. It is a flowchart which shows the operation | movement of a memory recall support in case a conversation pattern classification is a chat. It is a flowchart which shows the operation | movement of a memory recall support in case a dialogue pattern classification is a user memory recall.

  The present invention will be described below with reference to the drawings. Hereinafter, the case where the present invention is realized as an automatic dialog pattern generation device will be described. However, the present invention can also be realized as an automatic dialog pattern generation method including steps for sequentially executing processing in the automatic dialog pattern generation device. It can also be realized as a program for causing a computer to function as each part of the dialog pattern automatic generation device.

  FIG. 1 is a functional block diagram showing an embodiment of an automatic dialogue pattern generation apparatus according to the present invention.

  The dialog pattern automatic generation device of this embodiment includes an experience information collection function unit 1, a dialog pattern automatic generation function unit 2, and a dialog format search function unit 3. These functional units 1 to 3 can be configured by one or a plurality of processors, and the configuration may be hardware or software.

  The experience information collection function unit 1 includes an experience information collection unit 1-1 and a text extraction unit 1-2.

  The experience information collection unit 1-1 collects the user's experience information, for example, blog articles, from a server on the network 4 such as the Internet.

  The text extraction unit 1-2 extracts the text included in the collected experience information. The extracted text is stored in the raw data storage database (DB) 5 together with information on bibliographic items such as date and time and titles other than the text. For example, in the case of experience information collected using a general blog service, since the advertisements and menus included in the HTML are information other than the text, they are not extracted by the text extraction unit 1-2. Note that the raw data storage DB 5 may be built in the dialog pattern automatic generation device or connected to the outside.

Extraction of the text can be performed, for example, by the following methods (1) and (2).
(1) HTML tag patterns are extracted and stored in advance for each Web site, and HTML tag patterns before and after the text are detected using this pattern, and the text is extracted based on the detected HTML tag pattern.
(2) For the part other than the tag included in HTML, the part where the text content appears most frequently is extracted as the text.

  The dialogue pattern automatic generation function unit 2 includes an experience element extraction unit 2-1, a superordinate concept extraction unit 2-2, a score calculation unit 2-3, and a dialogue pattern generation unit 2-4.

  The experience element extraction unit 2-1 extracts experience elements from the text of the experience information. Experience elements are words or idioms. The body of experience information is decomposed into sentences, and the experience elements are extracted from each sentence. The decomposition of the text into sentences can be realized by dividing the text into predicate term structures using a predicate term structure analyzer. The text can also be decomposed into sentences by the technique of dividing with punctuation marks. For extraction of experience elements, existing morphological analysis tools can be used, and thereby, experience elements can be extracted by identifying nouns and verbs.

  The superordinate concept extraction unit 2-2 extracts superordinate concept words for each noun by assigning superordinate concept words to the nouns extracted by the experience element extraction unit 2-1. For example, a high-order concept word can be assigned to each noun by preparing a tree-like high-order concept dictionary in advance and referring to it. If there are a plurality of high-level concept words for a noun, all the high-level concept words are extracted. Here, the reason why a higher concept word is assigned to each noun is that, in calculating a similarity score, which will be described later, not only the noun word itself but also the higher concept word is taken into account, and matches between words are counted.

  Based on the experience elements and higher-level concept words extracted by the experience element extraction section 2-1 and the higher-level concept extraction section 2-2, the score calculation section 2-3 uses the experience element text from the body of the experience information. A score obtained by quantifying the degree to which the decomposed sentence (hereinafter referred to as an experience sentence) is effective for memory recall support is calculated for each experience sentence.

  For example, as shown in the following formulas (1), (2), and (3), a similarity score and an emotion score are calculated, and a total score is calculated by the product of them, and the total score is used as the score of the experience sentence A. be able to.

  Similar score of experience sentence A = average value {number of matches ÷ number of experience elements included in experience sentence A} (1)

  Emotion score of experience sentence A = 1-number of emotion words included in experience sentence A ÷ number of experience elements included in experience sentence A (2)

  Total score of testimonial A = similarity score of testimonial A * sentiment score of testimonial A (3)

  The similarity score of experience sentence A is included in experience sentence A, as shown in the above equation (1), by finding the number of matches of experience elements and their broader concept words with each experience sentence of other experience information It is calculated by averaging the ratio of the number of matches to the number of experience elements that are received. The similarity score of the experience sentence A shows the difference, that is, the similarity between the experience sentence A and the experience sentence of other experience information, and becomes smaller as the experience sentence A includes characteristic items.

  The emotion score of the experience sentence A is calculated by the above formula (2) based on the number of emotion words included in the experience sentence A. The emotion score indicates the number of emotion words in the experience sentence A, and becomes smaller as the experience sentence A includes more emotional expressions.

  The total score of the experience sentence A is calculated by the product of the similarity score and the emotion score of the experience sentence A as shown in the above equation (3). The total score of the experience sentence A becomes smaller as the experience sentence A is not similar to the experience sentence of other experience information and contains more emotional expressions.

  What is useful for user memory recall is that (1) the experience is a characteristic that other experiences do not have, and the number of coincidence of experience elements and high-level concept words in the experience sentence and other experience sentences is small (2) The experience has a feeling that the experience is touching or sad, and the experience sentence has a lot of emotional expressions.

  From the above (1), (2), it can be said that the experience sentence with a smaller value of the score (overall score) is more effective for the memory recall support of the user, and from the viewpoint of supporting the recall of the experience with the score, A priority can be given to each experience sentence.

  The score of each experience sentence calculated as described above is arranged in ascending order for each experience information, and the experience sentence with a score below the predetermined threshold is given a higher priority as the score becomes lower. It accumulates in the experience information accumulation DB6 along with the sentences. As will be described later, when the bibliographic information such as the date and time of the experience information and the title is also used when generating the dialogue pattern, the information is also accumulated in the experience information accumulation DB 6. Information on these bibliographic items can be acquired from the raw data storage DB 5 without being stored in the experience information storage DB 6. In addition, the experience information storage DB 6 may be built in the dialog pattern automatic generation device or connected to the outside.

  The dialog pattern generation unit 2-4 generates a dialog pattern based on the experience sentence priority of the experience information stored in the experience information storage DB 6 and the dialog pattern generation template. The dialogue pattern generation template describes the fitting of experience sentences in the dialogue pattern with priority. For the dialog pattern generation template, prepare different templates depending on the dialogue pattern type, such as memory recall support by chat that asks the user from the agent (device) side or memory recall support that asks the agent from the user side. It is preferable to keep it. The dialogue pattern generation unit 2-4 applies the experience sentence with the priority given by the score calculation unit 2-3 according to the priority described in the dialogue pattern generation template to generate a dialogue pattern.

  The interactive search function unit 3 includes an interactive UI (User Interface) unit 3-1 that mediates between an interactive pattern generation unit (agent) 2-4 and a user. The dialogue UI unit 3-1 has a function of inputting a user's words and outputting an agent's words. With this function, for example, when a user inputs a character using a keyboard, the agent's words are displayed on the display. Alternatively, when the user inputs the speech into the microphone, the speech is recognized and converted into text, and the agent's response to the speech is converted into voice and output from the speaker.

  2 and 3 are flowcharts showing an operation of generating a dialogue pattern based on the experience sentence priority of the experience information and the dialogue pattern generation template. This operation is realized by the dialog pattern generation unit 2-4 (agent) interacting with the user via the dialog UI unit 3-1. In FIG. 2 and FIG. 3, the experience sentence with the highest priority of the experience information is described as [experience sentence with priority 1], and among the experience sentences after the next highest priority of the experience information, dialogue Is described as [Experience sentence after priority 2]. Note that the [experience sentence after priority 2] may be applied, for example, in descending order of priority.

  The template for generating a dialogue pattern is prepared in advance by leaving the columns of [Experience sentence with priority 1], [Experience sentence after priority 2] and [Proper noun] in the flowcharts of FIGS. The dialogue pattern generation unit 2-4 displays the corresponding experience according to the [Experience sentence with priority 1], [Experience sentence after priority 2], and [Proper noun] indicated in the blank of the template for generating the dialogue pattern. A dialogue pattern is generated by applying sentences and proper nouns. It is assumed that the experience sentence of the user experience information and the priority of each experience sentence have already been accumulated in the experience information accumulation DB 6.

  FIG. 2 is a flowchart showing the operation of the memory recall support when the agent (electronic secretary on the device side) asks the user, that is, when the conversation pattern type is chat.

  When the dialogue pattern type is chat, the user or the like instructs to start the dialogue (S21). Thereby, the dialogue pattern generation unit 2-4 selects an appropriate piece of experience information stored in the experience information storage DB 6 (S22). For selection of the experience information, for example, a genre, a time of experience, a region, and the like may be selected as conditions, and an item satisfying the conditions may be selected at random. Since the conversation pattern type is chat, it may be selected simply at random.

  Next, an experience sentence with priority 1 of the selected experience information is selected, and the experience sentence is applied to the blank of [Experience sentence with priority 1] to utter (S23). As a result, the agent utters, for example, “[the highest priority experience sentence]” to the user. In addition, in this utterance, “Let ’s say,” was added to the end of the utterance to create a conversational quality. The same applies to the subsequent utterances.

  Next, it is determined whether or not there is a user response to the agent utterance in (S23) (S24), and if it is determined that there is no user response, it is determined whether or not a certain time has passed (S25), Here, if it is determined that the user does not respond even after a certain period of time has elapsed, the user is informed that the conversation will proceed to another topic (S26). In (S26), the user may be informed that the dialogue is to be terminated, and the dialogue may be terminated. When it is informed that the conversation is to be switched to another topic, the process returns to the experience information selection (S22), and another experience information is selected to start the operation.

  If it is determined in (S24) that there is a user response, the user response (contents) is further determined (S27). The user's reaction can be recognized by recognizing the user's voice and comparing the pattern with a previously registered pattern.

  If the user's response (S27) is a utterance of a word similar to a high-priority (low score) experience sentence, the agent utters a nod word and already interacts with the user in the experience information. A search is made from the experience information storage DB 6 for experience sentences other than the experience sentence that has been prioritized, from the experience information storage DB 6, and the experience sentence is applied to the blank of the [experience sentence after priority 2] (S28). As a result, the agent utters, for example, “Yes, [Experience sentence after priority 2]”.

  Next, it is determined whether or not the experience sentence (with priority given) of the experience information remains (S29), and if it is determined that it remains, the process returns to (S24) and the agent The presence or absence of the user's reaction to the utterance and the reaction are determined, and the operation is continued according to the determination result.

  In the subsequent (S28), the agent sequentially searches and speaks experience sentences of priority 2 or later, other than the experience sentences that have already interacted with the user. If it is determined in (S29) that there is no experience sentence of the relevant experience information, the user is notified that the conversation will move to another topic. Alternatively, the dialogue may be terminated by informing the user that the dialogue is to be terminated (S26).

  In (S27), when the user utters a word similar to a high-priority (low score) experiential sentence, the experiential sentence of the relevant experience information (with priority assigned) no longer remains In (S28), for example, only “Yes, so” is uttered, and the user is informed that the conversation will shift to another topic. Alternatively, the dialog may be terminated by informing the user that the conversation is to be terminated.

  When the user's reaction (S27) is a speech that seems to be a reminder, the agent uses the experience information storage DB 6 for the experience information of priority 2 or later other than the experience sentences that have already interacted with the user in the experience information. , And the experience sentence is applied to the blank of [Experience sentence after priority 2] and uttered (S30). As a result, the agent utters a confirmation message to the user, for example, “Sure, [Experience sentence after priority 2]”.

  Next, it is determined whether or not the experience sentence of the experience information remains (S29) .If it is determined that the experience sentence remains, the process returns to (S24) and the user response to the agent utterance is returned. The presence / absence and the reaction are determined, and the operation is continued according to the determination result.

  In the subsequent (S30), the agent searches for an utterance sentence of priority 2 or later other than the empirical sentence that has already interacted with the user and speaks. If it is determined in (S29) that the experience information of the experience information does not remain, the user is informed of the transition to another topic conversation. Alternatively, the dialogue may be terminated by informing the user that the dialogue is to be terminated (S26).

  If the user's response (S27) is uttered in an unintelligible language or something other than the above, the agent further extracts proper nouns from the experience information and utters utterances to recall the experience. (S31). Any number of proper nouns may be extracted here, and in addition to proper nouns, dates and titles may be extracted and spoken. In (S31), the agent speaks to the user, for example, “Do you remember [proprietary noun] or [proprietary noun]?”.

  After that, it is determined whether or not the experience sentence of the experience information remains (S29), and if it is determined that it remains, the process returns to (S24) and the presence or absence of the user's reaction to the utterance of the agent And the reaction is determined, and the operation is continued according to the determination result.

  FIG. 3 is a flowchart showing the operation of the memory recall support when the user asks the agent, that is, when the conversation pattern type is the user memory recall.

  When the dialogue pattern type is user memory recall, first, the user starts to utter an experience that the user wants to remember. The agent detects the user's utterance (S31), and determines whether the utterance is related to user memory recall or other than user memory recall (S32).

  Here, when it is determined that the user's utterance is not related to the memory recall, it is determined that the conversation is impossible for the utterance, and the operation is terminated. Alternatively, another dialogue pattern type may be referred to and operated using a dialogue pattern generation template of the dialogue pattern type for the utterance (S33).

  When it is determined that the user's utterance is related to memory recall (S32), the agent searches the experience information storage DB 6 for experience information including nouns included in the utterance, and the experience sentence with the priority 1 of the experience information. Select to speak. For example, when a user's “(word to recall something)” is detected (S31), the agent utters “Maybe it means [experience sentence of priority 1]?”.

  Next, it is determined whether or not there is a user response to the agent utterance in (S34) (S35), and if it is determined that there is no user response, it is determined whether or not a certain time has passed ( Here, if it is determined that the user does not respond even after a lapse of a certain time, the user is informed that the user is prompted to move to another topic conversation (S37). In (S37), the user may be informed that the dialogue is to be terminated, and the dialogue may be terminated. If it is informed that it is urged to move to another topic conversation, in S31, the user's utterance is newly detected and the operation is started.

  When it is determined in (S35) that there is a user's reaction, the user's reaction (content) is further determined (S38). The user's reaction can be recognized by recognizing the user's voice and comparing the pattern with a previously registered pattern.

  If the user's response (S38) is utterance of a word similar to a high-priority (low score) experience sentence, the agent utters a nodding word and already interacts with the user in the experience information. A search is made from the experience information storage DB 6 for experience sentences other than the experience sentence that has been prioritized, from the experience information storage DB 6, and the experience sentence is applied to the blank of the [experience sentence after priority 2] (S39). As a result, the agent utters, for example, “Yes, [Experience sentence after priority 2]”.

  Next, it is determined whether or not the experience sentence (with priority assigned) of the experience information remains (S40), and if it is determined that it remains, the process returns to (S35) and the agent The presence or absence of the user's reaction to the utterance and the reaction are determined, and the operation is continued according to the determination result.

  In the subsequent (S39), the agent searches for an utterance sentence with a priority of 2 or later other than the empirical sentence already interacted with the user and speaks.

  If it is determined in (S40) that the experience information of the experience information does not remain, the user is informed that the user is prompted to move to another topic conversation. Alternatively, the dialog may be ended by informing the user that the dialog is to be ended (S37).

  In (S38), when the user utters a word similar to a high-priority (low score) experience sentence, the experience sentence (with priority assigned) no longer remains In (S39), for example, only “Yes, yes” is spoken, and the user is notified that the conversation will move to another topic. Alternatively, the dialog may be terminated by informing the user that the conversation is to be terminated.

  If the user's reaction (S38) is a speech that seems to be a reminder, the agent uses the experience information storage DB to store experience sentences of priority 2 or later, other than those already interacted with the user, in the experience information. And utters by applying the experience sentence to the blank of [Experience sentence of priority 2] (S41). As a result, the agent utters a confirmation message to the user, for example, “Sure, [Experience sentence of priority 2]”.

  Next, it is determined whether or not the experience sentence of the experience information remains (S40) .If it is determined that the experience sentence remains, the process returns to (S35) and the response of the user to the agent utterance is returned. The presence / absence and the reaction are determined, and the operation is continued according to the determination result.

  In the subsequent (S41), the agent searches for an utterance sentence with a priority of 2 or later other than the empirical sentence already interacted with the user and speaks. If it is determined in (S40) that the experience information of the relevant experience information does not remain, the user is informed that it is urged to move to another topic conversation. Alternatively, the dialogue may be terminated by informing the user that the dialogue is to be terminated (S26).

  If the user's response (S38) is an utterance of an unrecognizable word or something other than the above, the agent further extracts a proper noun from the experience information and utters an utterance to recall the experience. (S42). Any number of proper nouns may be extracted here, and in addition to proper nouns, dates and titles may be extracted and spoken. In (S42), the agent speaks to the user, for example, “Do you remember [proprietary noun] or [proprietary noun]?”.

  After that, it is determined whether or not the experience sentence of the relevant experience information remains (S40), and if it is determined that it remains, the process returns to (S35) and the presence or absence of the user's reaction to the utterance of the agent And the reaction is determined, and the operation is continued according to the determination result.

Hereinafter, the case where the conversation pattern type is user memory recall will be described with a specific example. Here, the experience information of the user's experience is as follows.
・ Title: Travel to Hokkaido ・ Date: October 14, 2014 ・ Text:
(A) It was when I was driving in Hokkaido.
(B) I found a little girl harvesting an onion mixed with an adult.
(C) Especially the scene of a girl helping Grandpa is impressive,
(D) I was impressed and asked me to take a picture.

  Here, in the sentence (B), “I found a little girl harvesting an onion mixed with an adult.” Is a rare combination of words such as “onion”, “harvest”, and “girl”. Because it does not include emotional expressions and describes the facts of the experience, it is judged as the experience sentence with the lowest score and priority 1.

  (C) sentence “Especially the scene of a girl who helps Grandpa is impressive,” and (D) sentence “I was impressed and asked to take a picture.” (B ) Is a combination of words that are more likely to appear than sentences, but since it includes emotional expressions, it is determined as an experience sentence with a score of not more than a predetermined threshold and a priority of 2 or later.

  In the sentence (A), “When I was driving in Hokkaido,” the word combination is normal and does not include emotional expressions, so the score exceeds the specified value and no priority is given. A sentence.

  Based on the above determination results, the dialog pattern generation unit (agent) generates a dialog pattern as follows using the dialog pattern generation template of the user memory recall dialog pattern type.

(1) User's utterance Word that tries to remember something (eg, “I think I went to Hokkaido a long time ago”)

(2) Search for experience information including “Hokkaido” of the user's utterance and agent's utterance.
2-1: Select an experience sentence with a priority of 1 from the searched experience information (the above experience information) and utterance (example: “I found a place where a small girl is mixed with an adult and harvesting an onion. ? ")

(3) User utterance in response to agent utterance (3-1 or 3-2 or 3-3)
3-1: Words similar to experience sentences after priority level 2 (Example: “I ’ve asked you to take a picture without hesitation”)
3-2: Words I remembered (example: “Oh! That happened”)
3-3: Words that have no idea (example: “Oh, what is it?”)
If there is no user utterance for a certain period of time, utter the word that ends the conversation (ex. "Did you remember?"

(4) Agent processing and utterances for user utterances (4-1 or 4-2 or 4-3)
For user utterance 3-1
4-1: Select a sentence with a priority of 2 or higher that has not been used in the dialogue yet (eg: yes, “The scene of a girl who helps Grandpa was particularly impressive”)
For user utterance 3-2
4-2: Select a test sentence of priority 2 or later that has not been used in the dialogue yet, and confirm the utterance (eg: “Sure, I was impressed and asked to take a picture. ")
For user utterance 3-3
4-3: Extract date and time, title of experience information, proper nouns, and utterances (ex. “Take a trip to Hokkaido last October. Biei no Oka?”)

  The user's utterances for the agent's utterances 4-1, 4-2, and 4-3 are the above (3). However, if there is no experience sentence of the relevant experience information in the course of 2-1 to 3-1 to 4-1, prompt the user to move to another topic conversation or Inform the end. In addition, in the user's utterance 3-1 for the utterance of the agent, a word similar to a high priority (small similarity score) experience sentence is uttered, and at that stage, if there is no experience sentence of the relevant experience information, You can urge them to move to another topic conversation after speaking such as "I'm sure," or "I'm nostalgic," or let them know that the conversation is over.

  Although the embodiment has been described above, the present invention is not limited to the above embodiment, and includes various modifications. For example, in the above embodiment, the product of the similarity score and the emotion score is calculated as the score of the experience sentence A. However, since the value of the similarity score or the emotion score is effective for memory recall support alone, the value May be used as the score of the experience sentence.

  Also, when calculating the score of experience sentence A, if the other experience information to be compared is past from the experience of experience sentence A, the experience is characteristic by the time of experience sentence A. If the other experience information to be compared is made past from the current time and the similarity score is updated, a score that the experience is characteristic up to the present time can be calculated. In the present invention, either similarity score may be adopted.

  In the above embodiment, the experience information is collected from a server on a network such as the Internet. However, the experience information may be collected from any location, and the experience that accumulates only the text of the experience information. When the information input function and server dedicated to information storage can be used, extraction of the text from the experience information can be omitted.

  Furthermore, in the generation of the experience pattern, since the experience sentence with a low priority is not applied, it is not necessary to accumulate all the experience sentence of the experience information in the experience information DB.

  1 ... Experience information collection function unit, 1-1 ... Experience information collection unit, 1-2 ... Text extraction unit, 2 ... Interaction pattern automatic generation function unit, 2-1 ... Experience element Extraction unit, 2-2 ... higher concept extraction unit, 2-3 ... score calculation unit, 2-4 ... dialog pattern generation unit, 3 ... interactive format search function unit, 3-1.・ Dialogue UI part, 4 ... Network, 5 ... Raw data storage DB, 6 ... Experience information storage

Claims (9)

An interactive pattern automatic generation device that automatically generates an interactive pattern for a user based on user experience information,
An experience element extraction means for extracting an experience element from each sentence of the body of one or more accumulated experience information;
Based on the experience element extracted by the experience element extraction means, a score is calculated for each sentence, and the degree of whether each sentence is effective for memory recall support is calculated for each sentence, and priority is given to each sentence according to the score. A score calculation means;
A dialogue pattern automatic generation apparatus comprising dialogue pattern generation means for generating a dialogue pattern based on the priority given by the score calculation means and the dialogue pattern generation template.   2. The dialogue pattern automatic generation according to claim 1, wherein the experience element extraction unit decomposes the text of the experience information for each sentence, and extracts words and idioms given part-of-speech information from each sentence as experience elements. apparatus.   The score calculation means calculates a similarity score with each sentence of other experience information based on the experience element extracted by the experience element extraction means and its superordinate concept, and sets the similarity score as the score The dialog pattern automatic generation device according to claim 1 or 2, characterized by the above. 4. The dialogue pattern automatic generation apparatus according to claim 3, wherein the score calculation means calculates a similarity score of each sentence by the following formula.
Similar score of sentence A = average value {number of matches / number of experience elements included in sentence A}
Here, the number of matches is the number of experience elements and their superordinate concepts that match between the sentence A and the sentences of other experience information. The score calculating means further calculates an emotion score of each sentence based on the number of emotion words included in each sentence, and calculates a total score of each sentence by a product of the emotion score and the similarity score. The dialog pattern automatic generation apparatus according to claim 3, wherein the total score is the score.
Emotion score of sentence A = 1-number of emotion words / number of experience elements included in sentence A 3. The score calculation unit according to claim 1, wherein the score calculation unit calculates an emotion score of each sentence based on the number of emotion words included in each sentence, and uses the emotion score as the score. Dialog pattern automatic generation device.
Emotion score of sentence A = 1-number of emotion words / number of experience elements included in sentence A   The dialog pattern generation template describes sentence fitting in the dialog pattern with priority, and is prepared separately according to the type of the dialog pattern. The dialog pattern generation means includes the dialog pattern generation means. 7. The dialogue pattern according to claim 1, wherein the dialogue pattern is generated by applying the sentence of the priority given by the score calculation means according to the priority described in the template. Automatic generator. An interactive pattern automatic generation method for automatically generating an interactive pattern for a user based on user experience information,
An experience element extraction step for extracting an experience element from each sentence of the body of one or more accumulated experience information;
Based on the experience element extracted in the experience element extraction step, a score is calculated for each sentence, which is a numerical value indicating whether each sentence is effective for memory recall support, and priority is given to each sentence according to the score. A score calculating step;
A dialog pattern automatic generation method comprising a dialog pattern generation step of generating a dialog pattern based on the priority given in the score calculation step and the dialog pattern generation template. A program for automatically generating an interaction pattern for a user based on the user's experience information, comprising:
Experience element extraction means for extracting experience elements from each sentence of the body of one or more accumulated experience information,
Based on the experience element extracted by the experience element extraction means, a score is calculated for each sentence, and the degree of whether each sentence is effective for memory recall support is calculated for each sentence, and priority is given to each sentence according to the score. Score calculation means,
A program for functioning as an interaction pattern generation unit that generates an interaction pattern based on the priority given by the score calculation unit and an interaction pattern generation template.