JP6858721B2 - Dialogue controls, programs and methods capable of conducting content dialogue - Google Patents

Description

The present invention relates to a technology of a dialogue system for having a dialogue with a user or the like.

In recent years, the development of a dialogue system capable of interacting with a user and providing various services to the user has been actively promoted. For example, many smartphones that are currently in widespread use are always equipped with a dialogue response function that responds to inquiries spoken by the user by voice and provides information such as search results. In addition to such task-type dialogues, communication robots, which are currently a hot topic, engage in dialogues such as chats with users and provide services that meet various needs in recreation and nursing care. It is possible.

At present, in the development site of such a dialogue system, how to improve the satisfaction obtained by the user as a result of the dialogue is one of the issues. In response to such a problem, for example, in Patent Document 1, the type of response having self-disclosure, empathy, and non-empathy as elements is determined for the user's utterance, and words are input from the database according to the response type. An interactive device that is selected and generates a response on the device side is disclosed.

Specifically, the user's utterance is analyzed to determine the dialogue action type, the domain knowledge database is referred to to obtain a responsive dialogue action type sequence, and self-disclosure and empathy dialogue actions are performed from the setting parameters. The probability is calculated. After that, the acquired dialogue action type sequence and the feature amount of self-disclosure and empathy expressed by the occurrence probability are compared, the dialogue action type sequence in which both feature amounts match is selected, and the accompanying information is added. A dialogue action sequence is generated, and the generated dialogue action sequence is converted into a natural sentence and presented to the user by voice or text.

Further, for example, Non-Patent Document 2 includes a dialogue system in which a classifier learns a user's utterance converted into a feature amount and a system's aizuchi for the user's utterance, and predicts an aizuchi suitable for the user's utterance. It is disclosed.

Further, for example, Non-Patent Document 3 discloses a dialogue system capable of giving a response indicating consent to the content of a user's utterance, a response of repeating a part of the user's utterance, or a response asking the details of the preceding utterance. Has been done.

Japanese Unexamined Patent Publication No. 2010-14582

Takashi Yamaguchi, Koji Inoue, Koichiro Yoshino, Katsuya Takanashi, Nigel G. Ward, Tatsuya Kawahara, "Generation of various forms of aizuchi based on linguistic and prosodic information for listening dialogue systems", Journal of the Japanese Society for Artificial Intelligence, 2016, Vol. 31, No. 4, Page: C-G31 10 Shinya Ishida, Koji Inoue, Shizuka Nakamura, Katsuya Takanashi, Tatsuya Kawahara, "Generation of Diverse Listener Responses for Listening Dialogue Systems", IPSJ National Convention Proceedings, 2016, Vol. 78, No. 2, 2513 ~ 2514 pages

However, even with the above-mentioned conventional technology, it cannot be said that the problem that the user's satisfaction is not increased and the activation of the dialogue is not promoted is sufficiently solved.

For example, the technique disclosed in Patent Document 1 employs a method of selecting a word from a domain knowledge database when generating a response on the device side. However, with such a method, only words registered in the database can be selected. As a result, the number of variations in the response is limited.

The problem that the number of such response variations is limited also exists in the technique disclosed in Non-Patent Document 2. In this technology, an aizuchi suitable for the user's utterance is predicted and used for the response on the system side. However, the number of aizuchi that can be trained by the learning machine is limited to a predetermined number of minutes. As a result, the variation of the response is scarce.

In any case, the limited number of response variations makes it impossible to implement a wide variety of responses, and in many cases the user is unable to empathize as a result of such dialogue. As a result, user satisfaction may decrease and the activation of dialogue may be hindered.

Further, the technique disclosed in Non-Patent Document 3 is the same as that of Non-Patent Document 2, but the system side responds positively or negatively to the positive or negative utterance on the user side. Can not do it. In many cases, as a result of such a dialogue, the user cannot obtain empathy, and as a result, the satisfaction of the user may decrease and the activation of the dialogue may be hindered.

Therefore, an object of the present invention is to provide a dialogue control device, a dialogue control program, and a dialogue control method capable of carrying out a response in which the user's empathy in dialogue is further improved.

According to the present invention, it is a dialogue control device capable of carrying out dialogue regarding contents.
Related text acquisition means to acquire related text that is the text related to the content, and
A polarity determining means for determining the polarity of the acquired related text and determining the polarity of the user's utterance when the user's utterance regarding the content is received.
An utterance generation means that receives the polarity determined in the user's utterance regarding the content and generates a content-related text utterance that is the utterance related to the related text from the related text having the polarity determined based on the received polarity. ,
As a response to the user's utterance regarding the content, a dialogue control device including an utterance control means for outputting the content-related text utterance is provided.

Further, in one embodiment of the dialogue control device according to the present invention, the utterance generation means may generate the content-related text utterance from the related text having the same polarity as the polarity determined in the user's utterance regarding the content. preferable.

Further, as another embodiment of the dialogue control device according to the present invention, the present device extracts words of a predetermined part of speech included in the utterance of the user side regarding the content and the acquired related text, and the device of the user side. It also has a scoring means for determining the score, which is the degree of matching of the word between the utterance and the relevant text.
It is also preferable that the utterance generation means generates the content-related text utterance from the related text indicating the highest or higher score than the user's utterance regarding the content.

Further, as a further embodiment of the dialogue control device according to the present invention, the device determines whether or not the honorific expression is included in the utterance on the user side regarding the content and the acquired related text. It also has a means of judging honorific expressions,
It is also preferable that the utterance generation means generates the content-related text utterance from the related text having the same determination result as the determination result in the user's utterance regarding the content.

Further, in the dialogue control device according to the present invention, the utterance generation means generates a content presentation utterance which is a utterance that presents the content from the title or summary in the data of the content.
It is also preferable that the utterance control means outputs the content presentation utterance and outputs the content-related text utterance as a response to the user's utterance regarding the content, which is performed after the output.

Further, as a further embodiment of the dialogue control device according to the present invention, the present device obtains information related to the user's hobbies and tastes from the user's profile information or information related to posting in a predetermined SNS (Social Networking Service). It further has a user information acquisition means for acquiring or determining, and a content acquisition means for acquiring content related to or associated with information related to the user's hobbies and tastes.
It is also preferable that the utterance generation means generates a content presentation utterance related to the acquired content.

Furthermore, as yet another embodiment of the dialogue control device according to the present invention, the device provides conditions relating to the number of characters and / or conditions relating to whether or not a predetermined character or symbol is included from the acquired related text. Further has a related text selection means for selecting related texts to be satisfied,
It is also preferable that the utterance generation means generates the content-related text utterance from the selected related texts.

Further, as a further embodiment of the dialogue control device according to the present invention, when the related text contains characters or character strings that are continuous more than a predetermined value, the device is such that the characters or character strings are continuous with respect to the related text. It is also preferable to have a related text changing means for making changes to a smaller degree.

Further, as a further embodiment of the dialogue control device according to the present invention, the device discriminates the first-person words included in the acquired related texts, and obtains one type of related texts including a plurality of types of first-person words. It is also preferable to have a related text changing means for changing to one containing the first person word of.

Further, in the dialogue control device according to the present invention, it is also preferable that the related text acquisition means acquires the post related to the content as the related text from the posting site or the posting database including the posting related to the content.

According to the present invention, it is also a program for operating a computer mounted on a device capable of performing dialogue on contents.
Related text acquisition means to acquire related text that is the text related to the content, and
A polarity determining means for determining the polarity of the acquired related text and determining the polarity of the user's utterance when the user's utterance regarding the content is received.
An utterance generation means that receives the polarity determined in the user's utterance regarding the content and generates a content-related text utterance that is the utterance related to the related text from the related text having the polarity determined based on the received polarity. ,
As a response to the user's utterance regarding the content, a dialogue control program for operating the computer as an utterance control means for outputting the content-related text utterance is provided.

According to the present invention, further, it is a dialogue control method implemented in a computer mounted on a device capable of performing dialogue regarding contents.
Steps to get the related text, which is the text about the content,
The steps of determining the polarity of the acquired related text and determining the polarity of the user's utterance when the user's utterance regarding the content is received.
A step of receiving a polarity determined in the user's utterance regarding the content and generating a content-related text utterance which is an utterance related to the related text from the related text having the polarity determined based on the received polarity.
As a response to the user's utterance regarding the content, a dialogue control method including a step of outputting the content-related text utterance is provided.

According to the dialogue control device, the dialogue control program, and the dialogue control method of the present invention, it is possible to carry out a response in which the user's empathy in the dialogue is further improved.

It is a functional block diagram which shows the functional structure in one Embodiment of the dialogue control apparatus by this invention. It is a schematic diagram which shows one Embodiment about the handling of the polarity determination result by the polarity determination part. It is a schematic diagram which shows one Embodiment about the handling of the honorific expression determination result by the honorific expression determination unit. It is a functional block diagram for demonstrating one Embodiment of the score determination process by a score determination part. It is a schematic diagram which shows the example of the utterance input / output when the user interface is a touch panel display. It is a flowchart which shows one Embodiment of the dialogue control method by this invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[Dialogue control device]
FIG. 1 is a functional block diagram showing a functional configuration according to an embodiment of the dialogue control device according to the present invention.

According to FIG. 1, the dialogue control device 1 as an embodiment of the present invention is a device capable of performing dialogue with a user via a user interface (UI) 102 such as a speaker / microphone or a touch panel display. In particular, it is possible to output a response on the device side that further improves the user's sympathy in dialogue.

Moreover, the dialogue carried out in this embodiment is a dialogue concerning some kind of content. Here, the content is data including news articles, commentary articles, critiques, and texts containing some matters such as essays, such as websites and databases acquired via a communication network. Data can be adopted.

Similarly, as shown in FIG. 1, the dialogue control device 1 includes the communication interface 101 in the present embodiment and is installed on the communication network.
(A) A content server 2 capable of distributing and providing various contents (for example, news article data) operated by a digital content distribution company, and
(B) SNS server 3 capable of distributing and providing SNS (Social Networking Service) posted data
You can get content and SNS post data by communicating with.

Incidentally, as will be described in detail later, when the dialogue control device 1 acquires the "related text" which is the text related to the content related to the dialogue and generates a response for further improving the user's empathy, the "related text" is generated. It is characterized by utilizing.

Here, as the "related text", the SNS post acquired from the SNS server 3 in (b) above can be adopted. Of course, other posts, blogs, bulletin board comments and threads obtained from various posting site servers, blog (Web-log) servers, bulletin board servers, review site servers, and even twitter (registered trademark) site servers. , Reviews, and even tweets can be adopted.

In addition to acquiring posts and comments from the distribution / providing server, it is also possible to acquire posts / comments from a pre-installed / prepared post / comment database (DB) and use them as “related texts”. Furthermore, it is also preferable to acquire the content and the "related text" from a site server in which the content and the corresponding "related text" are managed in association with each other and distributed / published, for example, a YouTube (registered trademark) site server. ..

Specifically, the dialogue control device 1
(A) Acquire "related text", which is the text related to the content related to the dialogue.
(B) From the acquired "related text", a "content-related text utterance", which is an utterance related to the "related text", is generated.
(C) It is characterized in that the generated "content-related text utterance" is output as a response to the user's utterance regarding the content.

Here, the "related text" of (B) can be obtained from a posting site (for example, an SNS posting site provided by the SNS server 3) including posts related to the content, as described above. For example, a user who saw a news article (content) containing the title "** Players won a gold medal!" Said their impressions and posted an SNS saying "** I thought players would win medals." Is the "related text" for this news article.

Further, as a modification mode, it is also possible to adopt a “related text” in which a voice comment made by the user on the content is converted into text data. For example, when a commentator makes a voice comment in a news program on television, radio, or the like, the voice data of this comment may be dropped into a text and used as a "related text".

Furthermore, enter a keyword that will be the title name of a news article into the API (Application Programming Interface) of a twitter (registered trademark) site server, etc. to acquire the news article of the news site and correspond to a comment on the news. It is also possible to acquire the tweet to be done as "related text".

In any case, such "related texts" are what various people feel, think, think, and even think about the content that is the topic (or part of it) in the dialogue. Includes what was stated about. The dialogue control device 1 generates a "content-related text utterance" for responding to the utterance of the user who is the conversation destination from the "related text" having such an assertion content, thereby causing the user's impression of the content and the content. It makes it possible to carry out a response that is in line with or in line with the idea, in other words, a response that further improves the user's empathy.

In addition, in a dialogue regarding the content of a specific content, the device side responding to the utterance on the user side so that the user can sympathize with the user, which improves the user's satisfaction and activates the communication using the device. It also leads to letting them do it.

Furthermore, when, for example, an SNS post is adopted as the "related text", it indicates whether the content is usually a large number and (positive or negative) with respect to the content that becomes the topic (or a part thereof) of the dialogue. ) Posts with various contents regarding polarity can be obtained as appropriate. Therefore, it is possible to realize higher empathy by selecting a more suitable post from such posts in consideration of polarity and the like and using it for response generation.

[Device function configuration]
Similarly, according to the functional block diagram of FIG. 1, the dialogue control device 1 includes a communication interface 101, a user interface (UI) 102, a content database (DB) 103, an SNS posting DB 104, a dialogue scenario DB 105, and a processor. Has memory.

Here, the processor memory stores one embodiment of the dialogue control program according to the present invention, has a computer function, and executes the dialogue control process by executing the dialogue control program. To do. Therefore, the dialogue control device 1 can be, for example, a smartphone, a tablet-type or notebook-type computer, a wearable terminal, a personal computer (PC), or the like, which is equipped with the dialogue control program according to the present invention. Incidentally, this dialogue control program may be a dialogue application distributed from a predetermined application distribution server when, for example, the device 1 is a terminal such as a smartphone.

Further, the processor memory includes a user information acquisition unit 111, a content acquisition unit 112, a post acquisition unit 113 as a related text acquisition means, a post selection unit 114 as a related text selection means, and the like as a functional configuration unit. Post change unit 115 as related text changing means, polarity determination unit 116, honorable word expression determination unit 117, score determination unit 118, content presentation utterance generation unit 121a, and (as utterance generation means related to related text). It has an utterance generation unit 121 including a post utterance generation unit 121b, an utterance control unit 122, an input / output control unit 123, and a communication control unit 131.

It should be noted that these functional components can be regarded as the functions of the dialogue control program stored in the processor memory, and the processes shown by connecting the functional components of the dialogue control device 1 in FIG. 1 with arrows. Is also understood as an embodiment of the dialogue control method according to the present invention.

Similarly, in FIG. 1, the user information acquisition unit 111 describes information related to the user's hobbies and tastes from the profile information registered in a predetermined SNS or from the information related to the SNS posting by the user. To acquire or decide. The information related to the hobbies and tastes is used in the content acquisition unit 112, which will be described later, when acquiring the content that is the topic of dialogue.

Specifically, the user information acquisition unit 111, for example, from a server related to a predetermined SNS (for example, SNS server 3), the text data of the profile information of the user registered by the user and the text data of the post posted by the user. And, for example, using a predetermined API provided by the server or by a browser.

Next, the user information acquisition unit 111 combines these text data into one document data, and then extracts the feature degree from this document by using, for example, a TF-IDF (Term Frequency --Inverse Document Frequency) method. From the words having a high tf-idf score, for example, a dictionary of words related to hobbies and tastes may be used to select information related to the user's hobbies and tastes.

The content acquisition unit 112 acquires content that is a topic of dialogue from the content server 2 via the communication interface 101 and the communication control unit 131. Alternatively, the content may be acquired from the content DB 103 that stores and manages the content acquired in advance from the outside. Here, for example, the content acquisition unit 112 may acquire content suitable as a topic by using a predetermined keyword related to the topic as a search key and using the search API provided by the content server 2.

Further, as one preferred embodiment, it is also preferable that the content acquisition unit 112 acquires the content related to or associated with the information related to the user's hobbies and tastes acquired by the user information acquisition unit 111. Specifically, when the content includes text data composed of a title, a summary, a text, etc., the word representing the acquired information related to the user's hobbies and tastes is used as a search key, and the word is added to a predetermined text data portion. (For example, a predetermined frequency or more) may be selected and acquired. For example, when the content is a news article or public movie information and the user's hobby / preference is "automobile", a search is performed using "automobile" or a related word "passenger car" as a search key, and the article or the relevant word is concerned. You can get one piece of information that includes these search keys in the title.

By outputting the content presentation utterance based on the selected content as described above, it is more likely that the user actively participates in the dialogue and makes an utterance expressing his / her feelings. It is possible to create a situation that can increase empathy in dialogue.

Further, as a modification mode, for example, when a keyword characterizing the content is given in advance for each content stored in the content DB 103, the content acquisition unit 112 matches or is related to the information related to the user's hobbies and tastes. It is also possible to acquire the content to which the keyword is added from the content DB 103.

Similarly, in FIG. 1, the content presentation utterance generation unit 121a of the utterance generation unit 121 generates a “content presentation utterance” which is an utterance that presents the content (content) from the content acquired by the content acquisition unit 112. Specifically, for example, although it is a simple method,
(A) A title or summary in the text data of the content, or a word (group) extracted from it (for example, using a dictionary in which current terms are registered).
(B) Content presentation utterances may be generated using a predetermined utterance template. For example, if the content is a news article and the title is "** Car is burning on the highway!", The content presentation utterance is "** There is news that the car is burning on the highway!" Can be generated. It is also possible to generate utterances such as "** The highway is being reported."

Next, the generated content presentation utterance is output to the user via the utterance control unit 122, the input / output control unit 123, and the UI 102, which will be described in detail later. Here, for example, when the UI 102 is a touch panel display, the content presentation utterance may be converted into display display data by the input / output control unit 123 and displayed as an utterance sentence on the display. When the UI 102 is a speaker / microphone, the input / output control unit 123 can convert it into voice data and output it as voice from the speaker.

Further, the utterance control unit 122 outputs the content presentation utterance as described above, and then receives the utterance on the user side regarding the content, which is made in response to this output (device side utterance), via the UI 102. At this time, when the UI 102 is a touch panel display, the utterance on the user side is input as, for example, touch panel input data by the user, converted into text data by the input / output control unit 123, and output to the utterance control unit 122. When the UI 102 is a speaker / microphone, the utterance on the user side is input as voice data via the microphone, converted into text data by the input / output control unit 123, and output to the utterance control unit 122. Become.

Similarly, in FIG. 1, the post acquisition unit 113 is a related text acquisition means for acquiring an SNS post as a “related text”. Here, the "related text" related to the content that is the topic of the dialogue can be acquired by various methods as described above, but in the present embodiment, the post acquisition unit 113 is on the web page of the content site related to the content. The SNS post generated via the SNS link icon of is acquired as the related text. In any case, the post acquisition unit 113 usually acquires a plurality of (often a large number) SNS posts (related texts).

The post selection unit 114 is based on the acquired SNS posts (related texts).
It is a related text selection means for selecting an SNS post (related text) that satisfies either one or both of (a) a condition related to the number of characters and (b) a condition related to whether or not a predetermined character or symbol is included. .. In the present embodiment, the utterance generation unit 121, which will be described later, generates a content-related text utterance from at least one of the SNS posts (related texts) selected here.

Specifically, as an example of the above condition (a), the post selection unit 114 can select SNS posts whose number of constituent characters is within a predetermined threshold (for example, 60 characters) (exclude those exceeding the predetermined threshold). This is a situation in which the amount of content-related text utterance data generated from SNS posts increases after that, and the user needs to read a long sentence or listen to a considerable amount of voice, which reduces the comfort of dialogue. It is a measure to avoid. Here, the character number threshold value may be set to a different value depending on, for example, the attribute information of the user (acquired together with the hobby / preference information) or the type of content.

Further, as an example of the above condition (b), the post selection unit 114 excludes SNS posts containing characters such as "http: //" for indicating a URL (Uniform Resource Locator) (selects those not included). It is also preferable. In addition to the URL, SNS posts containing characters and symbols that make the generated content-related text utterances unsuitable as utterances on the device side, or characters and symbols that are difficult to utter (verbalize) are excluded. You may.

The post change unit 115 describes the acquired SNS post (related text) (or further selected by the post selection unit 114).
(A) When the SNS post (related text) contains characters or character strings that are continuous more than a predetermined value, the SNS post (related text) is changed to reduce the degree of continuity of these characters or character strings. The change processing to be performed, and (b) the SNS post (related text) containing a plurality of types of first-person words by discriminating the first-person words included in the SNS post (related text), including one type of first-person word A related text-changing means that implements one or both of the change processes to change to one.

Specifically, as an example of the above condition (a), the post change unit 115 is a portion (for example, "!!!!!") in which characters or character strings appear continuously by a number exceeding a predetermined threshold value (for example, three) in an SNS post. If "!!!") exists, it is also preferable to change this continuous part to a continuous one (for example, "!!!!!") by three. As a result, unnecessary display and unnecessary audio output can be suppressed in advancing the dialogue.

Further, as the above condition (b), if there are a plurality of different first-person words such as "I" and "I" in the SNS post, for example, the post change unit 115 uses one of those first-person words. It is also preferable to carry out a change process that replaces the first person word, for example, "I". This makes it possible to make the utterances on the device side consistent. For example, after the device side utterance "I was surprised about the matter," the device side utterance "I will try it in the future." Was output, and the other party changed. It is possible to avoid a situation in which the user feels such unnaturalness.

Similarly, in FIG. 1, the polarity determination unit 116 is
(A) The utterance on the user side regarding the content output from the utterance control unit 122, and
(B) The "polarity" of each of the acquired (or further selected by the post selection unit 114 and passed through the post change unit 115) SNS post (related text) is determined.

Here, in the present embodiment, three classification models of positive (Pos), negative (Neg), and neutral (Neu) are adopted as the “polarity”. Of these, positive SNS posts have positive content for the content, negative SNS posts have negative content for the content, and neutral SNS posts have content that is neither of them. It becomes a thing. Naturally, another index may be adopted as the polarity. For example, Ekman's 7 classification model and other known emotion / facial expression category models can be adopted.

In any case, as will be described in detail later, the utterance generation unit 121 transmits the content-related text utterance from the SNS post (related text) having the polarity determined based on the polarity of the user's utterance regarding the content. It will generate.

Here, the polarity determination in the user-side utterance (a) and the SNS posting in (b) above is performed.
For example, it is also preferable to use a determination device such as SVM (Support Vector Machine). In this case, a large number of pairs of text data uttered by the user or posted on the SNS and polarity as the correct answer to the data are prepared, the discriminator is trained in advance with these pairs, and the learned discriminator is used for polarity. The judgment will be made. Alternatively, as another method, it is possible to perform morphological analysis on the text data and use a dictionary containing the polarity of the word to determine the polarity of the entire data from the polarity of the words constituting the text data. is there.

FIG. 2 is a schematic view showing an embodiment regarding the handling of the polarity determination result by the polarity determination unit 116.

According to FIG. 2, the SNS posting DB 104 (FIG. 1) inputs the result of the polarity determination for each SNS posting by the polarity determination unit 116 from the polarity determination unit 116, and records the result in association with the SNS posting. More specifically, the SNS post DB 104 stores the SNS posts (text data) related to the contents (information) in association with each content (information), and each of these SNS posts is stored. The polarity determination results are linked and recorded (marking). Here, "Pos", "Neg", and "Neu" in FIG. 2 indicate that the polarities are positive, negative, and neutral, respectively.

As shown in FIG. 2, the polarity determination in the SNS posting in (b) above is a plurality of (many) acquired (or further selected by the post selection unit 114 and passed through the post change unit 115). It is also preferable to carry out the SNS posting in advance. Of course, when a response is generated in response to an utterance on the user side, it is also possible to determine the polarity of the SNS post, which is a candidate for generating the response utterance, in real time.

However, by performing the polarity determination for a plurality of (many) SNS posts in advance, when the user side utterance is received, the SNS utterance for generating the response utterance is immediately selected, and the response utterance is output in a shorter time. Therefore, it becomes possible to realize a smoother dialogue. Incidentally, in this case, it is also preferable that the utterance control unit 122 outputs the content presentation utterance to the user after the polarity determination process by the polarity determination unit 116 is completed and the determination result is registered in the SNS posting DB 104.

Returning to FIG. 1, the honorific expression determination unit 117
(A) The utterance on the user side regarding the content output from the utterance control unit 122, and
(B) Whether the honorific expression is included in each of the acquired SNS posts (related texts) that have been acquired (or further selected by the post selection unit 114, passed through the post change unit 115, or have undergone polarity determination processing). Judge whether or not.

Incidentally, as will be described in detail later, the utterance generation unit 121 generates the content-related text utterance from the SNS post (related text) having the same judgment result as the honorific expression judgment result in the user's utterance regarding the content. ..

For example, when the user side speaks in honorific expressions in a dialogue, the device side outputs the utterances in a normal expression other than the honorific expressions, which makes the dialogue unnatural. On the contrary, outputting honorific expressions to ordinary expressions also brings about an unnatural dialogue. Therefore, in order to have a natural dialogue, if the user side utters the honorific expression / normal expression, the device side should also output the honorific expression / normal expression utterance accordingly. In this respect, by generating the device-side utterance using the honorific expression determination result as described above, it is possible to realize a natural dialogue that does not make the user feel uncomfortable.

The honorific expression determination unit 117 specifically determines whether or not the expression in the sentence (particularly at the end of the sentence) constituting the user-side utterance in (a) and the text of the SNS post in (b) above is an honorific expression. ..

Here, for example, the text of the text is decomposed into morphemes using MeCab, which is an open source morphological analysis engine, and used as honorific expressions such as morphemes whose part of speech is an auxiliary verb and whose prototype is "desu" or "masu". It is also preferable to determine whether or not a morpheme satisfying the predetermined condition is contained. In this case, if such a morpheme is included, the content or SNS post is determined to contain the honorific expression (respect), while if it is not included, the honorific expression is not included (normal: normal expression). It will be judged.

FIG. 3 is a schematic diagram showing an embodiment regarding the handling of the honorific expression determination result by the honorific expression determination unit 117.

According to FIG. 3, the SNS posting DB 104 (FIG. 1) inputs the result of the honorific expression determination for each SNS post by the honorific expression determination unit 117 from the honorific expression determination unit 117, and records the result in association with the SNS post. ing.

More specifically, the SNS post DB 104 is an SNS post related to the content (information) for each content (information), and the SNS post (text data) to which the above-mentioned polarity determination results are associated with each is displayed. It is saved in association with each other. The SNS post DB 104 further records (marks) each of these SNS posts in association with the result of the honorific expression determination for the SNS post. Here, "Kei" in FIG. 3 indicates that the SNS post includes the honorific expression, while "Pair" indicates that the SNS post does not include the honorific expression.

As shown in FIG. 3, the honorific expression determination in the SNS posting described in (b) above is obtained (or further selected by the post selection unit 114 and passed through the post change unit 115, or further subjected to polarity determination processing. It is also preferable to carry out in advance for a plurality of (many) SNS posts (received). Of course, when a response is generated in response to a user's utterance, it is also possible to determine the honorific expression of the SNS post, which is a candidate for generating the response utterance, in real time.

However, by performing the honorific expression determination for a plurality of (many) SNS posts in advance, when the user side utterance is received, the SNS utterance for generating the response utterance is immediately selected and the response utterance is output in a shorter time. Therefore, it becomes possible to realize a smoother dialogue. By the way, also in this case, after the honorific expression determination process by the honorific expression determination unit 117 is completed and the determination result is registered in the SNS posting DB 104, the utterance control unit 122 may output the content presentation utterance to the user. preferable.

Returning to FIG. 1, the score determination unit 118
(A) The utterance on the user side regarding the content output from the utterance control unit 122, and
(B) In each of the acquired SNS posts (related texts) that have been acquired (or further selected by the post selection unit 114, passed through the post change unit 115, or further subjected to the polarity determination process and / or the honorific expression determination process). A word having a predetermined part of speech included is extracted, and a "score", which is the degree of matching of the word between the utterance on the user side and the SNS post (related text), is determined.

FIG. 4 is a functional block diagram for explaining an embodiment of the score determination process by the score determination unit 118.

As shown in FIG. 4, the score determining unit 118 has a user-spoken verb extraction unit 118a, a posted verb extraction unit 118b, and a score giving unit 118c as functional constituent units. Of these, the user-spoken verb extraction unit 118a and the posted verb extraction unit 118b have part of speech that appears in the user-spoken text data of (a) and the SNS-posted text data of (b), respectively. Extract words (score words) that are verbs, adjectives, or adjective verbs.

Next, the scoring unit 118c matches the ratio of all the score words extracted from the SNS post that match any of the score words extracted from the user's utterance (for example, 1 if all match). If there is nothing, a value that becomes 0) is calculated, and the ratio is used as a score and given (marked) to this SNS post.

Here, as a modification mode, it is also possible to use a preset synonym dictionary and set the ratio of score words that match or are similar to any of the score words extracted from the user's utterance as the score. In any case, it is preferable that the score for the SNS post is given and registered for each of the SNS posts stored in the SNS post DB 104, as in the case of the polarity in FIG. 2 and the honorific expression in FIG.

By the way, verbs, adjectives and adjectives are set as predetermined parts of speech for score calculation in the text of ordinary conversation, and in many cases, the part of speech of words related to the speaker's emotions is the verb, adjective or adjective. Because it is a verb. The empathy held by the user is the degree of sensation of sharing emotions with the dialogue destination (device 1), and these part of speech that are likely to contain emotional expressions in the user's utterance. It is possible to further enhance the user's empathy by responding to the word in a manner that the device side repeats or responds to.

However, in some cases, for example, noun words that point to objects related to the speaker's emotions (such as "dog" and "cat") may be repeated in the utterance on the device side, increasing the empathy of the dialogue. obtain. Therefore, it is also preferable to extract a score word by adding a noun to a verb, an adjective, and an adjective verb as a predetermined part of speech for score calculation, and calculate the score.

Next, after the scoring process described above, the utterance generation unit 121 utters the content-related text from the SNS post (related text) showing the highest or higher score with the utterance on the user side regarding the content. To generate. Here, as shown in FIG. 4, the score determination unit 118 has a score selection unit 118d, and the score selection unit 118d selects an SNS post to which the highest score or a predetermined score or higher is given and the utterance generation unit 121. You may hand it to. Of course, it is also preferable that the utterance generation unit 121 makes a selection based on the score.

Returning to FIG. 1, the posted utterance generation unit 121b of the utterance generation unit 121 has been acquired (or further selected by the post selection unit 114 and passed through the post change unit 115, or further polarity determination processing, honorific expression determination processing, and / Or, from the SNS post (which has undergone the score-giving process), a "content-related post utterance" (content-related text utterance), which is an utterance related to the SNS post, is generated.

Here, the post utterance generation unit 121b can use the text data of the text of the SNS post as it is as the content-related post utterance, but uses the text data of the text of the SNS post and a predetermined utterance template. , Content-related post utterances may be generated.

For example, from an SNS post saying "** players thought they would win medals", it is possible to generate content-related post utterances such as "Yes! ** I thought players would win medals." Regarding the above-mentioned predetermined utterance template, for example, a plurality of types are prepared in advance in association with the presence or absence of polarity and honorific expression, and a suitable one is selected according to the polarity and honorific expression determination result given to the user's utterance. It is also preferable to select.

Specifically, the posted utterance generation unit 121b is the content (information) that has become a hot topic in the user's utterance from the SNS post DB 104 in which the polarity determination result, the honorific expression determination result, and the score are linked and saved for each SNS post. ) Is an SNS post associated with
(A) A polarity that matches the polarity of the user's utterance is given, and (b) a honorific expression determination result that matches the honorific expression determination result (for example, "respect") of the user's utterance (same "respect"). ) Is added to the search for SNS posts.

For example, in the embodiment shown in FIG. 3, assuming that the dialogue related to the content information 1 is performed and the polarity of the user-side utterance is positive (Pos), SNS posts 1 to 16 are searched from the SNS post DB 104. Will be.

Next, the post utterance generation unit 121b is among the searched SNS posts.
(C) Select one SNS post that has the highest score or a score equal to or higher than a predetermined threshold. Here, when a plurality of scores can be selected, for example, one may be randomly determined. The post utterance generation unit 121b generates content-related post utterances from the SNS posts selected in this way.

In the embodiment described above, the SNS posting satisfying all of the above conditions (a), (b) and (c) is selected as a suitable embodiment, but the above conditions (a), (b) and It is also possible to select one that satisfies any one or two of (c). It is also possible to first search for SNS posts with a score equal to or higher than the predetermined threshold under the above condition (c), and then narrow down to SNS posts that satisfy the above conditions (a) and / or (b).

Further, in the above condition (a), polarity matching is a selection condition as a preferable form, but as a change mode, a polarity determined in advance according to the polarity of the user's utterance (for example, pos / neg) On the other hand, it is also possible to select an SNS post having neg / pos and neu for neu). Alternatively, SNS posts having a predetermined polarity (for example, pos) may be selected regardless of the polarity of the user's utterance.

The utterance generation unit 121 including the posted utterance generation unit 121b described above reads out the utterance text data stored in the dialogue scenario DB 105 (or the dialogue scenario DB 4 installed outside the device 1) (or obtains it by communication). It is also preferable to generate appropriate utterances in response to continuous dialogue with the user. This dialogue scenario DB 105 is a database that stores utterance text data groups that constitute a dialogue between the assumed device 1 and the user.

Similarly, in FIG. 1, the utterance control unit 122 outputs the generated content-related text utterance to the UI 102 via the input / output control unit 123 as a response to the user-side utterance (related to the content being talked about). At this time, the output timing is controlled so that the content-related text utterance can be presented to the user (for example, screen display and / or voice presentation) at a timing such that a normal human who receives the user-side utterance responds. Is also preferable.

The input / output control unit 123 converts the data related to the user-side utterance input from the user interface 102 into, for example, text data and outputs the data to the utterance control unit 122, while the input / output control unit 123, for example, the text related to the utterance input from the utterance control unit 122. The data is converted into a form that can be output by the user interface 102, and is output to the UI 102 at the timing and order specified by the utterance control unit 122.

For example, when the UI 102 is a touch panel display, the input / output control unit 123 converts the utterance into display display data and outputs the utterance. Further, the user-side utterance input via the touch panel can be converted into, for example, text data and output to the utterance control unit 122. Further, when the UI 102 is a speaker / microphone, the input / output control unit 123 converts the utterance into voice data by using a known voice data conversion means and outputs the utterance. Further, in this case, the voice data of the user-side utterance input via the microphone may be converted into, for example, text data by using a known voice recognition means and output to the utterance control unit 122.

FIG. 5 is a schematic diagram showing an example of utterance input / output when the UI 102 is a touch panel display.

According to FIG. 5, the “content presentation utterance” on the device 1 side is displayed on the display screen of the UI 102 in the form of a balloon from the left end, and then presented in the form of a balloon from the right end below it. "User utterance" related to the content is displayed. Further below that,
"Content-related text utterance" on the device 1 side is displayed in the form of a balloon from the left end. In this way, the utterances exchanged between the device 1 and the user are displayed alternately from the top, and the state of the dialogue can be understood at a glance.

Of these, it is also preferable that the input of the utterance on the user side is performed by using a touch panel, for example, by inputting characters into the input form provided at the bottom of the display screen of the UI 102 of FIG. 5 and tapping the transmission icon. After this input, "user-side utterance" is immediately displayed on the screen. As another method, the input of the user's utterance may be a voice input by the user via a microphone provided in the UI 102.

Further, the utterance on the device 1 side may be output as voice by using the speaker provided in the UI 102 together with or without the display display. Of course, both the input and output of the utterance may be only the voice input and output. Incidentally, for example, when the dialogue control device 1 is a smartphone that has downloaded the dialogue application, which is the dialogue control program according to the present invention, it is possible to realize the input / output described above as a function of this application.

As another embodiment of the UI 102, one provided with a keyboard or one provided with a Braille input / output device may be adopted, and the input / output control unit 123 may be configured accordingly.

[Dialogue control method]
FIG. 6 is a flowchart schematically showing an embodiment of the dialogue control method according to the present invention. In this embodiment, news articles and SNS posts are adopted as the contents and related texts, respectively.

(S101) Acquire a news article based on user information.
(S102) Acquire a plurality of SNS posts linked to the acquired news article (for example, linked with a link icon).
(S103) From the acquired plurality of SNS posts, select an SNS post that satisfies a predetermined condition related to the number of characters and the like.

(S104) Among the selected SNS posts, a change process for reducing the degree of continuity is performed on the SNS posts containing a predetermined number of consecutive characters or more.
(S105) Among the SNS posts subjected to the above-mentioned change processing, a process of changing an SNS post containing a plurality of types of first-person words to one containing one type of first-person word is performed.
(S106) For each SNS post, a determination relating to polarity and honorific expressions is performed, and verbs, adjectives, and adjective verbs are extracted.

(S107) A news article presentation utterance (content presentation utterance) is generated from the acquired news article and output to the user via the UI 102.
(S108) The user-side utterance is input via the UI 102.

(S109) The polarity of the input user-side utterance is determined, and an SNS post having the same polarity as the polarity is selected from the SNS posts prepared in the step S106.
(S110) Performs honorific expression determination for the input user-side utterance, and further selects an SNS post having the same honorific expression determination result as the honorific expression determination result in this user-side utterance from the selected SNS posts. ..

(S111) The verbs, adjectives, and adjective verbs included in the input user-side utterance are extracted, and for each selected SNS post, the extracted verbs, adjectives, and adjective verbs are combined with the user-side utterance. Make a comparison and calculate the score.
(S112) Based on the calculated score, the SNS post for generating the news article-related post utterance (content-related post utterance) is determined from the selected SNS posts.
(S113) From the determined SNS post, a news article-related post utterance is generated and output to the user via the UI 102.

As described above, in steps S101 to S113, one dialogue flow is carried out in which news article presentation utterances are generated and output, user-side utterances are input, and then news article-related utterances are generated and output. This dialogue flow may be repeated as appropriate in a series of dialogues.

Further, the flow of steps S107 to S113 for inputting the utterance on the user side and generating and outputting the utterance related to the news article may be appropriately repeated in a series of dialogues on the content of a certain content. Further, as a modification mode, for example, as a response based on an SNS post in which another content information is found in the user's utterance and registered in advance in association with the content information, or an SNS post searched in response to the content information. It is also possible to output the content-related post utterances of.

As described in detail above, the "related text" according to the present invention refers to the feelings, thoughts, and thoughts of various human beings with respect to the content that is the topic (or a part thereof) in the dialogue. It also includes content that expresses what you think. According to the present invention, by generating a device-side utterance as a response to a user-side utterance about the content from a related text having such an assertion content, the user's impressions and thoughts about the content are met or followed. In other words, it is possible to carry out a response in which the user's empathy is further improved.

In addition, in a dialogue regarding the content of a specific content, the device side responding to the utterance on the user side so that the user can sympathize with the user, which improves the user's satisfaction and activates the communication using the device. It also leads to letting.

With respect to the various embodiments of the present invention described above, various changes, modifications and omissions within the scope of the technical idea and viewpoint of the present invention can be easily made by those skilled in the art. The above explanation is just an example and does not attempt to restrict anything. The present invention is limited only to the scope of claims and their equivalents.

1 Dialogue control device 101 Communication interface 102 User interface (UI)
103 Content database (DB)
104 SNS Post DB
105 Dialogue scenario DB
111 User information acquisition unit 112 Content acquisition unit 113 Post acquisition unit (related text acquisition means)
114 Post Sorting Department (Related Text Sorting Means)
115 Post change section (related text change means)
116 Polarity judgment unit 117 Respectful expression judgment unit 118 Score determination unit 118a User utterance verb extraction unit 118b Post verb extraction unit 118c Score assignment unit 118d Score selection unit 121 Utterance generation unit 121a Content presentation utterance generation unit 121b Post utterance generation unit ( Content-related text utterance generation means)
122 Speech control unit 123 Input / output control unit 131 Communication control unit 2 Content server 3 SNS server 4 Dialogue scenario DB

Claims (12)

A dialogue control device that can perform dialogues related to content.
Related text acquisition means to acquire related text that is the text related to the content, and
A polarity determination means for determining the polarity of the acquired related text and determining the polarity of the user's utterance when the user's utterance regarding the content is received.
An utterance generation means that receives the polarity determined in the user's utterance regarding the content and generates the content-related text utterance that is the utterance related to the related text from the related text having the polarity determined based on the received polarity. ,
A dialogue control device comprising: an utterance control means for outputting a text utterance related to the content as a response to the utterance on the user side regarding the content. The dialogue control device according to claim 1 , wherein the utterance generation means generates the content-related text utterance from the related text having the same polarity as the polarity determined in the user's utterance regarding the content. A word with a predetermined part of speech included in the user's utterance related to the content and the acquired related text is extracted, and the score, which is the degree of matching of the word between the user's utterance and the related text, is obtained. It also has a means of determining the score to determine,
The utterance generating means according to claim 1 or 2 , wherein the utterance generating means generates the content-related text utterance from the related text showing the highest score or a predetermined score or more with the user's utterance related to the content. The interactive control device described. Further, it has a honorific expression determination means for determining whether or not the honorific expression is included in the utterance on the user side regarding the content and the acquired related text.
Any one of claims 1 to 3 , wherein the utterance generation means generates the content-related text utterance from the related text having the same determination result as the determination result in the user's utterance regarding the content. The interactive control device according to. The utterance generation means generates a content presentation utterance, which is a utterance that presents the content, from a title or a summary in the data of the content.
From claim 1, the utterance control means outputs the content presentation utterance, and outputs the content-related text utterance as a response to the user's utterance regarding the content, which is performed after the output. The dialogue control device according to any one of 4. A user information acquisition means for acquiring or determining information on a user's hobbies and tastes from a user's profile information or posting information on a predetermined SNS (Social Networking Service).
It further has a content acquisition means for acquiring content related to or associated with information related to the user's hobbies and tastes.
The dialogue control device according to claim 5 , wherein the utterance generation means generates a content presentation utterance related to the acquired content. Further having a related text selection means for selecting related texts that satisfy the conditions related to the number of characters and / or the conditions related to whether or not a predetermined character or symbol is included from the acquired related texts.
The dialogue control device according to any one of claims 1 to 6 , wherein the utterance generation means generates the content-related text utterance from the selected related text. When the related text contains characters or character strings that are continuous more than a predetermined value, the claim is characterized by further having a related text changing means for changing the related text to reduce the degree of continuity of the characters or character strings. Item 2. The dialogue control device according to any one of Items 1 to 7. It is characterized by having a related text changing means for discriminating the first-person word contained in the acquired related text and changing the related text containing a plurality of types of first-person words to one containing one type of first-person word. The dialogue control device according to any one of claims 1 to 8. The related text acquisition means according to any one of claims 1 to 9 , characterized in that, as the related text, a post relating to the content is acquired from a posting site or a posting database including the posting relating to the content. The interactive control device described. A program that allows a computer installed in a device that can engage in content dialogue to function.
Related text acquisition means to acquire related text that is the text related to the content, and
A polarity determination means for determining the polarity of the acquired related text and determining the polarity of the user's utterance when the user's utterance regarding the content is received.
An utterance generation means that receives the polarity determined in the user's utterance regarding the content and generates the content-related text utterance that is the utterance related to the related text from the related text having the polarity determined based on the received polarity. ,
A dialogue control program characterized in that a computer functions as an utterance control means for outputting a text utterance related to the content as a response to the utterance on the user side regarding the content. A dialogue control method implemented on a computer mounted on a device capable of performing dialogue on content.
Steps to get the related text, which is the text about the content,
A step of determining the polarity of the acquired related text and determining the polarity of the user's utterance when the user's utterance regarding the content is received.
A step of receiving a polarity determined in the user's utterance regarding the content and generating a content-related text utterance which is an utterance related to the related text from the related text having the polarity determined based on the received polarity.
A dialogue control method comprising a step of outputting a text utterance related to the content as a response to a utterance on the user side regarding the content.

Images