JP6831522B2 - Communication system - Google Patents

Description

The present invention relates to an interactive system that automatically communicates with a user who has accessed through a telecommunication line such as the Internet.

With the spread of social networking services, it has become possible for a plurality of users to transmit and share information using devices such as smartphones, tablet terminals, and personal computers. As a result, the amount of information that an individual can acquire or transmit has increased, and the scope has expanded significantly. By operating the application software installed on the terminal device, it is possible to easily share information with multiple people at the same time. It can be said that the amount and quality of information is increasing because it is possible to communicate not only by voice and text but also by still images and videos.

Social networking services allow you to overcome time and distance obstacles and share large amounts of information with multiple people in real time, while each user has private time. The reality is that it is not always possible to exchange information in real time because it is normal. In addition, users of social networking services can simultaneously send information to multiple people associated as friends, but it is difficult to communicate individually and simultaneously with each of the multiple friends. is there.

By the way, there is disclosed an automatic interactive system in which document data of answers to expected questions are stored in a recording medium and the answers are output in response to the user's questions (see Patent Documents 1 and 2). ). However, in these automatic interactive systems, since the response documents are prepared in advance, the dialogue is uniform and it is not possible to have a conversation according to the ever-changing social situation. In addition, there is a problem that the individuality and emotions of the person asking the question cannot be expressed.

Japanese Unexamined Patent Publication No. 2006-07278 Japanese Unexamined Patent Publication No. 2006-172280

In view of such a problem, one embodiment of the present invention aims to provide a communication system and a communication method capable of expressing the individuality and personality of an individual user.

A communication system that generates a virtual first user on a computer that reflects the individuality of the first user according to an embodiment of the present invention has an information processing module that generates answers to questions. Artificial intelligence is a first layer in which the answer to a question is inferred and learned by the first social data input by the first user to the social networking service, and a plurality of users input to the social networking service. The answer to the question is inferred by the second layer that infers and learns the answer to the question by the second social data, and the third social data generated by multiple text data collected from the social networking service. Includes a third hierarchy to learn. The first layer, the second layer, and the third layer are layered from the top to the bottom in this order, and the artificial intelligence is a virtual that reflects the individuality of the first user by the first social data. First user is generated, and an appropriate answer to the question of the second user is obtained and output in order from the upper layer among the first layer, the second layer, and the third layer. ..

The first social data and the second social data may be accumulated over time.

A trend analysis of the first social data may be performed and a second social data update may be performed.

The first social data and the second social data may be text data and text data generated from any one or more of audio data, photographic data, and video data.

Artificial intelligence may parse the text data contained in the first social data and infer or learn the answers to the questions to determine.

It may have a function of setting whether or not to generate an answer by artificial intelligence for the question by the second use.

An image data generation module that generates three-dimensional image data of the first user may be included.

A phoneme data generation module that generates phoneme data from the voice data of the first user and a voice generation module that generates speech voice using the phoneme data may be included.

The first social data may be stored in the first database, the second social data may be stored in the second database, and the third social data may be stored in the third database.

The first tier may be linked to the first database, the second tier may be linked to the second database, and the third tier may be linked to the third database.

The communication system may be placed in a state in which it can communicate with a plurality of users in both directions through a telecommunication line.

According to one embodiment of the present invention, artificial intelligence (AI) learns the user's thoughts by acquiring data from various social networking services used by the user and storing it in a database. , Can provide a system that provides appropriate answers to questions.

It is a figure which shows the relationship between the communication providing system which concerns on one Embodiment of this invention, a social networking service, and the user who uses this communication providing system. It is an example diagram of the screen for registering in the communication providing system which concerns on one Embodiment of this invention. It is an example diagram of the screen for registering in the communication providing system which concerns on one Embodiment of this invention. It is an example diagram of the screen for registering in the communication providing system which concerns on one Embodiment of this invention. (D) is an example diagram of a screen for registering in the communication providing system according to the embodiment of the present invention. It is an example diagram of the screen for registering in the communication providing system which concerns on one Embodiment of this invention. It is a figure which shows the relationship between the communication providing system which concerns on one Embodiment of this invention, a social networking service, and the user who uses this communication providing system. It is an example figure of the screen for asking a question in the communication providing system which concerns on one Embodiment of this invention. It is an example figure of the screen for asking a question in the communication providing system which concerns on one Embodiment of this invention. It is an example diagram of the screen in which the answer is provided in the communication providing system which concerns on one Embodiment of this invention. In the communication providing system according to the embodiment of the present invention, when a virtual first user answers a question asked by a plurality of second users on a terminal of the first user, a plurality of second users It is a figure explaining the function of setting the automatic answer possibility for each user of. In the communication providing system which concerns on one Embodiment of this invention, it is a flowchart explaining the flow of the process of determining the propriety of answer for each use which asked a question, and answering. It is a figure which shows the functional structure of the communication providing system which concerns on this embodiment. It is a figure explaining the structure of the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the structure of the communication providing system which concerns on one Embodiment of this invention. It is a flowchart explaining the flow of processing executed in the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the similar database included in the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the database included in the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the database included in the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the mode of evaluating the answer in the communication providing system which concerns on one Embodiment of this invention. It is a flowchart explaining the flow of processing executed in the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the mode of candidate and editing of the answer to a question in the communication providing system which concerns on one Embodiment of this invention. It is a figure which shows the structure for searching the answer from the past conversation and predicting an appropriate answer when a question is received in the communication providing system which concerns on one Embodiment of this invention. It is a flowchart explaining the flow of processing executed in the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the mode of ranking the answer to a question in the communication providing system which concerns on one Embodiment of this invention. It is a figure explaining the mode of ranking the answer to a question in the communication providing system which concerns on one Embodiment of this invention. In the communication providing system according to the embodiment of the present invention, a mode of integrating contexts by grouping questions and answers in conversation is shown. It is a figure which illustrates what kind of order of a keyword is important in seeking a match in the communication providing system which concerns on one Embodiment of this invention. It is a figure which shows the mode in which a question and the answer are clustered by the similarity of the answer, and each cluster constitutes one group in the communication providing system which concerns on one Embodiment of this invention. In the communication providing system according to the embodiment of the present invention, it is a figure explaining that the vector (keyword, concept, type) constituting each question has a predictor for estimating an equivalent answer vector. It is a figure explaining the structure of the communication providing system which concerns on one Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to drawings and the like. However, the present invention can be implemented in many different modes and is not construed as being limited to the description of the embodiments illustrated below. Further, in order to clarify the explanation, the drawings may schematically represent the width, thickness, shape, etc. of each part as compared with the actual embodiment, but this is just an example, and the interpretation of the present invention is used. It is not limited. Further, in the present specification and each figure, the same elements as those described above with respect to the above-mentioned figures may be designated by the same reference numerals, and detailed description thereof may be omitted as appropriate.

[System overview]
FIG. 1A shows the relationship between the communication providing system according to the embodiment of the present invention, the social networking service (hereinafter, also referred to as “SNS”), and the user who uses the communication providing system. The communication providing system 100 according to the present embodiment has a function of acquiring data from various SNSs used by users and storing the data in a database. The communication providing system 100 further has a function of inferring and learning the user's thoughts (function as artificial intelligence) based on the information stored in the database, and can provide an appropriate answer to the question. The details will be described below.

In FIG. 1A, the communication providing system 100 is placed in a state of being accessible to a plurality of users 202 through a telecommunication line. The plurality of users 202 can directly and indirectly communicate with each other by the service provided by the communication providing system 100.

FIG. 1A shows that a plurality of users 202 are placed in a state in which each user terminal 200 can communicate with both of the communication providing system 100 and the telecommunication line. Here, it is preferable that the plurality of users 202 are a set of users registered in advance in order to receive the service provided through the communication providing system 100. By making the use of the service provided through the communication providing system 100 a registration system, the safety and reliability of communication can be enhanced.

It should be noted that it is not always necessary for all of the plurality of users 202 to be registered in advance in order to receive the service provided through the communication providing system 100. For example, at least one of the plurality of users 202 may be able to use the communication providing system 100 as an anonymous user.

The communication providing system 100 is placed in a state in which it can communicate with one or more of the SNS 300 provided to the public. Examples of the SNS 300 include Twitter (registered trademark), Facebook (registered trademark), and the like, and various communication services provided on the Internet are exemplified. It should be noted that the modes of communication provided by the SNS are various and are not limited to the services exemplified above. In the present embodiment, the SNS 300 linked with the communication providing system 100 may be any as long as the social data of the user is provided, and it is preferable that the SNS 300 can share information among a plurality of users.

In addition, in this specification, "social data" means a character (including an emoticon) transmitted by a user to a specific person (for example, another user registered as a friend) or an unspecified person through some SNS. ), Symbols, sounds, still images, moving images, etc. In addition, "personal social data" refers to the social data of a specific individual among a plurality of users.

The communication providing system 100 acquires social data accumulated in each of at least one, preferably a plurality of SNSs 300. The acquisition of social data performed by the communication providing system 100 may be executed in real time, or may be performed at regular intervals or at regular intervals. The communication providing system 100 stores the acquired social data in a database. In order to increase the quantity and quality of the information stored in the database, it is preferable that the communication providing system 100 acquires social data in cooperation with a plurality of SNS 300s.

Specific examples of the acquisition of social data and the process of accumulating the acquired social data in the database are as follows. First, the first user 204 registers the attributes of the first user 204 in the communication providing system 100 by using a personal computer, a smartphone, or the like. For this purpose, for example, as shown in FIG. 1B, the first user 204 displays the screen shown in "Welcome to personalized AI" using a browser or an application, and presses the "Register" button 1. Push. Then, a screen for inputting the attributes of the first user 204 is displayed. The attributes of the first user 204 include information for specifying from which SNS of the SNS 300 the social data is to be acquired (SNS specific information), and information of the first user 204 in the SNS to acquire the social data. Credentials may be included. For example, as shown in FIG. 1C, the SNS specific information is entered in the fields 2, and the user name and password are entered in the fields 3 and 4 as authentication information. Then, press the "Register" button 5.

The attributes of the first user 204 are stored in a storage device of the communication providing system 100, for example, a secondary storage for storing a database or the like. The communication providing system 100 uses the SNS specific information and the SNS authentication information to access the SNS specified by the SNS specific information, and logs in to the SNS using the authentication information of the first user 204. After that, the communication providing system 100 acquires information about the first user 204 stored in the SNS, that is, social data. Information about the first user 204 includes messages exchanged by the first user 204 with a specific user on the SNS, articles, photos, audio information, and videos posted on the SNS by the first user 204. (Can include audio information) is included.

The communication providing system 100 stores the information about the first user 204 acquired from the SNS in the database. In this case, the communication providing system 100 can directly store the information about the first user 204 acquired from the SNS into a database. Alternatively, the communication providing system 100 may convert the information about the first user 204 into other information and store it in the database according to the attributes and contents of the acquired information.

For example, when information indicating that the first user 204 in the SNS has answered a question from another user is acquired, the question from another user and the first user 204 for the question are obtained. Answers can be associated with each other and stored in the database. In this case, the database is composed of a plurality of areas (subdatabases, tables, etc.), and can include (1) an area where questions are accumulated and (2) an area where answers are accumulated. Therefore, the question from another user is accumulated in the area where the question is accumulated, and the answer of the first user 204 is accumulated in the area where the answer is accumulated. Further, the question from another user and the answer of the first user 204 are associated with each other, and for example, the information for association is stored in the third area. For example, a set of a primary key for a question and a primary key for an answer is stored in a third area.

When all the information about the first user 204 is acquired from the SNS specified by the SNS specific information and is not stored in the database, the communication providing system 100 is the first of the information stored in the database. The ratio of all information about the user 204 may be presented to the first user 204. As a result, the first user 204 can know the degree of information accumulation, and for example, the communication providing system 100 responds to a question from any of the plurality of users 202 by the first user 204. You can decide whether or not to answer on your behalf. Further, by presenting the degree of accumulation of such information to a plurality of users 202, it is possible to predict the appropriateness of the answer to the question.

For example, as shown in FIG. 1D, when the ratio of all the information about the first user 204 to the information stored in the database is displayed as a pie chart 6, and 60% of the information is stored in the database. Is displayed as "60% of you." If the first user 204 is considered to be 60% or more, the "public" button 7 is pressed. As a result, the communication providing system 100 answers questions from other users by using the information stored in the database.

Also, if the first user 204 thinks that 60% is dissatisfied, then all (100%) information is stored in the database, for example, as shown in FIG. 1E, "I have become you." Message is displayed. If the first user 204 is satisfied with this, press the "public" button 8. As a result, the communication providing system 100 answers questions from other users by using the information stored in the database.

The first user 204 may be able to ask the communication providing system 100 some trial questions. After seeing the answers to some attempted questions, you may be able to press the "Publish" buttons 7 and 8.

The privacy of users should be ensured even for social data provided by SNS. The communication providing system 100 according to the present embodiment can reflect the information individually set by the user in each SNS 300. For example, if a user who uses a certain SNS 300 sets his / her personal profile to be private and the message posted on the SNS to be made public, the range of public / private / public is set for each message. In that case, the communication providing system 100 can take over the setting.

Therefore, when the communication providing system 100 acquires the information regarding the first user 204 and stores it in the database, the range of disclosure of the information is also stored in the database. For example, as shown in FIG. 1F, in the case of one-to-one message exchange 301-1 with user 207 in SNS6 (301), the question and answer by this message exchange 301-1 are sent to user 207. Only disclosed for. In other words, if the question asked by user A other than user 207 using the communication providing system 100 appears only in the one-to-one message exchange with user 207 by the first user 204. , The answer to the question will not be provided to user A.

Regarding the answer that the first user 204 answered only in a specific group in SNS6, even if a question is asked by a user other than the user belonging to the specific group, the answer is not provided. For example, as shown in FIG. 1F, in SNS6, when the first user 204 receives a question in group 301-2 and answers, the answer is given to user 208 belonging to group 3010-2. Will be disclosed.

On the other hand, when the first user 204 answers a question in a state where anyone can see it on the SNS, the answer is the communication providing system 100 regardless of which user asks the question. Provided by. For example, when the first user 204 answers the question under the state (301-3) that can be viewed by any user belonging to SNS6 (301), the first user 204 is special. If a similar question is asked by a user 209 who has no relation and belongs to the same SNS6, the answer is provided.

In the present embodiment, it is preferable that the plurality of users 202 to which the service is provided from the communication providing system 100 are users of one or a plurality of SNS 300s at the same time. This is because the communication providing system 100 acquires the social data of each of the plurality of users 202 from the SNS 300 provided to the public. That is, the plurality of users 202 can receive the service through the communication providing system 100 by participating in the community of the SNS 300.

FIG. 1A shows a mode in which the second user 206 interacts with the first user 204 via the communication providing system 100. However, one of the features of the present embodiment is that the first user 204 and the second user 206 do not directly exchange messages, but are virtual generated by the communication providing system 100. It takes the form of a dialogue between a typical first user 204b and an existing second user 206. As a result, even if the same question is asked to the first user 204 multiple times, the communication providing system 100 answers as a virtual first user 204b instead of the first user 204. It can be carried out. As a result, the time and effort of the first user 204 can be reduced. For example, when the first user 204 is a celebrity, even if many users ask the same question, it is possible to prevent the first user 204 from being burdened. Further, even if a special question is asked, if the first user 204 has answered such a question in the past, the first user 204 does not search for the past answer. The communication providing system 100 can answer as a virtual first user 204b. This also reduces the time and effort of the first user 204.

An example of dialogue in this case is as follows. First, the second user 206 accesses the communication providing system 100, logs in if necessary, and then identifies a partner (first user 204b) who wishes to have a dialogue. For example, as shown in FIG. 1G, a screen for asking a question is displayed, and the "Login" button 9 is pressed to log in. Alternatively, when asking a question as an anonymous user, press the "anonymous" button 10. Further, in order to specify the person who wants to have a dialogue, as shown in FIG. 1H, the user name of the person who wants to have a dialogue is input in the field 11. Alternatively, a list of acquaintances in the communication providing system 100 of the second user 206 can be displayed to specify a person who wants to have a dialogue.

Then, the user terminal 200 transmits a question to the other party (first user 204b). For example, as shown in FIG. 1H, by inputting a question in the field 12 and pressing the enter key or the like, the communication providing system 100 accepts the question. The communication providing system 100 that has received the question transmits an appropriate answer to the question to the user terminal 200. For example, if the question "Who are you?" Is entered in the field 12, the answer "I am the president" is displayed in the field 14, as shown in FIG. 1J. Further, at this time, the face photo data 13 of the first user 204 may be displayed. The facial photograph data 13 may be data generated by three-dimensional data, as will be described later.

At this time, the response transmitted by the communication providing system 100 is generated based on the personal social data of the first user 204 as the first example. As a second example, when the communication providing system 100 cannot find an appropriate answer from the personal social data, it refers to the social data of a plurality of users to generate an appropriate answer and responds.

In this case, due to the characteristics of the computer system, the response of the virtual first user 204b to the time required for the second user 206 to ask a question (time for operating the user terminal 200) is extremely fast. It becomes. Therefore, as described above, the plurality of users 202 can simultaneously and frequently interact with the virtual first user 204b.

In the present embodiment, the virtual first user 204b generated based on the personal social data of the first user 204 can be considered to be embodied via the communication providing system 100. From another point of view, it can be considered that the virtual first user 204b exists on the computer program or application program executed by the communication providing system 100. Alternatively, the virtual first user 204b can be regarded as an entity embodied by the hardware resources constituting the communication providing system 100 and the software resources executed on the hardware resources.

FIG. 2 shows a functional configuration of the communication providing system 100 according to the present embodiment. The communication providing system 100 includes an information processing module 104 and a database 102 having a function as artificial intelligence.

In the communication providing system 100 shown in FIG. 2, the database 102 is realized by a memory module or a storage device such as a hard disk, a semiconductor memory, and a magnetic memory, and a central processing unit (CPU) or an arithmetic processing circuit having a function equivalent thereto is realized. The information processing module 104 can be realized by the device realized by. Here, the information processing module can be regarded as a hardware resource or a functional block realized by a hardware resource and a software resource, and may be referred to as an information processing unit or an information processing means. Hereinafter, the same applies to those referred to as modules in the present specification.

Here, artificial intelligence is realized by using hardware resources and software resources for intellectual functions such as inference and judgment, and may be recognized as a concept including a database for storing data as knowledge. .. In addition, artificial intelligence has a learning function, and may have the ability to predict the future from past information (data). Also in this embodiment, the information processing module 104 having a function as artificial intelligence includes at least one having a function as normal artificial intelligence as described above.

The database 102 has at least an area for storing social data. FIG. 2 shows an embodiment in which the database 102 includes a first database 102a and a second database 102b that store social data acquired from the SNS 300. Further, the third database 102c may include social data in the same manner, or words, vocabulary, and fixed phrases used in communication may be stored in advance.

The first database 102a, the second database 102b, and the third database 102c store the contents of the dialogue contained in the social data in association with each other, such as a question and an answer (or a question and an answer). It may also include social data that is unilaterally transmitted, such as personal tweets (tweets).

In general, tweets are often not the answer to a question. In this case, the syntax of the tweet (tweet) is analyzed and decomposed into the subject, object, adverbial phrase representing place, time, and mode, and the subject, object, adverbial phrase representing place, time, and mode, etc. You may generate a question asking for and an answer to it and store information about the first user 204 in the database. For example, suppose the first user 204 tweeted, "My sister signed a contract to buy a car yesterday." At this time, the subject is "my sister", the object is "car" and "contract", and the adverbial phrase representing time is "yesterday". So we generate the question "Who did it?", The answer "My sister", the question "When did you do it?", And the answer "Yesterday." Generate, generate the question "What did you contract to buy?", Generate the answer "It's a car.", Generate the question "What did you do?" And "Buy a car." It is a contract to do. ”Can be generated and each can be stored in the database.

The similar database 106 has similar questions as one group, and the answers to the group are associated and stored. The similar database 106 may be divided into a first similar database 106a, a second similar database 106b, and a third similar database 106c corresponding to the first to third databases described above.

The information processing module 104 having a function as artificial intelligence exerts a function as artificial intelligence in cooperation with a database 102 and a similar database 106, and also has a message (answer information, etc.) editing function and a message (answer information, etc.). It includes an evaluation function, an evaluation result notification function, a phoneme generation function for reproducing the user's voice, a question generation function for generating a new question, and a three-dimensional visualization function for generating a user's three-dimensional image.

The communication providing system 100 is placed in a state in which it can bidirectionally communicate with a plurality of users (user terminals 202a and 202b illustrated in FIG. 2). In other words, the user terminals 202a and 202b can communicate with the communication providing system 100 and the SNS 300, while the communication providing system 100 can communicate with a plurality of user terminals 202a and 202b, and the social data can be transmitted from the SNS 300. It is in an available state.

The communication providing system 100 can generate a virtual first user 204b for the first user 204 on the system, and can have a conversation with a plurality of other users sequentially or simultaneously. is there. In this case, the virtual first user 204b can have an equal conversation with all the accessing users. On the other hand, for the first user 204, among the plurality of second users 202, depending on the degree of intimacy (family, friendship), organizational relationship (whether or not they belong to the same corporation or group), etc. There is a request to limit the scope of answers. That is, the first user 204 may want to reserve an answer to the question in relation to the second user 206.

In the communication providing system 100 according to the present embodiment, whether or not the first user 204 may automatically respond to a plurality of users 202 by the virtual first user 204b. It has a function to set. FIG. 1K shows an example of a screen display of a user terminal operated by the first user 204. In the screen display shown in FIG. 1K, whether or not the first user 204 may automatically answer for each user registered to receive the service provided through the communication providing system 100. Indicates the stage for setting. In FIG. 1K, a display 210 for identifying a user by a user name, a thumbnail image of the user, or the like, and a switch 211 for selecting whether or not to answer are shown. The first user 204 can automatically set whether or not to answer for each user while observing such a screen display. It should be noted that such a setting can be changed in a timely manner, and the first user 204 can change the setting by operating the timely answer selection switch 211.

FIG. 1L shows the operation of the communication providing system 100 when the answerability is individually set for the plurality of second users 202. For example, when a question is asked by the second user 206 (S31), the user is identified and whether or not the answer can be automatically determined (S32). When the second user 206 who asked the question is set as "answerable" by the first user 204, the virtual first user 204b generates an answer (S33), and the second user 204 Answer the user 206 (S34). On the other hand, when the second user 206 who asked the question is set as "cannot answer" by the first user 204 (S32), the virtual first user 204b reserves the answer. Become.

In this way, the communication providing system 100 can provide a function in which the first user 204 individually sets whether or not to respond to the plurality of second users 202. That is, the communication providing system 100 has a function of setting whether or not to generate an answer in the information processing module 104 in response to the question by the second use. Thereby, the range of communication by the virtual first user 204b can be set. As a result, for the first user 204, communication with a specific user can be restricted, and in the communication providing system 100, the virtual first user 204b is a plurality of second users. Since it is not necessary to answer all of 202 and all of the questions, the load on the system can be reduced.

Hereinafter, the configuration for the communication providing system 100 to generate a virtual individual of a user and enable the virtual individual to communicate with another user will be described in detail.

[Database]
FIG. 3A is a diagram illustrating a database 102 configuration in the communication providing system 100 according to the embodiment of the present invention. As shown in FIG. 3A, the database 102 may be constructed by being divided into a plurality of layers. The first database 102a is a database corresponding to each user, and personal social data is stored. The personal social data stored in the first database 102a is also used as basic data when generating virtual user data in the communication providing system 100. Therefore, personal social data and the information generated from it are stored in the database for each user. In addition, as described later, the information generated from personal social data includes vectors obtained by analyzing personal social data by keywords, vectors obtained by analyzing personal social data by concept, and personal social data by type. Includes one or more of the analyzed vectors and one or more of the personal social data analyzed by the group.

The data stored in the first database 102a includes personal social data acquired from a plurality of SNS 300s. For example, information such as a comment posted on an SNS by a user, the content of communication exchanged between the user and another user on the SNS, and a tweet (tweet) of the user is stored. There is. Also, as described above, questions and answers can be generated so that they are stored.

The second database 102b stores social data of a plurality of users. The data stored in the second database 102b includes the social data of each user acquired from the plurality of SNS 300s. The amount of data stored in the second database 102b is larger than that in the first database 102a. The second database 102b is hierarchically located below the first database 102a.

It is possible to change the data stored in the second database 102b based on the change in the social data stored in the first database 102a. For example, when it is detected that the length, style, and content of answers to similar questions change by a large number of users, the data stored in the second database 102b may be changed according to the change. It is possible. That is, as shown in FIG. 3B, the tendency analysis for the first database 102a can be performed and the second database 102b can be updated. As an example of such a case, in response to the question "What kind of person are you?", Many users have traditionally asked their personality, such as "I have a short temper." In contrast to the response, recently, when a change that responds to the occupation is detected, such as "I am an office worker.", A second database responds to this change. It is also possible to change the data stored in 102b so that the occupation is answered to the question "What kind of person are you?". In addition, such a change may be referred to as "trend analysis of the answer to the question".

The third database 102c stores a plurality of text data. The data stored in the third database 102c may be arbitrary text data set in advance, or may be a collection of frequently appearing texts among the texts appearing on the SNS. .. For example, it may be a collection of frequently appearing texts from real-time communication called chat exchanged on SNS.

Specifically, the social data of the first user stored in the first database 102a is stored in data in real-time communication called chat, tweets uploaded to Twitter (registered trademark), Facebook (registered trademark), etc. Uploaded comments and comments of other users on the comments are stored. Further, the social data of the plurality of users stored in the second database 102b may include the same data as described above, and may be stored in association with the text data of the question and the answer to the question. Further, also in the third database 102c, a plurality of text data may be stored in association with each other in the form of a question and an answer to the question.

The social data stored in the first database 102a and the second database 102b is preferably text data. However, this text data may include not only text data created by an SNS user but also text data generated from audio data, photographic data, and video data.

For example, the audio data included in the audio data and the video data can be stored in the database as text data by converting the audio into text. In addition, when a user updates a photo or video to SNS, a comment may be added at the same time, and another user's comment may be added to the comment. In addition, the shooting location and its contents may be added as tags to photos, moving images, and sounds. Therefore, when generating text data from photo data and video data, data such as tags and comments attached to the text data can be converted into text data and stored in a database. In addition, the location information (country name, place name, etc.) and date information from which the photo data and the video data have been acquired can be stored in the database as text data. In other words, text data can be generated and stored in a database based on any one or more of audio data, photographic data and video data.

Such a database 102 operates in cooperation with or in cooperation with the information processing module 104 having a function as artificial intelligence, as described with reference to FIG. Since the first database 102a stores the user's personal social data, the one using this hierarchy can also be expressed as "private AI" or "personalized AI" as artificial intelligence that reflects the individual. it can. Further, since the second database stores social data of all users, what is realized by using this hierarchy can be expressed as "Everyone AI" or "Common sense AI".

[Information processing module (artificial intelligence)]
The information processing module 104 having a function as artificial intelligence generates a virtual individual by using the social data stored in the database 102, and communicates with a real user. The information processing module 104 having a function as artificial intelligence has, for example, a function of recognizing text data stored in a database 102 and creating or generating an answer to a question. At this time, the information processing module 104 having a function as artificial intelligence has a function of inferring or learning an appropriate answer to a question from the text data stored in the database 102 and determining the answer.

The information processing module 104 having a function as artificial intelligence receives a question of a second user 206 for a virtual first user 204b, and creates or generates an answer to the question. At this time, the information processing module 104 having a function as artificial intelligence performs a process of requesting an answer in order from the upper layer of the database 102.

FIG. 4 is a flowchart showing an example of a process in which the information processing module 104 having a function as artificial intelligence searches the database 102 and creates an answer to a question. Hereinafter, the processing flow will be described with reference to FIG.

First, when an information processing module 104 having a function as artificial intelligence creates or generates an answer to a question, it first searches the first database 102a for a suitable answer to the question (S01). If an answer is obtained from the first database 102a, it will most accurately reflect the thoughts, thoughts, feelings, etc. of the first user 204 (S06).

When searching for a suitable answer to the question from the first database 102a, among the questions stored in the first database 102a, the question that matches the question as a character string is searched and searched. You can search for the answer associated with the question asked. Alternatively, among the questions stored in the first database 102a, even if the question does not exactly match the question as a character string, it is determined whether or not the question matches in consideration of some fluctuation in the notation, and a search is performed. You can also. It is also possible to search for the question having the largest number of words contained in the question among the questions stored in the first database 102a. In this case, in addition to the number of words, the search can be performed in consideration of the order in which the words are arranged. It is also possible to analyze the meanings of the questions stored in the first database 102a and the questions, and search for the one with the closest meaning. The analysis of meaning can include, for example, deriving a logical conclusion from the answer as a conclusion or deriving the conditions that the answer presupposes based on a predetermined logical system. In this way, when detecting an answer, if it includes a case where the character string does not exactly match, there is a case where "the answer is inferred and determined". In addition, when there is an exact match as a character string, there is a case of "learning and deciding the answer".

When the information processing module 104 having a function as artificial intelligence cannot obtain an answer from the first database 102a (S02), the information processing module 104 searches the second database 102b in the lower hierarchy (S03). The answer obtained from the second database 102b can show the tendency of how the majority of users answer a specific question, and can obtain an average and reasonable answer. For example
"who are you"
In response to the question
"I am XX (name)."
When many answers are given, the information processing module 104 having a function as artificial intelligence can determine a similar answer as an answer to the question. The above is an example of a simple question, but based on the second database 102b, the answer tendency of a large number of users is reflected, so that the answer can be obtained in line with the current trend (S06).

When the information processing module 104 having a function as artificial intelligence cannot find an answer in the second database 102b (S04), the information processing module 104 searches the third database 102c (S05). Since a huge amount of text data is stored in the third database 102c, an answer can be selected from the stored text data. Then, the selected one is used as the answer (S06).

[Judgment of similar questions]
The database 102 stores the question and the answer to the question in association with each other. However, questions (questions) in communication in human society and answers (answers) to them are usually not uniform. For example, when trying to ask the name of a person you meet for the first time, you may ask "Who are you?" Or "Please tell me your name."

In one embodiment of the present invention, when such a similar question is asked, a mechanism for selecting a response to the corresponding question is provided. For example, in response to the above-mentioned "Who are you?" And "Please tell me your name", there is no sense of discomfort even if you receive the answer "I am Annie." This mechanism may be constructed as a similar database 106 as described in FIG.

FIG. 5 shows an example of a similar database 106. The similar database 106 includes a question database 107 that stores questions and a response database 108 that stores responses. Alternatively, an area for storing a question and an area for storing a response may be provided in the similar database 106.

A plurality of questions are stored as data in the question database 107. A plurality of questions are stored as a group in which similar questions are associated with each other. For example, the above-mentioned "Who are you?" And "Please tell me your name" are associated and remembered as similar questions.

The response database 108 stores the response contents to the question, that is, a plurality of responses as data. For example, the above-mentioned reply "I am Annie" is remembered. In the response database 108, the response data is associated with a particular question. According to the example above, the questions "Who are you?" And "Please tell me your name" are grouped together and associated with the response "I am Annie." ing.

FIG. 6A shows another aspect of the similar database 106. FIG. 6A shows an aspect in which the similar database 106 is layered. For example, the similar database 106 is a first similar database 106a corresponding to the first database 102a and a second similarity corresponding to the second database 102b. The database 106b and the third similar database 106c corresponding to the third database 102c may be layered.

The first similar database 106a is created for each specific user based on the personal social data of that user. Text data corresponding to a question sentence and text data corresponding to a reply sentence are stored from text data such as chat included in personal social data and the content of a text-based conversation. In this case, as described with reference to FIG. 5, among the text data corresponding to the question text, similar question text data is grouped and stored in the question database, and the response text data for the text data is stored in the response database and associated. Has been done.

The second similar database 106b is created based on the social data of a plurality of users. Text data corresponding to question sentences and text data corresponding to reply sentences are stored from text data such as chats included in social data of a plurality of users and the contents of text-based conversations. In the second similar database 106b, as described with reference to FIG. 5, among the text data corresponding to the question sentences, similar question sentence data is grouped and stored in the question database, and the response sentence data for the text data is stored in the response database. Is remembered and associated with.

It is possible to change the data stored in the second similar database 106b based on the change in the social data stored in the first similar database 106a. That is, as shown in FIG. 6B, the tendency analysis of the first similar database 106a can be performed, and the result can be used to update the second similar database 106b. For example, suppose that a large number of users may answer the nominations to each of the questions "Who are you?" And "Please tell me your name." In this case, if it is detected in the first similar database 106a that "Who are you?" And "Please tell me your name" are similar questions for a large number of users, the first is In the similar database 106b of 2, the second similar database 106b can be updated so that both the question "Who are you?" And the question "Please tell me your name" are similar questions. it can.

The third similar database 106c stores a plurality of text data. The plurality of text data may be text data of a question sentence prepared in advance and a response sentence to the question sentence. These text data may be a collection of text data having a high frequency of appearance among the text data appearing on the SNS. In the third similar database 106c, as described with reference to FIG. 5, among the text data corresponding to the question sentences, similar question sentence data is grouped and stored in the question database, and the response sentence data for the text data is stored in the response database. Is remembered and associated with.

In this way, the similar database 106 is included in the first similar database in which data categorizing the contents of questions and answers from the social data of the first user is stored, and in the social data of a plurality of users. Hierarchized into a second similar database that stores data that categorizes the contents of questions and answers from, and a third similar database that stores data that categorizes the contents of questions and answers from text data. It is preferable that it is.

According to the present embodiment, those that are judged to have the same answer to a plurality of questions are associated as a combination of the question and the answer. By such a procedure, it is not necessary to store the entire combination of the question and the answer to the question, and the storage capacity can be saved. You can also quickly access the answers to your questions.

Further, in the communication providing system according to the present embodiment, the information processing module 104 having a function as artificial intelligence refers to the second database 102b, and many of the answers from many users have the same contents. If it is determined, it may have a function of automatically adding and updating the question to the answer to the similar database 106. In this way, by updating the similar database by referring to the social data of multiple users, the combination of questions and answers judged to be similar can be made more accurate and accurate, and the accuracy of communication can be improved. it can.

[Generation of phonemes]
In the mode shown in FIG. 1A, the communication providing system according to the present embodiment recognizes the voice by the second user 206 speaking by voice, and the voice (or pseudo) of the first user 204 himself / herself. It may have a function to reply by the voice of the person himself / herself. Alternatively, when the second user 206 asks a question non-voiced by text information or the like, it has a function of replying with the voice of the first user 204 (or a pseudo voice of the person). You may be. As such a method of replying by voice, a method may be adopted in which a large number of voices of the first user 204 are recorded in advance and a reply sentence is created as voice data using the recording. However, such a method forces the user himself / herself to input a large amount of voice data in advance, which is not preferable to be applied to all users and impairs the convenience of the user.

Therefore, in the communication providing system 100 according to the present embodiment, data of user's voice communication performed via SNS is recorded at any time, and phoneme data is created based on the data. For example, a user takes out the words spoken in voice communication and creates a set of a plurality of words.

Then, this set of words is stored as data for creating phonemes. To create a phoneme, first, the characteristics (waveform) of the voice data prepared in advance are compared with the voice waveform of the actual user according to the attributes of the user (for example, gender, age, etc.). Then, the waveform of the voice data prepared in advance is adjusted so as to approach the waveform of the voice of the user. The adjusted voice data is used as the user's phoneme data.

Alternatively, when a method is used in which a large number of voices of the first user 204 are recorded in advance and a response sentence is created as voice data using the voices, the following can be performed. .. That is, when recording a large number of voices of the first user 204 in advance, it is assumed that the first user 204 reads a predetermined sentence to acquire the voice data. At this time, the data of the user's voice communication performed via the SNS is recorded at any time, and the voice emitted by the first user 204 is recognized. Then, when a predetermined sentence or a part of a word constituting a part thereof is recognized, the voice data of the recognized part is acquired and stored. If the voice data of the entire sentence can be obtained by repeating this process and appropriately connecting the acquired and stored voice data, the connected voice data is registered. As a result, even if a method of creating voice data by recording a large number of voices of the first user 204 in advance is used, the first use without imposing a burden on the first user 204. It is possible to acquire voice data obtained by recording a large number of voices of the person 204 himself / herself.

The creation of such phoneme data may be performed by the information processing module 104 and the database 102 having a function as artificial intelligence, or may be executed in cooperation with an external server that creates phonemes.

In the information processing module 104 having a function as artificial intelligence, when the second user 206 asks a question by voice, the text data of the reply sentence is voice data using the phoneme data created as described above. To create. Then, the created voice data is transmitted to the second user 206 as a reply sentence.

In this way, by acquiring the voice data of the conversation performed via the SNS and creating the phonetic data that meets each user, the communication providing system 100 can generate the communication providing system 100 without imposing a special load on the user. Even if it is a pseudo, it is possible to create voice data close to the user himself / herself and establish voice communication.

[Evaluation of answers]
The communication providing system according to the present embodiment can have an evaluation module in which the second user 206 evaluates the content of the answer transmitted from the communication providing system 100. The evaluation of the answer can be performed by operating the user terminal 200 of the second user 206. The user terminal 200 can display the answer content on the screen and display an icon (or button) for evaluation, and the second user 206 can execute the evaluation by operating according to the screen display. it can. The evaluation screen can be constructed by displaying icons (or buttons) (buttons 15 and 16 in FIG. 1J) indicating “Like” and “Disclike” together with the response message. In addition, the evaluation may be in a format in which the user gives a score instead of the alternative method.

FIG. 7 is a diagram illustrating a mode of evaluating the answers answered by the communication providing system 100 to the questions asked by the second user 206 to the virtual first user 204b.

In FIG. 7, a question is asked in order for the second user 206 (any of Mr. A, Mr. B, and Mr. C) to communicate with the virtual first user 204b generated on the communication providing system 100. Is shown (on the left side of the figure). The second user 206 (any of Mr. A, Mr. B, and Mr. C) evaluates the response provided by the communication providing system 100. For example, in the case of Mr. A, it is evaluated as "Discike", and in the case of Mr. B and Mr. C, it is evaluated as "Like" (substantially in the center of the figure).

The information processing module 104 having a function as artificial intelligence can aggregate the evaluation information input to the evaluation module for each response sentence and store it as evaluation data. For example, the evaluation data may be stored on an equal footing with the response sentence data as a point system. The information processing module 104 having a function as artificial intelligence can be ordered based on the evaluation value so that the response data having a high evaluation value is preferentially selected when responding to the question.

Further, when the second user 206 gives a low evaluation to a certain response content, the information may be notified to the first user 204. The first user 204 who received the notification can edit the reply content and update the reply text.

[Edit answer to question]
The communication providing system according to the present embodiment has an editing module in which the first user 204 edits the response sentence created by the communication providing system 100 by himself / herself. For example
"Who are you?" Or "Please tell me your name."
In response to the question
"I am XX (name)."
It is assumed that the information processing module 104 having a function as artificial intelligence has been learned so as to reply. However, some users may think that the response content derived in this way is not appropriate. For example, in response to the above question
"I am the CEO of XX company."
You may think that it is appropriate to answer.

In addition, as shown in FIG. 7, the first user 204 may receive a notification of low evaluation for the response content (left side in the figure). In that case, the first user 204 may want to modify and edit the response content.

In such a case, the communication providing system 100 according to the present embodiment individually sets a response sentence when the user wants to set a response sentence different from the response sentence created by the information processing module 104 having a function as artificial intelligence. It can be edited by the function of the editing module.

As a method of editing the response content, the user may input the text data of the response content from his / her own user terminal 200 and update the content by the update module. Further, as a preferred embodiment, the user may interact with the information processing module 104 having a function as artificial intelligence to learn.

For example, the first user 204 accesses the communication providing system 100 by using the user terminal 200, and the pseudo first user 204 created on the communication providing system 100, for example,
"Please tell me your name"
Ask. The information processing module 104, which has a function as artificial intelligence, provides an answer prepared in advance.
"I am XX (name)."
Respond. The first user 204 can determine that there is no problem with this response content. In this case, the first user 204 does not have to use the editing function. Then, as another question, "Who are you?"
In response to the question, the information processing module 104, which has a function as artificial intelligence, provides an answer prepared in advance.
"I am XX (name)."
When the response is made, the first user 204 determines that this response is not appropriate and uses the editing function. For example, the icon is operated according to the screen display of the user terminal 200 to enter the edit mode. Then, the response content "I am XX (name)" is set to the response content "I am the representative director of the XX company." As a result, it is possible to set to answer "I am the representative director of XX company" to the question "Who are you?".

In this case, when "Tell me your name" and "Who are you" are registered in the same similar group in the similar database 106 shown in FIG. 5, the contents of the similar database 106 are updated. For example, "I am XX (name)" should be answered to "Please tell me your name", and "I am XX company" to "Who are you?" You can re-register as another group so that you can answer "I am the representative director of."

FIG. 8A shows a flowchart illustrating a question and answer evaluation by the second user 206 and a mode in which the first user 204 edits.

In FIG. 8A, the second user 206 accesses the virtual first user 204b in which the communication providing system 100 exists and asks a certain question (question) (S11). The communication providing system 100 generates an answer to the question (S12) and transmits the answer to the second user 206 (S13). Then, the second user 206 evaluates the obtained answer (S14).

The communication providing system 100 aggregates the evaluation values (S15). If the evaluation value is normal or high, or the evaluation value is higher than a certain level, the answer may be weighted as a preferable answer (S16, S17). In this case, the weighted answer may be designed to be more likely to be selected in the next search.

In addition, if the evaluation of the response is low, the first user 204 is notified to that effect (S16, S18). The first user 204 who received the notification (S19) edits the content of the answer so as to give a favorable answer (S20). When the communication providing system 100 receives the edited answer, the communication providing system 100 corrects and updates the corresponding answer recorded in the database (S21). If the first user 204 determines that the content of the answer does not need to be modified, it can be left as it is.

In FIG. 8A, the second user 206 asks a question, and the first user 204 edits the answer to the question. However, one embodiment of the present invention is not limited to this, and the first user 204 talks with the virtual first user 204b appearing in the communication providing system 100, asks a question by himself, and evaluates himself. And the process of editing can be performed. By such a process, the first user 204 can know the content of the answer to the question, evaluate the answer, and edit the answer so that a preferable answer can be obtained. That is, the information processing module 104 having a function as artificial intelligence can be learned.

As described above, the communication providing system 100 according to the present embodiment allows the user to edit the data associated with the question and the response to create a new response sentence. In this case, the user can learn the artificial intelligence function so as to give an appropriate response to the question while interacting with the pseudo self generated on the communication providing system 100.

In this case, the first user 204 edits the answer to the question while interacting with the virtual first user 204b generated on the communication providing system 100, and also functions as artificial intelligence. The information processing module 104 can be trained.

[Display and selection of answer candidates]
FIG. 8B shows a mode in which the communication providing system 100 generates candidates for answers to questions and the user selects them. The communication providing system 100 generates candidates for answers to questions based on the first database 102a in which personal social data is stored. At this time, the first information processing module 104a having a function as artificial intelligence has a function of generating a plurality of answer candidates based on different logics. The first information processing module 104a having a function as artificial intelligence weights the contents of recent conversations from the personal social data of the first user, generates answer candidates, and generates another logic. As a result, answer candidates are created based on the contents of conversations that frequently appear from all the data. For example, in response to a question about a favorite food, there is a logic for answering the food that was recently eaten and delicious, and a logic for answering the food that the first user has liked for a long time.

Further, the second information processing module 104b having a function as artificial intelligence stores in the second database 102b when the first information processing module 104a cannot find an appropriate answer in the first database 102a. Create answer candidates based on the data provided. In the second database 102b, in addition to storing the social data of the entire user, for example, various information on the Internet is crawled and collected and stored. The answer to the question generated by the second information processing module 104b having a function as artificial intelligence is not an answer that reflects the individuality of the individual user, but an answer that is generalized according to social conventions. For example, a question about a player in your own country who has won an international match will be answered as happy.

The answer candidates generated by the information processing module 104 having a function as artificial intelligence are output to the terminal 204a of the first user. For example, the first information processing module 104a having a function as artificial intelligence generates seven kinds of answer candidates, and outputs the answer candidates to the terminal 204a of the first user. FIG. 8B shows an aspect in which answers A1 to A7 are displayed on the terminal 204a of the first user. On the terminal 204a of the first user, the answer candidates generated by the information processing module 104 having a function as artificial intelligence are in descending order of priority (in order of the answer judged to be the most accurate). Answer candidates A1 to A7 are displayed. The first user can change the order of the answers at his / her own discretion by referring to the answers A1 to A7. In FIG. 8B, the order of answer candidates A1 and A3 is changed.

The order of answer candidates changed by the first user is reflected in the information processing module 104 having a function as artificial intelligence. As a result, when the same question is asked next time, answer candidates are generated in the order of A3, A2, and A1. This result is used as an answer to the pseudo first user 204b generated by the information processing module 104 having a function as artificial intelligence. In this way, the information processing module 104 having a function as artificial intelligence generates a plurality of answer candidates for the question, and the first user selects the answer candidates, whereby the pseudo first user The answer in 204b can be made more appropriate.

Such an answer candidate editing function can edit the answer candidate by the function of the above-mentioned editing module, and can update the answer candidate to the question based on the content edited by the editing module by the function of the update module.

[Three-dimensional user photo]
The communication providing system 100 according to the present embodiment can acquire the figure of the user as image data, convert it into three-dimensional image data, and display it on the user terminal. For example, in the configuration shown in FIG. 1A, when the second user 206 interacts with the virtual first user generated on the communication providing system 100, the image of the first user is displayed on the user terminal 200. Can be displayed.

The image data of the user can be acquired by using the imaging function of the user terminal 200. For example, the user takes a picture of himself / herself using the user terminal 200 and uploads the image data to the communication providing system 100. Alternatively, if information is recorded in the SNS or the like in which the user has photographed the user and exchanged a message with another user, the photographed information may be used to extract image data. it can. As a result, the latest image data of the user can be acquired, and the user photograph can be made three-dimensional based on the latest image data of the user.

The communication providing system 100 can infer and calculate the three-dimensional shape of the person in the two-dimensional photographic data to generate the three-dimensional image data. As the method, for example, three-dimensional image data is created by a method of obtaining depth information from the texture density distribution of an object or a method of calculating a normal direction from the intensity of light reflection on the surface of an object.

As described above, the user's 3D image data may be updated as appropriate. As another configuration, for example, image data is searched from the personal social data of a specific user recorded in the first database 102a, and the image data showing the user is captured by using the face recognition function. To identify. Then, it is possible to provide a configuration in which the latest data is selected from the image data in which the user is captured and the three-dimensional image data of the user is updated based on the data.

In this way, when interacting with a virtual individual existing on the communication providing system 100, the communication can be given a sense of substance by displaying the image of the other party as a three-dimensional image on the user terminal. In this case, by updating the three-dimensional image of the virtual individual to the most recent data, the appearance of the talking partner becomes more concrete, and the feeling of intimacy can be enhanced through pseudo communication.

[Automatic question generation]
The communication providing system 100 of the present embodiment has a function of automatically generating and registering a question sentence using the social data stored in the database 102.

For example, the first database 102a may contain many personal "tweets" as personal social data of the first user 204. Personal "tweets" include, for example, "I went shopping in Tokyo today" and "I'm meeting in Shibuya." Although such personal social data does not have a form of response to a question, a "tweet" contains at least one piece of information.

In the above sentence example, information about place names such as "Tokyo" and "Shibuya" is included. In addition, various information such as food, weather, clothes, stock prices, news keywords, and personal names can be found in personal social data.

The information processing module 104 having a function as artificial intelligence decomposes the text data of such personal social data into nouns, verbs, adjectives, etc., and creates and registers question data from the information obtained by syntactically interpreting it. It has a function to keep it. The information processing module 104 having a function as artificial intelligence refers to the social data of a plurality of users registered in the second database 102b, and learns what kind of conversation is taking place. For example, if there are many conversations such as "The function went to XX (place name)" in response to the question "Where did you go?", The question text for the place name is "Where did you go?" A question sentence "?" Is generated and registered in the first database 102. If a plurality of question sentences can be created, they can be registered in the similar database 106.

In this way, the information processing module 104 having a function as artificial intelligence can enhance the dialogue ability by automatically creating a question sentence from personal social data.

[How to express humanity]
You may want to give fluctuations to the answers to the questions. For example
"How are you?"
In response to the question
"I'm fine. How are you?"
Even if the reply is set, it is not always necessary to make this reply. For example, when many other users are replying "It's really good!", You may be influenced by it and reply in the same way.

For this fluctuation, parameters such as a person who has a high probability of occurrence of fluctuation or a person who is not easily affected by other users may be assigned to the user from the personality analysis of the user. As a result, it is possible to produce the personality of each user or the personality of the user depending on how the fluctuation is reflected. Such a parameter can be stored in the communication providing system 100 at the time of registering the 100 in the communication providing system as an attribute of the user, and can be read out as needed. Further, as for the result of such parameter and personality analysis, the result of detecting the temporal change of the user's answer accumulated in the first database 102a is stored in 100 in the communication providing system, and if necessary. It can be read. Similarly, the result of detecting the temporal change of the information about the user stored in the first similar database 106a can be stored in the communication providing system 100 and read out as needed.

[Derivation of answers to questions]
The information processing module 104 having a function as artificial intelligence can respond to a question as follows.

-Overall configuration-
For example, it is assumed that the following conversation is held between the user 1 and the user 2. Here, T1, T2 ... Represent the time.

User 1: "Hello" [T1]
User 2: "Hello" [T2]
User 1: "How are you?" [T3]
User 2: "I'm in good shape, you" [T4]
User 1: "Today is wonderful" [T5]

Tables 1 and 2 summarize the contents of such conversations between the user 1 and the user 2.

Such, in a typical conversation that is exchanged between the user 1 and user 2, for example, the answer to the question of the user 1, is a "good mood How is it" for the "Hello" It turns out. This is, the user 1 to the user 2, can be understood from the context in which has already said, "Hello".

In some cases, the user gives an answer without receiving a question, that is, a one-sided conversation such as a tweet. For example, the following tweet.

User 1: "What a wonderful match" [T1]
User 1: "The future of computing is ubiquitous artificial intelligence" [T2]

Table 3 summarizes such tweets by user 1.

The combination of questions and answers can be obtained from past conversations. The communication providing system 100 can learn past conversations. The communication providing system 100 according to the present embodiment can generate a new question, infer a response, and predict it by learning past conversation data (social data stored in a database). FIG. 9 shows a configuration for searching past conversations for answers and predicting appropriate answers when a question is received.

An important aspect of predicting answers is that artificial intelligence must mimic human behavior. The response to a question depends on the person asking the question. Therefore, in order to predict the response to a question asked by a user, artificial intelligence needs to refer to the past conversation data of the user and predict the high-priority response content obtained from the data.

However, if the appropriate response sentence cannot be detected from the past conversation data, the artificial intelligence finds the appropriate response sentence from the conversation that the first user is exchanging with other users. There is a need. When doing this, it is necessary to be careful not to leak confidential information such as personal information of other users by searching for conversations with other users.

In such a case, in order to eliminate the risk of leakage of personal information, artificial intelligence will estimate the confidentiality score of each answer in the content of conversations exchanged by other users.

For example, a response that has already been shared with another user will have a low confidentiality score. In addition, even if a general answer sentence such as the above example sentence is not shared with other users, the confidentiality score can be set to be low.

FIG. 10 shows a flow of processing for selecting an answer to a question from user 2 to user 1. Artificial intelligence first searches for a suitable answer from the conversations between the first user and the second user that have been exchanged in the past (S21). As a result, if a suitable answer is found, it is selected as the answer (S22, S26). On the other hand, when a suitable answer is not found, one or a plurality of learned answers are selected from the conversations of all users (S22, S23). The confidentiality score is evaluated for the answers obtained here (S24). Then, the one with a low confidentiality score is selected and selected as an answer (S25, S26).

-Search for answers-
Further, as shown in FIG. 11, the answers to the questions may be ranked as follows. For example, as shown in FIG. 11, when an answer is found in the conversations that the first user and the second user have had in the past, the answer is of "individual level" that can be shared between the two. Can be the answer. In addition, when the first user finds an appropriate response sentence from the conversations with other users, the confidentiality score is evaluated for one or more of the obtained answers. Then, the answer with a low confidentiality score can be regarded as a "public level" answer. Furthermore, as shown in FIG. 12, the social data of a plurality of users is learned, the confidentiality score of one or more answers obtained from the social data is evaluated, and the answer with a low score is "shared level". Can be the answer.

ここで、Ｍ_j ⁱは、利用者ｉと利用者ｊとの会話を表すサブモデルである。 When such a user is expressed as a mathematical formula, it can be expressed as the following equation.

Here, M _j ⁱ is a submodel representing a conversation between the user i and the user j.

ここで、Ｑ（ｉ，ｊ）は利用者ｊから利用者ｉに対してなされた質問を表し、Ａ（ｉ、ｊ）は利用者ｉから利用者ｊになされた回答を示し、ｔは対応する時間を示す。 Then, the public submodel M _public ⁱ that represents the conversation of all users can be expressed by the following equation.

Here, Q (i, j) represents a question asked by the user j to the user i, A (i, j) indicates the answer given by the user i to the user j, and t corresponds to the response. Indicates the time to do.

ここで、ｑは利用者ｉに対する利用者ｊの質問（またはクエリ）であり、Ａは予測される回答である。 The relationship between the answers to the questions can be expressed using the function f as follows.

Here, q is the question (or query) of the user j to the user i, and A is the expected answer.

And, for example, if the confidentiality score of the answer is higher than the first threshold, the predicted answer is an individual level answer. Also, if the confidentiality score is below the first threshold and the confidentiality score is lower than the second threshold, the predicted answer is a public level answer. If these are not the case, it can be treated as a shared-level answer. It should be noted that the fact that the confidentiality score is lower than the second threshold value can also be the openness score, which is a concept paired with confidentiality. For example, if the confidentiality score is C _private and the publicity score is C _public , it can be defined as C _private = 1-C _public .

ここで、Ｃ_privateは個人レベルの回答に対する機密性のスコアを与え、Ｃ_publicは公開性のスコアを与える。 Such an aspect of the answer level can be expressed by the following equation.

Here, C _private gives a confidentiality score for individual-level answers, and C _public gives a publicity score.

ここで、ｓｉｍ（ｑ，Ｑｘ）は、ｑ及びｔｕｐｌｅ（Ｑｘ，Ａｘ，ｔｘ）との間の類似性又はマッチングを示す関数である。 There is only one answer that is judged to be the most suitable answer to the question, and it can be shown by the following formula.

Here, sim (q, Qx) is a function indicating similarity or matching between q and tuple (Qx, Ax, tx).

− Contextual integration −
Understanding the context is important when predicting the answer to a question. Answers to the same question can rely on past questions and answers made in the conversation. This is especially true for common questions, such as answering "yes". The answers to such questions can be applied to the functions mentioned above. To integrate the context, learning data of questions and answers in past conversations can be grouped and applied. FIG. 13 shows an aspect of integrating contexts by grouping questions and answers in a conversation. Matching queries with questions and answers has been shown to be a weighted sum.

ここで、ａ_iは重み係数であり、Ｗは文脈ウィンドウに含まれているタプル数を示す。 The similarity when integrating the contexts can be shown by the following equation.

Here, a _i is a weighting factor and W is the number of tuples contained in the context window.

− Matching function −
In order to find the suitability of questions and answers from past conversations, it is necessary to evaluate the similarities between them. Questions and answers can be represented in each query as a set of vectors. The similarity between a set of two vectors can be expressed as a weighted sum of the degree of similarity between the corresponding vectors. Similarities between vectors can be estimated as cosine distance, Euclidean distance, Mahalanobis distance.

Each query for a question and an answer is composed of a set of vectors of keywords (W), concept (C), type (T) and group (G), as shown in the following equation.

-Keywords Keywords are obtained by constructing words from all messages exchanged between users. The similarity of the vectors between the two keywords is weighted (TF-IDF: Robertson, SE, Walker, S., Beaulieu, MM, Gatford, M., & Payne, A. Okapi at TREC-4. In Proceedings of The 4th Text REtrieval Conference (TREC-4), pp. 73-96, 1995]) can be obtained as the cosine distance of the keyword. Alternatively, Okapi BM25 (Harman, D. Ranking algorithms. In WB Frakes & R. Baeza-Yates (Eds.), Information retrieval: Data structures and algorithms, pp. 363-392. Englewood Cliffs, New Jersey, USA: Prentice Hall, 1992]), or any other ranking function used in information retrieval can be used. Similarities between keywords include semantic similarities. It is considered that the two keywords that are semantically related are similar. This semantic relationship is related to WordNet (GA Miller, R. Beckwith, CD Fellbaum, D. Gross, K. Miller. WordNet: An online lexical database. Int. J. Lexicograph. 3, 4, pp. 235-244, 1990), BabelMet (R. Navigli and S. P Ponzetto. BabelNet: The Automatic Construction, Evaluation and Application of a Wide-Coverage Multilingual Semantic Network. Artificial Intelligence, 193, Elsevier, pp. 217-250, 2012) It can be obtained using the data of the analysis (Harris ZS Co-occurrence and transformation in linguistic structure, Language, 33, pp. 283-340., 1957). When estimating similarity, the order and location of keywords in the document must be considered. Use distribution of index representing location as a measure to integrate the order in the system. FIG. 14 shows what keyword order is important in seeking a match.

-Concepts Concepts can be obtained by compressing the dimensions of the vocabulary. Latent Semantic Analysis can be used to provide a limited list of concepts (Thomas K. Landauer, Peter W. Foltz et Darrell Laham. Introduction to Latent Semantic Analysis, Discourse Processes, vol. . 25, p. 259-284, 1998). Each question and answer is projected into the conceptual space. This projection gives a conceptual vector. Then, the cosine distance can estimate the similarity between the concept vectors.

-Types Types are defined as different groups of messages in terms of grammatical structure and reputation analysis. This group includes questions, people, general locations, order, romantic messages, happy messages, and more. These classifications, manually labeled feature vectors, can be obtained by artificial intelligence learning data, concepts and keywords that are grammatical structures.

-The tuples (Q, A) of all users in the group are clustered based on the similarity of the texts between the answers. Each cluster is given a name. The group vector is a list of clusters to which tuples (Q, A) belong. Obviously, the two questions in the same group are similar, as they both have similar answers. FIG. 15 shows a mode in which questions and answers are clustered by the similarity of the answers, and each cluster constitutes one group.

-Template generation Artificial intelligence can be trained using all available data. The input vector is the question and the output vector is the answer. To get the answer to any question, the learned artificial intelligence estimates the answer template. Combined with query matching along with the questions stored in the database, you can match the answers in this template with the answers in the database. This is important so that you can find the answer when the question cannot be found in the data. Any supervised algorithm can be used in this case as a neural network, support vector machine, k-nearest-Neighbors, Gaussian mixed model.

Co-occurrence analysis of keywords between questions and answers can be performed on some of the keywords in the answer template. In FIG. 16, the vectors (keywords, concepts, types) that make up each question have predictors that estimate equivalent answer vectors. The group of answer templates is equivalent to that in the question template.

As described above, in the communication providing system according to the embodiment of the present invention, the communication providing system acquires data from various SNSs and stores it in a database, and an information processing module having a function as artificial intelligence is used by the user. Learning thinking can provide appropriate answers to questions.

The communication providing system according to the embodiment of the present invention can be conceived by one user by collecting social data from a plurality of SNSs and inferring and learning it by an information processing module having a function as artificial intelligence. It can also provide thoughts that are not.

Further, according to one embodiment of the present invention, if the database (first database 102a) that generates virtual users generated by the communication providing system 100 is acquired with specialized knowledge, the virtual users Various services can be provided by. Such expertise may be acquired from the SNS used by the relevant user.

For example, if at least one of the specialized knowledge in each field such as natural science, humanities, psychology, law, finance, and engineering is stored in a database that generates virtual users, the corresponding field is applicable. Expertise can be made available to virtual users. In this case, the specialized knowledge preferably includes not only uniform specialized knowledge accumulated in books and the like, but also information on specialized knowledge used by natural persons as conversations on social networks. By including information on such conversational expertise, the communication providing system 100 can make useful use of various expertise as an answer to a question.

It should be noted that such specialized knowledge can be acquired, for example, as pseudo first user knowledge generated by artificial intelligence, if desired by the first user. In that case, the pseudo first user can not only acquire specific specialized knowledge but also utilize it in a form that takes into account the individuality of the first user. For example, when the first user has knowledge in the medical field, the intellect is that the pseudo first user also has the expertise in the medical field, and the pseudo first user further has the expertise. If you acquire expertise in the field of law, you will be able to demonstrate your abilities in the field of forensic medicine.

In this case, the expertise stored in the first database 102a may be stored in the form of questions and answers (ie, Q & A). The social data (knowledge) stored in the first database 102 can be updated in a timely manner, and the data is not lost, so that the amount of knowledge can be increased.

Further, by reflecting the information of the second database 102b in which the social data of all users is stored in the first database 102a in which such specialized knowledge is stored and enhancing the learning effect, a more appropriate answer can be obtained. Can be output.

As specialized knowledge, it is possible to deal with knowledge in various fields such as justice, administration, medical care, environment, economy, etc., which meet the attributes of users. That is, according to one embodiment of the present invention, it is possible to make a virtual individual having high specialized knowledge appear on the communication providing system according to the attributes of the user and provide the service to other users.

[Knowledge sharing by group: Group AI]
FIG. 2 shows an information processing module (first information processing module) having a function as artificial intelligence, which is composed based on personal social data, and a function as artificial intelligence, which is composed based on the social data of all users. The relationship with the information processing module (second information processing module) having the above is shown. The present invention can further have an information processing module (third information processing module) having a function as artificial intelligence in which a plurality of first information processing modules are assembled to share information or knowledge levels.

FIG. 17 shows an embodiment in which a plurality of first information processing modules 104a having a function as artificial intelligence, which are constructed based on personal social data, exist (first information processing modules 104a_1 to 104a_n (n is 2 or more). integer)). The first information processing module 104a generates pseudo users called "private AI" and "personalized AI" as described with reference to FIG. On the other hand, the second information processing module 104b linked with the first information processing module 104a is based on the social data of the entire user and the fusion of knowledge obtained by crawling various information on the social network. It is artificial intelligence and is called "every one AI", "common sense AI" or "moral AI".

FIG. 17 shows that a plurality of first information processing modules 104a_1 to 104a_n form one set 110. Then, a third information processing module 112 having a function as artificial intelligence, which shares knowledge and information, is connected to the set 110. The third information processing module 112 is artificial intelligence generated by a common knowledge level generated by a plurality of first information processing modules 104a_1 to 104a_n. That is, the third information processing module 112 is a pseudo user having all the knowledge, individuality, and personality of the plurality of pseudo users generated by the plurality of first information processing modules 104a_1 to 104a_n. It can be said that there is. Such a third information processing module 112 can be called a "group AI".

The third information processing module 104c can exhibit the pseudo knowledge and individuality of each user generated by the plurality of first information processing modules 104a_1 to 104a_n. From another point of view, the pseudo-user generated by the third information processing module 112 is the pseudo-user generated by the first information processing module 104a, whereas the pseudo-user is at the individual level. Since it is generated by a collection of multiple individuals, it can be said that it has a corporate character.

The plurality of first information processing modules 104a_1 to 104a_n can talk with the third information processing module 112. As a result, the plurality of first information processing modules 104a_1 to 104a_n can share information and knowledge with each other. The third information processing module 112 provides the knowledge and individuality of the separated users even if one of the pseudo users generated by the first information processing modules 104a_1 to 104a_n leaves this group. Since it can be retained, it is possible to prevent the loss of intelligence due to the withdrawal of the member. For example, when the third information processing module 112 having a function as artificial intelligence is operated as a part of a function of a corporation, even if a member (natural person) belonging to the corporation leaves in the middle, the member who left the corporation. Skills can be retained in the legal entity.

The third information processing module 104c can acquire new information from the plurality of first information processing modules 104a_1 to 104a_n, whereby artificial intelligence grows. That is, by improving the pseudo individual skills generated by the first information processing module 104a_1 to 104a_n, the skills are used as the pseudo user skills generated by the third information processing module 112. It will be reflected.

As shown in FIG. 17, the third information processing module 112 may be placed in a state where it can be accessed from the second information processing module 104b through a telecommunication line. As a result, the third information processing module 112 can collect a wide range of information.

As described with reference to FIG. 17, according to the present invention, a set of a plurality of pseudo-users generated by artificial intelligence is used to form the set of knowledge, intelligence, and individuality of the plurality of users. It is possible to form and utilize a pseudo user who has everything.

100 ... Communication providing system, 102 ... Database, 104 ... Information processing module, 106 ... Similar database, 107 ... Question database, 108 ... Answer database, 200 ... User terminal, 202 ... Multiple users, 204 ... First user, 206 ... Second user, 300 ... SNS

Claims (11)

A communication system that creates a virtual first user on a computer that reflects the individuality of the first user.
Has artificial intelligence to generate answers to questions,
The artificial intelligence
A first layer in which the answer to a question is inferred and learned from the first social data input by the first user to the social networking service, and
A second layer in which the answers to questions are inferred and learned from the second social data entered by multiple users into the social networking service, and
Includes a third layer, which infers and learns the answers to questions by a third social data generated by multiple textual data collected from social networking services.
The first layer, the second layer, and the third layer are layered from upper to lower in this order.
The artificial intelligence
The first social data is used to generate a virtual first user that reflects the individuality of the first user.
Appropriate answers to the questions of the second user are obtained and output in order from the higher layer among the first layer, the second layer, and the third layer.
A communication system characterized by this. The communication system according to claim 1, wherein the first social data and the second social data are accumulated over time. The communication system according to claim 2, wherein the trend analysis of the first social data is performed and the second social data is updated. The communication system according to claim 1, wherein the first social data and the second social data are text data and text data generated from any one or more of voice data, photo data, and video data. The communication system according to claim 1, wherein the artificial intelligence analyzes the text data included in the first social data, infers or learns an answer to a question, and determines the answer. The communication system according to claim 1, further comprising a function of setting whether or not to generate an answer by the artificial intelligence in response to the question by the second user. The communication system according to claim 1, further comprising an image data generation module that generates three-dimensional image data of the first user. The communication system according to claim 1, further comprising a phoneme data generation module that generates phoneme data from the voice data of the first user, and a voice generation module that generates speech voice using the phoneme data. The first social data is stored in a first database, the second social data is stored in a second database, and the third social data is stored in a third database, claim 1. The communication system described in. The ninth layer, wherein the first layer cooperates with the first database, the second layer cooperates with the second database, and the third layer cooperates with the third database. Communication system . The communication system is placed in a communicable state in both directions through a plurality of users and telecommunication lines, communication system according to any one of claims 1 to 10.

Images