Information processing device and information processing method

Abstract

The present technology relates to an information processing device and an information processing method that make it possible to use a database for another purpose. An algorithm, that changes on the basis of the accumulation of first learning data, undergoes re-learning on the basis of: the first learning data; and specific learning data of second learning data forming another algorithm that changes on the basis of the accumulation of learning data. The first learning data includes data relating to information output from the algorithm pursuant to information input to the algorithm. The present technology is applicable to, for example, artificial intelligence.

Description

[Technical Field]

[0001] The present technology relates to an information processing device and an information processing method, and more particularly to an information processing device and an information processing method that are suitable for use when reusing a database, for example. [Background technology]

[0002] Conventionally, AI agent systems have been proposed that automatically respond to inputs from users by voice, etc. For example, Patent Document 1 describes a technology in which an AI agent responds by voice to utterance data from a user. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2017 / 191696 Summary of the Invention [Problem to be solved by the invention]

[0004] On the other hand, the responses of the above AI agents may be output based on algorithms based on accumulated learning data. In the future, as technologies related to algorithms based on accumulated learning data become more accessible to users, it is expected that users will encounter situations where they want to modify the accumulated learning data.

[0005] In addition, there may be cases where other users wish to use an algorithm that a particular user has used, and it is assumed that there may be a desire to be able to reuse such other algorithms (databases referenced to generate algorithms).

[0006] The present technology has been made in view of such circumstances, and makes it possible to reuse databases. [Means for solving the problem]

[0007] An information processing device according to one aspect of the present technology includes: first input information; The first one suitable for the first user First learning data relating to first output information based on an algorithm, second input information, and a learning process for the second input information. A second one suitable for the second user Integrating second learning data related to second output information based on the algorithm, and re-learning using the integrated learning data as new first learning data. Update the first algorithm .

[0008] An information processing method according to one aspect of the present technology includes: an information processing device; The first one suitable for the first user First learning data relating to first output information based on an algorithm, second input information, and a learning process for the second input information. A second one suitable for the second user Integrating second learning data related to second output information based on the algorithm, and re-learning using the integrated learning data as new first learning data. Update the first algorithm .

[0009] In one aspect of the present technology, an information processing device and an information processing method include: first input information; The first one suitable for the first user First learning data relating to first output information based on an algorithm, second input information, and a learning process for the second input information. A second one suitable for the second user The second learning data relating to the second output information based on the algorithm is integrated, and re-learning is performed using the integrated learning data as new first learning data. The first algorithm is updated. .

[0010] The information processing device may be an independent device or an internal block constituting one device. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 illustrates an information processing system and a user terminal according to a first embodiment. [Figure 2]3 is an example of a schematic configuration of learning history data stored in an information processing device. [Figure 3] FIG. 1 is a functional block diagram showing a configuration of an information processing device. [Figure 4] FIG. 2 is a functional block diagram illustrating an example of the configuration of a storage unit. [Figure 5] FIG. 2 is a functional block diagram showing the configuration of a processing unit. [Figure 6] FIG. 2 is a functional block diagram showing the configuration of a generation unit. [Figure 7] FIG. 2 is a functional block diagram showing the configuration of a user terminal. [Figure 8] 10 is a flowchart illustrating an example of transmission and reception of information between an information processing device and a user terminal. [Figure 9] 10 is a flowchart illustrating an example of transmission and reception of information between an information processing device and a user terminal. [Figure 10] 10 is a flowchart illustrating an example of transmission and reception of information between an information processing device and a user terminal. [Figure 11] 10A and 10B are diagrams showing examples of information recorded in an exchange DB, an update history of a knowledge DB, and an update history of a recommendation DB. [Figure 12] FIG. 10 is a diagram for explaining a method of using another user's DB. [Figure 13] FIG. 10 is a diagram for explaining a first method of reusing another user's DB. [Figure 14] FIG. 10 is a diagram illustrating a second method of reusing another user's DB. [Figure 15] FIG. 10 is a diagram illustrating a third method of reusing another user's DB. [Figure 16] FIG. 10 is a diagram illustrating a fourth method of reusing another user's DB. [Figure 17] FIG. 10 is a diagram illustrating a fifth method of reusing another user's DB. [Figure 18] FIG. 10 is a diagram for explaining how to integrate DBs. [Figure 19] FIG. 10 is a diagram illustrating a sixth method of reusing another user's DB. [Figure 20] FIG. 10 is a diagram for explaining how to integrate DBs. [Figure 21] FIG. 10 is a functional block diagram showing a configuration of an information processing device according to a second embodiment. [Figure 22] FIG. 10 is a functional block diagram showing the configuration of a storage unit according to a second embodiment. [Figure 23] 10A and 10B are diagrams illustrating an example of information recorded in an exchange DB, an update history of a knowledge DB, and an update history of a recommendation DB according to the second embodiment. [Figure 24] FIG. 10 is a functional block diagram showing the configuration of a processing unit according to a second embodiment. [Figure 25] FIG. 10 is a functional block diagram illustrating a configuration of a generation unit according to a second embodiment. [Figure 26] FIG. 10 is a flowchart illustrating an example of a parameter update process according to the second embodiment. [Figure 27] FIG. 10 is a diagram for explaining a method of using another user's DB. [Figure 28] FIG. 10 is a diagram illustrating a seventh method of reusing another user's DB. [Figure 29] FIG. 10 is a diagram illustrating an eighth method of reusing another user's DB. [Figure 30] FIG. 13 is a diagram illustrating a ninth method of reusing another user's DB. [Figure 31] FIG. 16 is a diagram for explaining a tenth method of reusing another user's DB. [Figure 32] FIG. 19 is a diagram for explaining an eleventh method of reusing another user's DB. [Figure 33] FIG. 19 is a diagram illustrating a twelfth method of reusing another user's DB. [Figure 34] FIG. 13 is a diagram illustrating a thirteenth method of reusing another user's DB. [Figure 35] FIG. 10 is a diagram for explaining extraction of information from a DB. [Figure 36]10A and 10B are diagrams showing output information generated before and after deletion of information related to an exchange, and processing contents based on changes in the output information before and after the deletion. [Figure 37] 10A and 10B are diagrams showing output information generated before and after deletion of information related to an exchange, and processing contents based on changes in the output information before and after the deletion. [Figure 38] 10 is a flowchart illustrating an update process of an exchange DB by an information processing device according to an embodiment of the present disclosure. FIG. [Figure 39] FIG. 10 is a diagram for explaining how to integrate DBs. [Figure 40] FIG. 10 is a diagram for explaining how to integrate DBs. [Figure 41] FIG. 10 is a diagram for explaining how to integrate DBs. [Figure 42] FIG. 10 is a diagram for explaining how an exchange DB is provided. [Figure 43] FIG. 10 is a diagram for explaining how to integrate DBs of different formats. [Figure 44] FIG. 1 is a functional block diagram showing a configuration of an information processing device. [Figure 45] 1 is a functional block diagram illustrating an example of a hardware configuration of an information processing device that constitutes a user terminal or an information processing system according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, modes for carrying out the present technology (hereinafter referred to as embodiments) will be described. In this specification and drawings, components having substantially the same functional configurations are denoted by the same reference numerals, and redundant description will be omitted.

[0013] <Configuration of Information Processing System in First Embodiment> FIG. 1 is a diagram showing an example of the configuration of an information processing system 1 according to the first embodiment. As shown in FIG. 1, the information processing system 1 according to the first embodiment is configured with an information processing device 10. Hereinafter, in the first embodiment, it is assumed that the information processing system 1 and the information processing device 10 are the same. As shown in FIG. 1, the information processing system 1 is configured to include the information processing device 10, a user terminal 20, and a network 30. In the information processing system 1, the information processing device 10 and the user terminal 20 are connected so as to be able to exchange data via the network 30.

[0014] The information processing device 10 has a function of generating output information using an algorithm generated based on accumulated learning data in response to input information from a user. Furthermore, the information processing device 10 re-learns the algorithm as necessary.

[0015] The user terminal 20 has a function of transmitting information input by a user to the information processing device 10 via the network 30 and providing various outputs (e.g., image or audio output) to the user in response to a response from the information processing device 10. In this embodiment, the user terminal 20 realizes output by an AI agent. Here, the AI ​​agent is a character that serves as a motif for audio or images output based on an algorithm. The character may be a fictional character or a real character.

[0016] The network 30 may include a public network such as a telephone network, the Internet, or a satellite communication network, a local area network (LAN), a wide area network (WAN), etc. The network 30 may also include a dedicated network such as an Internet Protocol-Virtual Private Network (IP-VPN).

[0017] The above algorithm is based on learning data accumulated in the information processing device 10. In other words, the above algorithm is a learning result based on the learning data. The information processing device 10 according to the present embodiment stores a learning history of the above algorithm.

[0018] Here, an example of learning history data 40 stored in the information processing device 10 according to the first embodiment will be described with reference to Fig. 2. Fig. 2 shows an example of a schematic configuration of the learning history data 40 stored in the information processing device 10 according to the first embodiment. The learning history data 40 shown in Fig. 2 is data configured by arranging the learning contents of an algorithm in chronological order. In Fig. 2, the learning contents of three learning sessions with learning numbers A to C are arranged in chronological order.

[0019] For example, in the learning of NoA, the content of learning content A is learned at time A. In addition, in the learning of NoB, the content of learning content B is learned at time B. Furthermore, in the learning of NoC, the content of learning content C is learned at time C. At each time, learning of the algorithm is performed based on the learning data corresponding to each learning. In this embodiment, learning of the algorithm is performed based on the accumulated learning data in this way.

[0020] 2 has a configuration in which learning data and learning contents are arranged in chronological order, but the configuration of learning history data is not limited to this. Also, while FIG. 2 shows three learning histories for learning Nos. A to C, the learning history data may have two or fewer learning histories, or may have four or more learning histories. Also, multiple learning data may be used in one learning, and multiple learnings may be performed.

[0021] Furthermore, the learning data is not particularly limited, and may be based on data accumulated in the environment in which the algorithm is used, for example. As a result, the algorithm based on the accumulated learning data can be an algorithm that is suited to the environment in which the algorithm is used by the user. This allows the information processing system 1 to more appropriately realize the state of the algorithm desired by the user.

[0022] The learning data may also include data on output information from the algorithm based on user input information for the algorithm. This allows the algorithm to learn based on daily input information from the user and output information based on the input information. The input information and output information may include information specific to the user. This allows the information processing system 1 to more appropriately realize the state of the algorithm desired by the user based on the learning data.

[0023] <Configuration of information processing device> The following describes the configuration of the information processing device 10. Fig. 3 is a functional block diagram showing the configuration of the information processing device 10 according to the first embodiment.

[0024] The information processing device 10 has a function of adjusting the influence of specific learning data in the accumulated learning data on an algorithm that changes based on the accumulation of learning data, and performing re-learning based on new learning data obtained after the adjustment. The functions of the information processing device 10 are realized by cooperation between a storage unit 110, a processing unit 120, an analysis unit 130, a generation unit 140, an output control unit 150, and a communication control unit 160 provided in the information processing device 10. Each functional unit provided in the information processing device 10 will be described below.

[0025] The storage unit 110 has a function of storing various types of information. The various types of information stored in the storage unit 110 are referenced by the processing unit 120, the analysis unit 130, the generation unit 140, or the communication control unit 160 as necessary.

[0026] Here, the storage unit 110 according to the present embodiment will be described in more detail with reference to Fig. 4. Fig. 4 is a functional block diagram showing an example of the configuration of the storage unit 110 according to the first embodiment. As shown in Fig. 4, the storage unit 110 has a knowledge DB (database) 111, a recommendation DB 112, an interaction DB 113, and a learning DB 114.

[0027] The knowledge DB 111 stores various types of information that are used by the analysis unit 130 to analyze input information from a user. For example, the knowledge DB 111 stores information about the meanings of various words. Also, for example, the knowledge DB 111 stores dictionary-like meanings of various words.

[0028] Furthermore, for example, the knowledge DB 111 records the meanings of words specific to each user. For example, the word "Gunma" generally means Gunma Prefecture. However, depending on the user, the word "Gunma" may mean a person's name (for example, a person's name "Iwasa"). In this case, the knowledge DB 111 stores that the word "Gunma" may refer to a person's name "Iwasa." The meanings of various words may be defined probabilistically in the knowledge DB 111. For example, the knowledge DB 111 may record that there is a 70% probability that the word "Gunma" means a person's name "Iwasa."

[0029] The knowledge DB 111 may also store user data, which is data related to the user. The user data may be included in learning data for learning the algorithm. This allows the user to modify information about themselves that is included in the learning data. As a result, the information processing device 10 can more appropriately re-learn the algorithm, and more appropriately achieve the state of the algorithm desired by the user.

[0030] The knowledge DB 111 may also store information about the weather or news. The knowledge DB 111 may also store information such as memos or reminders entered by the user. The knowledge DB 111 may also store information for web extraction or product extraction.

[0031] The recommendation DB 112 stores various data used by the recommendation information generation unit 142 (described later) to generate output information. For example, the recommendation DB 112 may store data related to the user's preferences. For example, assume that the user regularly listens to songs by a person named "Gunma," and information related to this fact is input to the storage unit 110. In this case, the recommendation DB 112 may store a playlist of songs by "Gunma." The recommendation DB 112 may also store a score list in which various songs are assigned recommendation scores, such as a recommendation score of 0.2 for song A and a recommendation score of 0.8 for song B. The recommendation DB 112 may also store information related to music or purchases recommended to the user.

[0032] The information recorded in the recommendation DB 112 is transmitted to the generating unit 140 and is used by the generating unit 140 to generate output information.

[0033] The interaction DB 113 stores data relating to input information from users and output information based on an algorithm for the input information. For example, suppose that one day a user inputs input information requesting that the song "Gunma" be played to the information processing system 1. As a result, the information processing system 1 generates output information for playing the song "Gunma" based on the algorithm and outputs it to, for example, the user terminal 20. At this time, the interaction DB 113 records the contents of the input information and the output information, the times when these pieces of information were input and output, and the like.

[0034] In this embodiment, the information recorded in the interaction DB 113 in this manner (for example, information related to input information and output information) is used as label information for extracting learning data recorded in the learning DB 114, which will be described later. In the first embodiment, the information recorded in the interaction DB 113 is also used to update the information recorded in the knowledge DB 111 or the recommendation DB 112.

[0035] When the information processing device 10 according to this embodiment acquires input information, it generates output information based on the information recorded in the knowledge DB 111 and the recommendation DB 112. Therefore, the algorithm for generating output information executed by the information processing device 10 uses the information recorded in the knowledge DB 111 and the recommendation DB 112. Therefore, when the information recorded in the knowledge DB 111 or the recommendation DB 112 is updated based on the accumulation of learning data recorded in the interaction DB 113, the above algorithm changes.

[0036] Note that, every time input information and output information are input / output, the information recorded in the knowledge DB 111 or the recommendation DB 112 may be updated. In this case, every time input information and output information are input / output, the algorithm by which the information processing device 10 generates the output information changes.

[0037] The learning DB 114 records learning data. The learning data may be recorded based on a user instruction or automatically in the background by the information processing system 1. The learning data may include various information necessary for output, such as the content of the user instruction, the user's situation, and the user's surrounding environment. The learning data may also include an index for estimating the appropriateness of a result (e.g., an analysis result by the analysis unit 130 described below or a recommendation result by the recommendation information generation unit 145). The index may be, for example, feedback from the user regarding the analysis or recommendation. A learning history of an algorithm used by the information processing device 10 to generate output information is recorded. The learning DB 114 may record a learning history in a format in which learning contents are arranged in chronological order, as shown in FIG. 2, for example. Note that the information recorded in the learning DB 114 does not necessarily include the learning data itself. In this case, the learning DB 114 may record information that associates the learning history with the learning data corresponding to the learning history.

[0038] (Processing section) The processing unit 120 has a function of performing various processes on the information stored in the storage unit 110. The processing unit 120 has a function of adjusting the influence of specific learning data in the accumulated learning data. The processing unit 120 also has a function of re-learning the algorithm based on new learning data obtained after the adjustment. The results of the processing by the processing unit 120 are transmitted to at least either the analysis unit 130 or the storage unit 110 as necessary. The adjustment of the influence and the learning of the algorithm will be described later with reference to FIG. 5.

[0039] In this embodiment, the influence degree is adjusted by the processing unit 120, and then the algorithm is re-learned. Note that the influence degree may be, for example, the degree of influence on output information based on the algorithm. Therefore, the output information is modified by adjusting the influence degree. Therefore, the information processing device 10 can generate output information that is more appropriate for the user by adjusting the influence degree.

[0040] The specific learning data may be specified by the user. By specifying the specific learning data, the influence of the learning data desired by the user is adjusted. As a result, the information processing device 10 can more appropriately realize the state of the algorithm desired by the user.

[0041] The specific learning data may also include user data, which is data related to the user. By adjusting the influence derived from the user data, the state of the algorithm desired by the user can be more appropriately realized. As a result, the information processing device 10 can generate output information that is more in line with the user data.

[0042] The user data may also include location information regarding the user's location. This allows the information processing device 10 to generate output information with content that is more in line with the user's location. The user data may also include information regarding the user's preferences. This allows the information processing device 10 to generate output information with content that is more in line with the user's preferences.

[0043] Furthermore, the processing unit 120 may re-learn the algorithm in response to changes in user data, so that when there is a change in the user data, the information processing device 10 can realize an algorithm state that corresponds to the change and generate more appropriate output information.

[0044] The processing unit 120 will be described in more detail with reference to Fig. 5. Fig. 5 is a functional block diagram showing the configuration of the processing unit 120 according to the first embodiment. As shown in Fig. 5, the processing unit 120 can acquire learning data and output the corrected learning data by correcting the learning data. Also, as shown in Fig. 5, the processing unit 120 includes an updating unit 121, an extracting unit 122, a determining unit 123, and a correcting unit 124. Information generated by these functional units may be transmitted between these functional units as appropriate.

[0045] The update unit 121 has a function of updating various pieces of information recorded in at least one of the knowledge DB 111 and the recommendation DB 112 of the storage unit 110. For example, the update unit 121 updates various pieces of information in response to input information from a user. Furthermore, the update unit 121 updates various pieces of information in response to changes in the learning data recorded in the learning DB 114. For example, the update unit 121 may update the recommendation score recorded in the recommendation DB 112 when the learning data recorded in the learning DB 114 is deleted, corrected, or the like.

[0046] The information processing device 10 according to this embodiment executes an algorithm that "acquires input information and generates output information based on various pieces of information recorded in the knowledge DB 111 or the recommendation DB 112." In this embodiment, updating various pieces of information recorded in the knowledge DB 111 or the recommendation DB 112 corresponds to relearning the algorithm.

[0047] The extraction unit 122 has a function of extracting various types of information recorded in the storage unit 110. More specifically, the extraction unit 122 extracts a specific learning history that meets a predetermined condition from the learning history of the algorithm based on a database in which data related to input information is recorded. The specific learning history that meets the predetermined condition may be, for example, a learning history that the user wants to delete. The specific learning history that meets the predetermined condition is used to re-learn the algorithm. For example, when the specific learning history is deleted, the algorithm is re-learned as if the learning history did not exist.

[0048] Furthermore, learning data may be associated with the learning history. The information processing device 10 according to the present embodiment may adjust the influence derived from the learning data. By adjusting the influence, the algorithm is re-learned. This allows the information processing device 10 to generate output information that is more appropriate for the user.

[0049] The extraction unit 122 according to the first embodiment extracts a specific learning history that meets a predetermined condition from the learning DB 114 based on the interaction DB 113 in which data related to input information is recorded. The specific learning history that meets the predetermined condition may be, for example, a history related to learning an algorithm based on learning data that includes a keyword specified by the user. Therefore, the extraction unit 122 extracts a learning history based on learning data that includes a keyword specified by the user.

[0050] For example, the extraction unit 122 acquires input information requesting extraction of the keyword "Gunma" from a user. At this time, the extraction unit 122 may extract information including the keyword "Gunma" from at least one of the knowledge DB 111, the recommendation DB 112, and the exchange DB 113 in the storage unit 110. Furthermore, the extraction unit 122 according to this embodiment may extract, from the learning DB 114, a learning history indicating that learning has been performed based on learning data including the keyword.

[0051] The determination unit 123 has a function of performing various determinations. For example, the determination unit 123 may determine the magnitude of change between the output information recorded in the exchange DB 113 and the output information generated by the generation unit 140. The result of the determination by the determination unit 123 is transmitted to the correction unit 124. As will be described later, the correction unit 124 deletes or corrects the learning data based on the determination result.

[0052] The correction unit 124 has a function of adjusting the degree of influence derived from the learning data. More specifically, the correction unit 124 according to the first embodiment has a function of adjusting the degree of influence derived from the learning data by deleting or correcting the learning data recorded in the learning DB 114. The correction unit 124 may, for example, delete or correct information recorded in the learning DB 114 that represents the output information that has been output.

[0053] As described above, the interaction information and the like recorded in the interaction DB 113 influence the algorithm. The correction unit 124 can adjust the degree of influence derived from the learning data by deleting or correcting the learning data recorded in the learning DB 114. More specifically, the correction unit 124 can eliminate the degree of influence derived from the learning data on the algorithm by deleting the learning data. Furthermore, the correction unit 124 can increase or decrease the degree of influence derived from the learning data on the algorithm by correcting the learning data. In this way, the correction unit 124 can adjust the degree of influence derived from the learning data by deleting or correcting the learning data recorded in the learning DB 114.

[0054] Furthermore, the information recorded in the knowledge DB 111 or the recommendation DB 112 is information based on the learning data recorded in the learning DB 114. Therefore, the correction unit 124 deletes or corrects the learning data (e.g., information related to input information and output information) recorded in the learning DB 114, whereby the information recorded in the knowledge DB 111 or the recommendation DB 112 is re-learned. In this way, the information processing device 10 according to the present embodiment adjusts the influence derived from the learning data and causes the algorithm to re-learn.

[0055] (Generation part) The generation unit 140 has a function of generating various types of output information based on the information stored in the storage unit 110. The generated output information is transmitted to the output control unit 150. The function of the generation unit 140 will be described in more detail with reference to FIG. 6. FIG. 6 is a functional block diagram showing the configuration of the generation unit 140 according to the first embodiment. As shown in FIG. 6, the generation unit 140 includes a confirmation information generation unit 141 and a recommendation information generation unit 142.

[0056] The confirmation information generating unit 141 has a function of generating output information for the user to confirm various points. For example, when the correction unit 124 deletes the learning data recorded in the learning DB 114, the confirmation information generating unit 141 may generate output information for the user to confirm whether the learning data may be deleted.

[0057] The recommendation information generation unit 142 generates output information for making various recommendations to the user. For example, the recommendation information generation unit 142 may generate output information for playing music desired by the user. In this case, the recommendation information generation unit 142 may determine the content to be recommended to the user based on the recommendation score recorded in the recommendation DB 112, and generate the output information.

[0058] (Output control section) The output control unit 150 has a function of controlling the output of output information. For example, the output control unit 150 may convert the output information acquired from the generation unit 140 into information to be output to another terminal. For example, when the output information is composed of text information, the output control unit 150 may convert the content of the text information into audio information to be output as audio. The output control unit 150 transmits various pieces of information acquired or generated to the communication control unit 160. Note that the output control unit 150 may transmit the output information transmitted from the generation unit 140 to the communication control unit 160 as is.

[0059] (Communication control unit) The communication control unit 160 has a function of controlling transmission and reception of various types of information between the information processing device 10 and various devices. For example, the communication control unit 160 controls transmission of information transmitted from the output control unit 150 from the information processing device 10 to the user terminal 20 via the network 30. The communication control unit 160 also controls reception of various types of information by the information processing device 10 from external devices (e.g., the user terminal 20). The received various types of information are transmitted via the communication control unit 160 to the storage unit 110, the processing unit 120, or the analysis unit 130.

[0060] <User device> Next, the configuration of the user terminal 20 according to the first embodiment will be described with reference to Fig. 7. Fig. 7 is a functional block diagram showing the configuration of the user terminal 20 according to the first embodiment. As shown in Fig. 7, the user terminal 20 includes a communication control unit 210 and an output control unit 220.

[0061] (Communication control unit) The communication control unit 210 has a function of controlling transmission and reception of various types of information between the user terminal 20 and various devices (for example, the information processing device 10). The communication control unit 160 acquires input information and controls transmission of the input information to the information processing device 10. Note that the input information may be input to the user terminal 20 based on an operation by a user, or may be automatically input from various devices. The communication control unit 210 also controls reception of information related to output information transmitted from the information processing device 10. The received information related to the output information is transmitted to the output control unit 220.

[0062] (Output control section) The output control unit 220 controls various outputs by the user terminal 20. For example, the output control unit 220 transmits information related to the output information transmitted from the information processing device 10 to an output device provided in the user terminal 20, thereby causing the output device to perform various outputs. For example, the output control unit 220 may control the output device to play music.

[0063] <Transmission and reception of information between an information processing system and a user terminal> Next, with reference to Figs. 8 to 10, transmission and reception of information between the information processing device 10 according to the first embodiment and the user terminal 20 will be described. Figs. 8 to 10 are flowcharts showing an example of transmission and reception of information between the information processing device 10 according to the first embodiment and the user terminal 20. First, with reference to Fig. 8, an example of transmission and reception of information (input information and output information) between the information processing device 10 according to the first embodiment and the user terminal 20 (hereinafter also referred to as "exchange between the information processing device 10 and the user terminal 20") will be described.

[0064] In the example shown in Fig. 8, the information processing device 10 generates output information in response to input information transmitted from the user terminal 20. The user terminal 20 receives the generated output information and outputs various pieces of information to the user in response to the output information. The transmission and reception of information between the information processing device 10 and the user terminal 20 will be described in more detail below with reference to Fig. 8.

[0065] First, the user terminal 20 acquires input information (step S102). More specifically, the communication control unit 210 included in the user terminal 20 acquires the input information from the user. For example, the communication control unit 210 acquires audio information such as "Gunma's favorite song is XX" as the input information. Next, the user terminal 20 transmits the input information to the information processing device 10 (step S104).

[0066] Next, the information processing device 10 receives the input information (step S106). The received input information is transmitted to the analysis unit 130 via the communication control unit 160.

[0067] Next, the analysis unit 130 analyzes the input information (step S108). More specifically, the analysis unit 130 analyzes the semantics of the input information based on various information stored in the knowledge DB 111. For example, it is assumed that the knowledge DB 111 stores information that the word "Gunma" only means "Gunma Prefecture." In this case, the analysis unit 130 cannot understand the semantics of the input information and outputs an analysis result to the storage unit 110 that the input information "Gunma's favorite song is XX" is contradictory. At this time, the storage unit 110 records the analysis result of the analysis unit 130 in the interaction DB 113. More specifically, the storage unit 110 stores the input information "Gunma's favorite song is XX" in association with the time when the input information was transmitted in the interaction DB 113.

[0068] Next, the generation unit 140 generates output information (step S110). More specifically, the generation unit 140 outputs the output information based on the result of the analysis by the analysis unit 130 and the information stored in the storage unit 110. For example, the generation unit 140 generates voice information of "What is Gunma?" as output information and transmits the output information to the output control unit 150. At this time, the storage unit 110 records the output information of "What is Gunma?" in the learning DB 114 in association with the information recorded in step S108. At this time, the storage unit 110 records in the interaction DB 113 that an interaction has occurred. More specifically, the storage unit 110 records in the interaction DB 113 the time when the interaction occurred and the content of the interaction.

[0069] Next, the information processing device 10 transmits the output information to the user terminal 20 (step S112).

[0070] Next, the user terminal 20 outputs the output information (step S114). More specifically, the communication control unit 210 acquires the output information transmitted to the user terminal 20 and transmits the output information to the output control unit 220. Based on the output information, the output control unit 220 causes an output device included in the user terminal 20 to output the output information. In this case, the output device outputs audio information such as "What is Gunma?" as the output information.

[0071] The above describes an example of information transmission and reception between the information processing device 10 and the user terminal 20. As described above, the input information and output information are recorded in the learning DB 114 and used as learning data.

[0072] Next, a second example of an interaction between the information processing device 10 according to the first embodiment and the user terminal 20 will be described with reference to Fig. 9. The interaction shown in Fig. 9 differs from the example of an interaction shown in Fig. 8 in that the information processing device 10 updates the information recorded in the knowledge DB 111 and the recommendation DB 112 based on input information.

[0073] First, the user terminal 20 acquires input information (step S202). More specifically, the communication control unit 210 acquires input information from the user. For example, the communication control unit 210 acquires voice information such as "Gunma is the nickname of my friend Iwasa." as input information.

[0074] Next, the processes of steps S204 and S206 are carried out. However, since the processes of steps S204 and S206 are substantially the same as the processes of steps S104 and S106, a description thereof will be omitted here.

[0075] When the process of step S206 is completed, the information processing device 10 analyzes the input information (step S208). More specifically, the analysis unit 130 analyzes the semantic content of the input information based on the information recorded in the knowledge DB 111. The analysis unit 130 transmits the analysis result to the processing unit 120.

[0076] Next, the update unit 121 updates the information recorded in the knowledge DB 111 (step S210). More specifically, the update unit 121 records in the knowledge DB 111 that "Gunma" is a friend of the user. Also, here, it is assumed that information that "Gunma's favorite song is XX" is recorded in the knowledge DB 111. At this time, based on the analysis result in step S208, the update unit 121 records in the knowledge DB 111 that "Gunma" likes the song XX. Also, the update unit 121 creates a playlist of songs that Gunma likes in the recommendation DB 112. Furthermore, the update unit 121 adds the song XX that Gunma likes to the playlist.

[0077] An example of the interaction between the information processing device 10 and the user terminal 20 has been described above with reference to Fig. 9. In the example shown in Fig. 9, the information processing device 10 updates the information stored in the storage unit 110 in response to input information from the user terminal 20. This enables the information processing device 10 to generate output information that is more in line with the user's wishes. For example, when input information such as "Please play Gunma's favorite song" is input, the information processing device 10 can generate output information for causing the user terminal 20 to output XX, which is Gunma's favorite song.

[0078] Next, a third example of communication between the information processing device 10 according to the present embodiment and the user terminal 20 will be described with reference to Fig. 10. In the third example, in addition to the processing according to the second example, processing is added in which the information processing device 10 transmits output information to the user terminal 20 and the user terminal 20 outputs the output information. Hereinafter, the third example will be described with reference to Fig. 10.

[0079] First, the user terminal 20 acquires input information (step S302). More specifically, the communication control unit 210 acquires input information from the user. For example, the communication control unit 210 acquires voice information such as "Play the song Gunma" as input information.

[0080] Next, the processes of steps S304 to S308 are carried out. However, since the processes of steps S304 to S306 are substantially the same as the processes of steps S204 to S208, a description thereof will be omitted here.

[0081] When the process of step S308 ends, the information processing device 10 generates output information (step S310). More specifically, the recommendation information generation unit 142 generates the output information based on the analysis result by the analysis unit 130 and the information recorded in the recommendation DB 112. For example, the recommendation information generation unit 142 generates output information for playing songs included in Gunma's favorite playlist stored in the recommendation DB 112. The song to be played may be the song with the highest recommendation score, or may be a song randomly selected from, for example, the top 5% of recommendation scores. The recommendation information generation unit 142 transmits the output information to the output control unit 150.

[0082] Next, the information processing device 10 transmits the output information to the user terminal 20 (step S312).

[0083] Next, the information processing device 10 updates the information stored in the storage unit 110 (step S314). For example, the update unit 121 increases the recommendation score of the song selected by the recommendation information generation unit 142. The update unit 121 also records the input information that has been input and the output information that has been output in the learning DB 114. The interaction DB 113 also records the time when an interaction occurred, the content of the interaction, and the like.

[0084] Next, the user terminal 20 outputs the output information (step S316). More specifically, the communication control unit 210 acquires the output information transmitted to the user terminal 20 and transmits the acquired output information to the output control unit 220. The output control unit 220 causes the output device to output the output information. As a result, the output device plays, for example, songs included in Gunma's favorite playlist.

[0085] 10, a third example of the interaction between the information processing device 10 according to the present embodiment and the user terminal 20 has been described. According to the third example, the information processing device 10 generates output information in response to input information from the user terminal 20, and transmits the generated output information to the user terminal 20. In addition, the information processing device 10 updates the information stored in the storage unit 110 in response to the input information and the output information.

[0086] 8 to 10, output information related to "Gunma" was generated based on input information including the noun "Gunma." Without being limited to this, output information related to "Gunma" may be generated based on input information that does not include the noun "Gunma." In other words, the user can implicitly instruct the information processing device 10 to generate output information related to "Gunma" without uttering the word "Gunma."

[0087] For example, in step S302, the user inputs voice information such as "Play a song with a similar style to the song I listened to yesterday." The voice information does not include the word "Gunma," but the phrase "the song I listened to yesterday" means "a song from Gunma." Then, in step S308, the analysis unit 130 analyzes the phrase "the song I listened to yesterday" contained in the user's input information based on the information stored in the knowledge DB 111 to determine that the phrase means "a song from Gunma." As a result, in step S310, the recommendation information generation unit 142 generates output information for causing the user terminal 20 to play a song with a similar style to the "Gunma song." In response, in step S314, the update unit 121 increases the recommendation score of the song with a similar style to the "Gunma song" stored in the recommendation DB 112. Furthermore, in step S316, the user terminal 20 plays the song with a similar style to the "Gunma song."

[0088] Here, an example has been described in which the user inputs voice information such as "Play a song with a similar style to the song I listened to yesterday" in step S302. However, voice information such as "Play a song that is completely different from the song I listened to yesterday" may also be input to the user terminal 20. In this case, the recommendation information generation unit 142 generates output information for causing the user terminal 20 to play a song with a completely different style from "Gunma songs." This allows the user terminal 20 to play a song with a completely different style from "Gunma songs." Furthermore, the update unit 121 may increase the recommendation score of a song with a completely different style from "Gunma songs" recorded in the recommendation DB 112.

[0089] In this way, with the information processing device 10 and the user terminal 20 according to this embodiment, the user can have the output information related to "Gunma" output from the user terminal 20 without directly uttering the noun "Gunma." Furthermore, the update unit 121 can also update various pieces of information related to "Gunma" stored in the storage unit 110.

[0090] The above has described the interactions between the information processing device 10 and the user terminal 20 according to this embodiment. Next, the update history of the interaction DB 113, the knowledge DB 111, and the recommendation DB 112, which are constructed based on the above interactions, will be described with reference to Fig. 11. Fig. 11 is a diagram showing an example of information recorded in the interaction DB 113, the update history of the knowledge DB 111, and the update history of the recommendation DB 112 according to the first embodiment.

[0091] FIG. 11 shows five pieces of information (No. A to E) for each of the update history of the interaction DB 113, the knowledge DB 111, and the recommendation DB 112. For example, No. A of the interaction DB 113 records the time information "2018 / 11 / 22 8:00 PM" and the interaction information that a user input at home, "Song E is one of Gunma's favorite songs." The analysis unit 130 analyzes the word "Gunma" in the input information to mean the name of a person named "Iwasa." The update unit 121 then increases the probability that the word "Gunma" means "Iwasa." More specifically, the update unit 121 increases the probability that the word "Gunma" recorded in the knowledge DB 111 means "Iwasa" from 80% to 81%. As a result, the analysis unit 130 analyzes the word "Gunma" to mean the name (nickname) of a person named "Iwasa" with an 81% probability. Meanwhile, the update unit 121 reduces the probability that the word "Gunma" stored in the knowledge DB 111 means the prefecture name "Gunma" from 20% to 19%. As a result, the analysis unit 130 will analyze the word "Gunma" to mean the prefecture name "Gunma" with a probability of 19%. The update unit 121 also adds and records song E to the favorite playlist of "Gunma" in the recommendation DB 112.

[0092] Similarly to No. A, the records in the knowledge DB 111 and the recommendation DB 112 are updated according to the interaction information related to Nos. B to E. Specifically, No. B in the interaction DB 113 records the time information as "2018 / 11 / 28 8:01 PM" and the interaction information as "Play Gunma's favorite song" input from the user at home. Accordingly, in the knowledge DB 111, the probability that "Gunma" means "Gunma" has been updated from 19% to 18%, and the probability that "Gunma" means "Iwasa" has been updated from 81% to 82%.

[0093] Additionally, in No. C of the interaction DB 113, the time information is recorded as "2018 / 11 / 28 8:02 PM," and the interaction information is recorded as the user playing songs A, B, and E from Gunma's favorite playlist at home. Accordingly, in the knowledge DB 111, the probability that "Gunma" means "Gunma" is updated from 18% to 17%, and the probability that "Gunma" means "Iwasa" is updated from 82% to 83%.

[0094] Additionally, in No. D of the interaction DB 113, the time information is recorded as "2018 / 11 / 28 8:15 PM," and the interaction information is recorded as the input information of "Like" that the user input at home. The input information of "Like" is presumed to be a response to the playing of songs A, B, and E in No. C. Therefore, in the recommendation DB 112, the recommendation score of song A has been updated from 0.2 to 0.3, the recommendation score of song B from 0.6 to 0.7, and the recommendation score of song E from 0.0 to 0.5.

[0095] Furthermore, in No. E of the interaction DB 113, the time information is recorded as "2018 / 11 / 28 8:20 PM," and the interaction information is recorded as the user inputting the input information "My favorite manga character is Gunma" at home. Accordingly, in the knowledge DB 111, the probability that "Gunma" means "Gunma Prefecture" is updated from 17% to 7%, and the probability that "Gunma" means "Iwasa" is updated from 82% to 73%. Furthermore, the knowledge DB 111 adds the fact that "Gunma" may be a manga character, and the probability that "Gunma" means a manga character is updated from 0% to 20%.

[0096] As described above, various pieces of information stored in the information processing device 10 according to this embodiment are updated through interactions between the information processing device 10 and the user terminal 20. More specifically, the information stored in the interaction DB 113 is used as learning data, and as the learning data accumulates, the information stored in the knowledge DB 111 and the recommendation DB 112 is updated. The generation unit 140 generates output information based on the information stored in the knowledge DB 111 and the recommendation DB 112. Therefore, as the learning data accumulates, the algorithm by which the information processing device 10 outputs the output information changes.

[0097] As described above, the information recorded in the learning DB 114, which is learning data, affects the information recorded in the knowledge DB 111 or the recommendation DB 112. The learning data is generated, for example, through interactions with a user and stored in the learning DB 114. Therefore, the knowledge DB 111 and the recommendation DB 112, which store data generated based on learning data generated through interactions with a user, are databases suited to the user. A database suited to the user is, for example, a database that reflects the user's preferences and lifestyle.

[0098] For example, assume the situation shown in Fig. 12. By performing the above-described processing, a knowledge DB 111A and a recommendation DB 112A are constructed for user A. Also, by performing the above-described processing, a knowledge DB 111B and a recommendation DB 112B are constructed for user B.

[0099] The knowledge DB 111A and recommendation DB 112A of user A are constructed by being updated by the update unit 121A based on feedback from user A. The feedback is information provided by user A, such as a reaction or reply from user A to information generated by the generation unit 140A with reference to the recommendation DB 112A after the analysis unit 130A analyzes information input by voice from user A, for example, by referring to the knowledge DB 111A.

[0100] The knowledge DB 111A and recommendation DB 112A of user A thus constructed are databases that reflect the preferences and lifestyle of user A. Furthermore, information recommended using such databases can be information suitable for user A.

[0101] Similarly, the knowledge DB 111B and recommendation DB 112B of user B are constructed by being updated by the update unit 121B based on feedback from user B. The feedback is user B's reaction to information that is input from user B, for example, information input by voice, which is analyzed by the analysis unit 130B with reference to the knowledge DB 111B, and generated by the generation unit 140B with reference to the recommendation DB 112B.

[0102] The knowledge DB 111B and recommendation DB 112B of user B thus constructed are databases that reflect the preferences and lifestyle of user B. Furthermore, information recommended using such databases can be information suitable for user B.

[0103] Here, it is assumed that user B is someone whom user A admires or looks up to. For example, user B may be a celebrity such as an entertainer or athlete, and user A may wish to obtain various information about such a celebrity, such as what they are interested in and what they do at what times.

[0104] Here, it is assumed that user A lives his life using knowledge DB 111B and recommendation DB 112B of user B. If user A lives his life using knowledge DB 111B and recommendation DB 112B of user B, it is considered that user A can simulate the life of user B. For example, in situation A, user B can recommend to user A information that is recommended to user B, such as listening to song A, purchasing item B, ordering menu C, going to place D, and using service E.

[0105] If user A wants to be like user B, he or she can use the knowledge DB 111B and recommendation DB 112B of user B to simulate the life of user B. If user B is an athlete and user A wants to be an athlete like user B, for example, the music that user B listens to during training, meal menus, and the like can be recommended to user A by using user B's knowledge DB 111B and recommendation DB 112B.

[0106] Furthermore, user B does not have to be a specific person such as a celebrity. For example, the knowledge DB 111B and recommendation DB 112B may be of user B who has been accepted into university A. For example, when user A is studying to take the entrance exam for university A, which user B has been accepted into, by using user B's knowledge DB 111B and recommendation DB 112B, it becomes possible to recommend to user A what subjects user B studied at what time of day, what music user B listened to during breaks, and so on.

[0107] As will be described later, multiple knowledge DBs 111 can be integrated into one knowledge DB 111, and multiple recommendation DBs 112 can be integrated into one recommendation DB 112. By using this integration method, it is possible to create knowledge DBs 111 and recommendation DBs 112 for multiple users who have been accepted into University A. For example, this database can be sold and distributed as a University A acceptance database, and User A can use such a database to receive various recommendations to help him or her get closer to being accepted.

[0108] As an example, it is assumed that a user desires to use the knowledge DB 111 and recommendation DB 112 learned by other users. Therefore, the following will explain a case where the knowledge DB 111 and recommendation DB 112 learned by other users can be used as the user's own knowledge DB 111 and recommendation DB 112.

[0109] In the following explanation, as explained with reference to FIG. 12, a case will be explained in which knowledge DB111A and recommendation DB112A are constructed as databases for user A, and knowledge DB111B and recommendation DB112B are constructed as databases for user B, and user A uses the database for user B (the database for user B is reused as the database for user A).

[0110] <The first method of using another user's database> As a first method of using another user's database, a case will be described in which the user replaces his / her own recommendation DB 112 with the recommendation DB 112 of the other user, thereby using the other user's database as his / her own database.

[0111] 13 is a diagram illustrating the first reuse method. Recommendation DB 112A for user A is replaced with recommendation DB 112B for user B. This replacement allows user A to receive recommendations that reference recommendation DB 112B' for user B. Here, the replaced recommendation DB 112B is written with a dash to indicate that it has been replaced.

[0112] By performing such substitution, the information input by user A, for example by voice input, is analyzed by analysis unit 130A by referring to knowledge DB 111A constructed for user A, and information to be recommended to user A is generated by generation unit 140A by referring to recommendation DB 112B' constructed for user B.

[0113] In this way, recommendations are made to user A with reference to recommendation DB 112B' constructed for user B. Therefore, recommendations are made to user A that match the preferences, lifestyle, etc. of user B.

[0114] In this way, after recommendation DB 112A constructed for user A is replaced with recommendation DB 112B' constructed for user B, if user A says, for example, "Play some music," the analysis unit 130A analyzes that an instruction to play music has been given, and the generation unit 140A (its recommendation information generation unit 142) refers to recommendation DB 112B', selects, for example, a song with a high recommendation score, and generates output information for playing that song.

[0115] In this case, since the recommendation score is information generated based on the preferences of user B, songs that match the preferences of user B are presented to user A.

[0116] Furthermore, for example, if user A says, "Play a song by Gunma," the analysis unit 130A refers to the knowledge DB 111A and analyzes that "Gunma" is a friend of user A, and interprets the instruction "Play a song by Gunma" as an instruction to play a song that user A's friend likes. Based on this analysis result, the generation unit 140A (the recommendation information generation unit 142 thereof) refers to the recommendation DB 112B' and reads out a playlist of Gunma's favorite songs.

[0117] However, because recommendation DB 112B' is a database constructed for user B, it does not store a playlist of Gunma's favorite songs, which is information about user A. Therefore, generation unit 140A generates a message such as, for example, "There are no songs about Gunma" or "Please tell me your favorite songs about Gunma." Alternatively, after the database has been replaced, a message such as, "May I play user B's favorite songs?" or "There are no songs about Gunma, so I will play an alternative song" may be generated to let user A know about the replacement.

[0118] For example, if user A says, "Play user B's song," the analysis unit 130A analyzes that user B is a person and that the instruction is to play that person's favorite song. If the analysis unit 130A cannot determine that user B is a person, a message such as "Who is user B?" may be generated by processing of the subsequent generation unit 140A. In this case, the processing is performed in the same manner as the processing described with reference to the flowchart of FIG. 8, and as a result, the knowledge DB 111A is updated and information such as user B = person is written.

[0119] The generation unit 140A (recommended information generation unit 142) refers to the recommendation DB 112B', references the playlist of songs that User B likes, selects songs with high recommendation scores, and generates output information for playing those songs. In this way, songs that User B likes can be recommended to User A.

[0120] As described above, the recommendation DB 112B' may become a database suited to the preferences of user A over time (as learning progresses) through the processing of update unit 121A. In other words, there is a possibility that recommendation DB 112B' may return to a state close to recommendation DB 112A before replacement. Even though user A willingly replaced his / her own recommendation DB 112A with recommendation DB 112B' of user B, it may not be a desirable state for user A if recommendation DB 112B' returns to recommendation DB 112A for user A.

[0121] Therefore, when such a database replacement is performed, some kind of limitation may be placed on the updates by the update unit 121 so that the data stored in the database after the replacement is not updated frequently. For example, limitations may be placed such that updates are performed only when instructed (permitted) by the user, or that updates are not performed for a predetermined period after the replacement, for example, one week.

[0122] In this way, by replacing the recommendation DB 112 with the recommendation DB 112 of a user desired by the user, it becomes possible to receive recommendations using the recommendation DB 112 after replacement.

[0123] <Second method of using other users' databases> As a second method of using another user's database, a case where one's own knowledge DB 111 is replaced with another user's knowledge DB 111 will be described.

[0124] 14 is a diagram illustrating the second reuse method. User A's knowledge DB 111A is replaced with user B's knowledge DB 111B. By replacing the knowledge DB 111A with user B's knowledge DB 111B, user A can receive recommendations using the results of semantic analysis performed by referring to user B's knowledge DB 111B'. Here, the replaced knowledge DB 111B is written with a dash to indicate that it has been replaced.

[0125] By performing such substitution, the information input by user A, for example by voice input, is analyzed by analysis unit 130A by referring to knowledge DB 111B constructed for user B, and information to be recommended to user A is generated by generation unit 140A by referring to recommendation DB 112A constructed for user A.

[0126] In this way, semantic analysis is performed on user A with reference to the knowledge DB 111B' constructed for user B. Therefore, semantic analysis is performed on user A using knowledge obtained from the life and friendships of user B, and a recommendation is made using the results of the semantic analysis.

[0127] In this way, after the knowledge DB 111A constructed for user A is replaced with the knowledge DB 111A' constructed for user B, if user A says, for example, "Play some music," the analysis unit 130A analyzes that an instruction to play music has been given, and the generation unit 140A (its recommendation information generation unit 142) refers to the recommendation DB 112A, selects a song with a high recommendation score, and generates output information for playing that song.

[0128] For example, if user A says, "Play a song by Gunma," the analysis unit 130A performs semantic analysis by referring to the knowledge DB 111B'. Since the knowledge DB 111A before the replacement stores information that "Gunma" is a friend of user A, the analysis unit 130A analyzes the instruction as an instruction to play a song that "Gunma" is a friend of user A, but since such information is not stored in the knowledge DB 111B' after the replacement, the result that the analysis is not possible (a result that there is a contradiction in the instruction content) is output to the generation unit 140A.

[0129] The generation unit 140A generates a message such as, for example, "What is Gunma?" The process of generating such a message can be performed in the same manner as the process described with reference to FIG. 8. When such a process is performed and there is a reply (feedback) from user A, the knowledge DB 111B' is updated based on the content of the reply. Although such an update may be performed, there is a possibility that the post-replacement knowledge DB 111B' may return to a state close to the pre-replacement knowledge DB 111A.

[0130] Even though user A has voluntarily replaced his / her own knowledge DB 111A with user B's knowledge DB 111B', it may not be a desirable state for user A if the knowledge DB 111A is reverted to that for user A. Therefore, when such a database replacement is performed, some kind of restriction may be placed on the updates of update unit 121A so that the data stored in the replaced database is not updated frequently. For example, a restriction may be placed so that updates are performed only when instructed (permitted) by the user, or so that updates are not performed for a predetermined period after the replacement, for example, one week.

[0131] After the knowledge DB 111 is replaced, the generating unit 140A may generate a message to let the user A know that the knowledge DB 111 has been replaced. For example, a message such as "There is no information about Gunma, but there is information about user B" or a message such as "May I play user B's favorite song?" may be generated.

[0132] As yet another example, when user A says, "Play user B's song," the analysis unit 130A refers to the knowledge DB 111B', reads information about user B's favorite song, and analyzes the instruction as an instruction to play that song. The knowledge DB 111B' stores information that "user B's favorite song is YY." For example, since the processes described with reference to FIGS. 8 and 9 are also executed on the information processing device 10 on user B's side, the database constructed as the knowledge DB 111B for user B also stores information that "user B's favorite song is YY."

[0133] Furthermore, by replacing the knowledge DB 111B', the update unit 121A can update the recommendation DB 112A based on the information recorded in the knowledge DB 111B'. For example, based on the information "User B's favorite song is YY" recorded in the knowledge DB 111B', a playlist of songs that User B likes can be created in the recommendation DB 112A.

[0134] As a result of such updates, the playlist of songs that user B likes is also recorded in recommendation DB 112A, so generation unit 140A (recommendation information generation unit 142) refers to recommendation DB 112A, references the playlist of songs that user B likes, selects songs with high recommendation scores, and generates output information for playing those songs.

[0135] In this way, by replacing the knowledge DB 111 with the knowledge DB 111 of a user desired by the user, it is possible to provide recommendations using the replaced knowledge DB 111.

[0136] <Third method of using other users' databases> As a third method for utilizing other users' databases, a case where one's own recommendation DB 112 and other users' recommendation DB 112 are used in combination will be described.

[0137] 15 is a diagram illustrating the third reuse method. A recommendation DB 112B' for user B is added to the storage unit 110 (FIG. 4) of the information processing device 10 of user A. As a result, the storage unit 110 of user A stores a recommendation DB 112A constructed for user A and a recommendation DB 112B constructed for user B. A DB switching unit 301 is added so that either one of the two recommendation DBs 112 can be referenced.

[0138] The DB switching unit 301 may be provided as a part of the function of the generating unit 140, or may be provided between the storage unit 110 and the generating unit 140 in the configuration of the information processing device 10 shown in FIG.

[0139] User A can receive recommendations that refer to recommendation DB 112B' for user B. When DB switching unit 301 switches the database to be referred to to recommendation DB 112B', the processing is performed as described in the first diversion method described above. Therefore, information input by user A by voice input or the like is analyzed by analysis unit 130A by referring to knowledge DB 111A constructed for user A, and information to be recommended to user A can be generated by generation unit 140A by referring to recommendation DB 112B' constructed for user B.

[0140] Furthermore, when the DB switching unit 301 switches the database to be referred to to the recommendation DB 112A, the processing is as described with reference to the flowcharts of Figures 8 to 10. Therefore, information input by user A by voice input or the like is analyzed by the analysis unit 130A by referring to the knowledge DB 111A constructed for user A, and information to be recommended to user A can be generated by the generation unit 140A by referring to the recommendation DB 112A constructed for user A.

[0141] In this way, recommendations are made to user A with reference to recommendation DB 112A constructed for user A or recommendation DB 112B' constructed for user B. Therefore, recommendations are made to user A that match information such as the preferences and lifestyles of each of user A and user B.

[0142] The DB switching unit 301 can be configured to switch when the location of user A changes significantly, for example, when the user A goes on a trip or moves. For example, when user A goes on a trip to area B, the DB is switched to recommendation DB 112B' constructed for user B who lives in area B. Then, recommended information is generated by referring to recommendation DB 112B'. In this case, it becomes possible to recommend to user A information closely related to area B, such as restaurants that user B regularly uses or places to play that user B frequently visits.

[0143] Furthermore, the DB switching unit 301 can be switched depending on, for example, time periods. For example, if user A is a student taking an exam and user B is a successful candidate at a school that user A wants to get into, for example, during the evening hours, the reference database is switched to recommendation DB 112B' constructed for user B, and user B can recommend to user A what subjects he studied and how he studied during the evening hours.

[0144] Furthermore, the DB switching unit 301 may be switched based on the analysis results of the analysis unit 130A, for example. For example, when user A says, "Play a song by Gunma," the analysis unit 130A refers to the knowledge DB 111A, analyzes that "Gunma" is a friend of user A, and interprets this as an instruction to play a song that user A's friend "Gunma" likes. Based on this analysis result, the generation unit 140A (the recommendation information generation unit 142 thereof) refers to the recommendation DB 112B' and reads out a playlist of songs that Gunma likes.

[0145] However, because recommendation DB 112B' is a database constructed for user B, it does not store a playlist of Gunma's favorite songs, which is information about user A. In such a case, DB switching unit 301 switches the reference database to recommendation DB 112A. By switching the reference database to recommendation DB 112A, songs that Gunma likes and have high recommendation scores are selected, and output information for playing those songs is generated.

[0146] The timing of switching by the DB switching unit 301 may of course be other than the above, and the above example is merely an example and is not intended to be limiting.

[0147] <The fourth method of using other users' databases> As a fourth method of using other users' databases, a case where one's own knowledge DB 111 and other users' knowledge DB 111 are used in combination will be described.

[0148] 16 is a diagram for explaining the fourth reuse method. A knowledge DB 111B' for user B is added to the storage unit 110 (FIG. 4) of the information processing device 10 of user A. As a result, a knowledge DB 111A constructed for user A and a knowledge DB 111B constructed for user B are stored in the storage unit 110 of user A. A DB switching unit 302 is added so that either one of the two knowledge DBs 111 can be switched and referenced.

[0149] The DB switching unit 302 may be provided as a part of the function of the generating unit 140, or may be provided between the storage unit 110 and the generating unit 140 in the configuration of the information processing device 10 shown in FIG.

[0150] User A can receive recommendations based on the analysis results obtained by referring to the knowledge DB 111B' for user B. When the DB switching unit 302 switches the database to be referred to to the knowledge DB 111B', the processing is performed as described in the second diversion method described above. Therefore, information input by user A by voice input or the like is analyzed by the analysis unit 130A by referring to the knowledge DB 111B' constructed for user B, and information to be recommended to user A can be generated by the generation unit 140A by referring to the knowledge DB 111A constructed for user A.

[0151] Furthermore, when the DB switching unit 302 switches the database to be referred to to the knowledge DB 111A, the processing is as described with reference to the flowcharts of Figures 8 to 10. Therefore, information input by user A by voice input or the like is analyzed by the analysis unit 130A by referring to the knowledge DB 111A constructed for user A, and information to be recommended to user A can be generated by the generation unit 140A by referring to the knowledge DB 111A constructed for user A.

[0152] In this way, semantic analysis is performed with reference to the knowledge DB 111A constructed for user A or the knowledge DB 111B' constructed for recommended user B, and recommendations using the results of the semantic analysis are made to user A. Therefore, recommendations are made to user A that match information such as the preferences and lifestyles of each of user A and user B.

[0153] The DB switching unit 302 can be configured to switch when user A's location changes significantly, for example, when the user goes on a trip or moves. For example, when user A goes on a trip to area B, the DB switching unit 302 switches to the knowledge DB 111B' constructed for user B who lives in area B. Then, the knowledge DB 111B' may be referenced to perform semantic analysis, and information to be recommended may be generated based on the analysis results. In this case, because information such as restaurants that user B who lives in area B regularly uses and places to play that he or she frequently visits is recorded in the knowledge DB 111B', semantic analysis suited to information specific to area B may also be performed, and recommendations based on the semantic analysis may be made to user A.

[0154] Furthermore, the DB switching unit 302 can be configured to switch based on, for example, time periods. For example, if user A is a student taking an exam and user B is a successful candidate at a school that user A wants to get into, for example, during the evening hours, the knowledge DB 111B' is switched to the knowledge DB 111B' constructed for user B. If information about songs that user B listened to during the evening hours is stored in the knowledge DB 111B', for example, when user A wants to listen to a song during a break from studying, the information about the song that user B listened to can be read from the knowledge DB 111B' and recommended to user A.

[0155] The DB switching unit 302 may also be configured to switch when, for example, a contradiction occurs in the analysis by the analysis unit 130A. For example, if user A says, "Play a song about Gunma," and the DB switching unit 302 has switched to refer to the knowledge DB 111B', it cannot be analyzed that "Gunma" is a friend of user A, and therefore the instruction "Play a song about Gunma" is analyzed to contain a contradiction. When such a contradiction is found in the analysis, the DB switching unit 302 switches the referenced database to refer to the knowledge DB 111A.

[0156] By referring to the knowledge DB 111A, the analysis unit 130A can analyze that "Gunma" is a friend of user A, and can interpret the instruction "Play a song by Gunma" as an instruction to play a song that is liked by a friend of user A. Based on this analysis result, the generation unit 140A (the recommendation information generation unit 142 thereof) refers to the knowledge DB 111A and reads out a playlist of songs that Gunma likes.

[0157] The timing of switching by the DB switching unit 302 may of course be other than the above, and the above example is merely an example and is not intended to be limiting.

[0158] <The fifth method of using other users' databases> As a fifth method of using other users' databases, a case where one's own recommendation DB 112 and the recommendation DB 112 of another user are integrated will be described.

[0159] 17 is a diagram for explaining the fifth reuse method, in which a recommendation DB 112A for user A and a recommendation DB 112B for user B are integrated to generate a recommendation DB 112AB for user A.

[0160] By performing such integration, information input by user A, such as by voice input, is analyzed by analysis unit 130A by referring to knowledge DB 111A constructed for user A, and information to be recommended to user A is generated by generation unit 140A by referring to recommendation DB 112AB, which is an integration of recommendation DB 112A constructed for user A and recommendation DB 112B constructed for user B.

[0161] In this way, recommendations are made to user A by referencing recommendation DB 112AB, which is an integration of different databases. Therefore, recommendations are made to user A that match the preferences and lifestyles of users A and B, respectively.

[0162] The databases are integrated, for example, by the method described with reference to Fig. 18. As described above, the recommendation DB 112 stores, for example, a playlist of songs, and the playlist contains recommendation scores. Here, the explanation will be continued by taking an example in which recommendation scores are integrated.

[0163] The left diagram in Fig. 18 shows an example of recommendation scores recorded in the recommendation DB 112A and the recommendation DB 112B before integration, and the right diagram in Fig. 18 shows an example of recommendation scores recorded in the recommendation DB 112AB after integration.

[0164] In recommendation DB 112A of user A before integration, the recommendation score for song A is "0.2", the recommendation score for song B is "0.9", the recommendation score for song C is "0.4", and the recommendation score for song D is "N / A (not applicable)". In recommendation DB 112B of user B before integration, the recommendation score for song A is "1.0", the recommendation score for song B is "0.5", the recommendation score for song C is "N / A", and the recommendation score for song D is "0.3".

[0165] When integrating these recommendation scores, one method is to use the average of the recommendation scores as the integrated recommendation score. The integrated value is the average value of the integrated recommendation DB 112AB shown in the right diagram of Fig. 18. First, the recommendation score for song A is "0.6", which is the average of "0.2" and "1.0". Similarly, the recommendation score for song B is "0.7", which is the average of "0.9" and "0.5".

[0166] The recommendation score for song C is the average of "0.4" and "N / A", but "N / A" may be calculated as "0" and the value may be "0.2". Alternatively, if there is an "N / A", the value other than "N / A", in this case "0.4", may be reflected as is. Here, the explanation will continue using an example in which the value other than "N / A" is reflected as is.

[0167] The recommendation score for song D is the average of "N / A" and "0.3", but since the value "0.3" that is not "N / A" is reflected as is, it becomes "0.3".

[0168] In this way, when the average value is used as the integrated score, if a score exists, the average value is calculated and that value is used as the integrated score, and if one score does not exist, the other score is used as the integrated score.

[0169] This technology can be applied when integrating two recommendation DBs 112, but it can also be applied when integrating two or more recommendation DBs 112. When integrating multiple recommendation DBs 112, an average value is calculated using only scores other than N / A among the scores associated with a specific song. Also, if there is only one score other than N / A among the scores associated with a specific song, that score is used as is.

[0170] Another method for integrating recommendation scores is to prioritize one's own score. In this case, user A is the "you" and user B is the "other." For a given song, if there is a recommendation score in recommendation DB 112A constructed for oneself (user A) and there is a recommendation score in recommendation DB 112B constructed for the other (user B), the recommendation score written in recommendation DB 112A constructed for oneself is used as the integrated recommendation score.

[0171] Furthermore, even if there is a recommendation score for a specific song in the recommendation DB 112A constructed for the user (user A) and there is no recommendation score in the recommendation DB 112B constructed for the other user (user B), the recommendation score written in the recommendation DB 112A constructed for the user is used as the recommendation score after integration. Furthermore, if there is no recommendation score for a specific song in the recommendation DB 112A constructed for the user (user A) and there is a recommendation score in the recommendation DB 112B constructed for the other user (user B), the recommendation score written in the recommendation DB 112B constructed for the other user is used as the recommendation score after integration.

[0172] The integrated value with self-priority in the integrated recommendation DB 112AB shown in the right diagram of Figure 18 is referenced. First, for song A, since there is a score in both your recommendation DB 112A and the other person's recommendation DB 112B, the recommendation score of "0.2" written in your recommendation DB 112A is used as the integrated recommendation score. Similarly, the recommendation score for song B becomes "0.9".

[0173] The recommendation score for song C is "0.4" in your recommendation DB 112A, but there is no corresponding score in the other person's recommendation DB 112B, so the score "0.4" is used as the recommendation score after integration. The recommendation score for song D is "0.3" in your recommendation DB 112A, but there is no corresponding score in the other person's recommendation DB 112B, so the score "0.3" is used as the recommendation score after integration.

[0174] In this way, the score recorded in one's own recommendation DB 112A may be preferentially used as the integrated score. In this case, a recommendation DB 112AB that is close to one's own preferences and lifestyle is constructed. Also, a recommendation DB 112AB is constructed in which the preferences and lifestyle of the other party (user B), which are not included in one's own preferences and lifestyle, are added.

[0175] This technology can be applied when integrating two recommendation DBs 112, but can also be applied when integrating two or more recommendation DBs 112. When integrating multiple recommendation DBs 112, if a score associated with a specific song is written in one's own recommendation DB 112, that score is used as the recommendation score after integration, and if no score is written in one's own recommendation DB 112, the score written in another user's recommendation DB 112 is used as the recommendation score.

[0176] Another method for integrating recommendation scores is to prioritize the other user's score. For a given song, if there is a recommendation score in recommendation DB 112A created for the user (user A) and there is a recommendation score in recommendation DB 112B created for the other user (user B), the recommendation score written in recommendation DB 112B created for the other user is used as the integrated recommendation score.

[0177] Furthermore, if there is a recommendation score for a specific song in the recommendation DB 112A constructed for the user (user A) and there is no recommendation score in the recommendation DB 112B constructed for the other user (user B), the recommendation score written in the recommendation DB 112A constructed for the user is used as the integrated recommendation score. Furthermore, if there is no recommendation score for a specific song in the recommendation DB 112A constructed for the user (user A) and there is a recommendation score in the recommendation DB 112B constructed for the other user (user B), the recommendation score written in the recommendation DB 112B constructed for the other user is used as the integrated recommendation score.

[0178] The integrated values ​​in the integrated recommendation DB 112AB shown on the right side of Figure 18 are referenced. First, for song A, since there is a score in both your recommendation DB 112A and the other person's recommendation DB 112B, the recommendation score of "1.0" written in the other person's recommendation DB 112B is used as the integrated recommendation score. Similarly, the recommendation score for song B becomes "0.5".

[0179] The recommendation score for song C is "0.4" in your recommendation DB 112A, but there is no corresponding score in the other person's recommendation DB 112B, so the score "0.4" is used as the recommendation score after integration. The recommendation score for song D is "0.3" in your recommendation DB 112A, but there is no corresponding score in the other person's recommendation DB 112B, so the score "0.3" is used as the recommendation score after integration.

[0180] In this way, the score recorded in the recommendation DB 112A of the other person may be preferentially used as the integrated score. In this case, a recommendation DB 112AB that is close to the other person's preferences and lifestyle is constructed. Also, a recommendation DB 112AB that retains the user's preferences and lifestyle that are not included in the preferences and lifestyle of the other person (user B) is constructed.

[0181] This technology can be applied when integrating two recommendation DBs 112, but can also be applied when integrating two or more recommendation DBs 112. When integrating multiple recommendation DBs 112, if a score associated with a specific song is written in the recommendation DB 112 for the other person, that score is used as the recommendation score after integration, and if no score is written in the recommendation DB 112 for the other person, the score written in the recommendation DB 112 for oneself is used as the recommendation score after integration.

[0182] As another method for integrating recommendation scores, although not shown, the recommendation with the higher score may be selected. Alternatively, the recommendation with the lower score may be selected. Furthermore, information that is not recorded in one's own recommendation DB 112, for example, information that has a score of "N / A" in FIG. 18, may not be recorded in the integrated recommendation DB 112.

[0183] Furthermore, the integration may be performed based on other calculation methods or rules not illustrated here.

[0184] <The sixth method of using other users' databases> As a sixth method of using other users' databases, a case where one's own knowledge DB 111 is integrated with another user's knowledge DB 111 will be described.

[0185] 19 is a diagram for explaining the sixth reuse method, in which a knowledge DB 111A for user A and a knowledge DB 111B for user B are integrated to generate a knowledge DB 111AB for user A.

[0186] By performing such integration, information input by user A by voice input or the like is analyzed by analysis unit 130A by referring to knowledge DB111AB, which is an integration of knowledge DB111A constructed for user A and knowledge DB111B constructed for user B, and information to be recommended to user A is generated by generation unit 140A by referring to knowledge DB111A constructed for user A.

[0187] In this way, semantic analysis is performed with reference to the knowledge DB 111AB, which is an integration of different databases, and recommendations based on the analysis results are made to user A. Therefore, recommendations are made to user A that match the preferences and lifestyles of users A and B, respectively.

[0188] The databases are integrated, for example, by the method described with reference to Fig. 20. As described above, for example, occurrence probability values ​​of predetermined words are recorded in the knowledge DB 111. Here, the explanation will be continued by taking an example in which probability values ​​are integrated.

[0189] The left diagram in Fig. 20 shows an example of probability values ​​recorded in the knowledge DB 111A and the knowledge DB 111B before integration, and the right diagram in Fig. 20 shows an example of probability values ​​recorded in the knowledge DB 111AB after integration.

[0190] In user A's knowledge DB 111A before the integration, the probability value of "Gunma → Gunma" is "8%, the probability value of "Gunma → manga character" is "20%, the probability value of "Gunma → Iwasa-san" is "72%, and the probability value of "Gunma → celebrity A-san" is "N / A (not applicable)". In user B's knowledge DB 111B before the integration, the probability value of "Gunma → Gunma" is "80%, the probability value of "Gunma → manga character" is "12%, the probability value of "Gunma → Iwasa-san" is "N / A", and the probability value of "Gunma → celebrity A-san" is "8%".

[0191] When integrating these probability values, one method is to use the average value of the probability values ​​as the integrated probability value. Refer to the integrated value using the average value of the integrated knowledge DB111AB shown in the right diagram of Figure 20. First, the probability value of "Gunma → Gunma" is "44", which is the average of "8" and "80". Similarly, the probability value of "Gunma → manga character" is "16", which is the average of "20" and "12".

[0192] The probability value for "Gunma → Iwasa-san" is the average of "72" and "N / A", but "N / A" is calculated as "0", so the result is "36". The probability value for "Gunma → Celebrity A-san" is the average of "N / A" and "8", which is "4".

[0193] In this way, the average value of the probability values ​​is calculated, and this value is used as the integrated probability value.

[0194] This technology can be applied when integrating two knowledge DBs 111, but it can also be applied when integrating two or more knowledge DBs 111. When integrating multiple knowledge DBs 111, the average value of the probability values ​​associated with predetermined information is used as the probability value after integration.

[0195] Another method for integrating probability values ​​is to prioritize one's own probability value. In this case, user A is the user and user B is the other user. For certain information, if there is a probability value in the knowledge DB 111A constructed for oneself (user A) and there is a probability value in the knowledge DB 111B constructed for the other user (user B), the probability value written in the knowledge DB 111A constructed for oneself is used as the integrated probability value.

[0196] Furthermore, if there is a probability value for a predetermined piece of information in the knowledge DB 111A constructed for the user (user A) but not in the knowledge DB 111B constructed for the other party (user B), the probability value written in the knowledge DB 111A constructed for the user (user A) is used as the probability value after integration. Furthermore, if there is no probability value for a predetermined piece of information in the knowledge DB 111A constructed for the user (user A) but there is a probability value in the knowledge DB 111B constructed for the other party (user B), the probability value written in the knowledge DB 111B constructed for the other party is used as the probability value after integration.

[0197] The value integrated with priority given to the user in the integrated knowledge DB 111AB shown on the right side of Figure 20 is referenced. First, for "Gunma → Gunma," since there is a probability value in both the user's knowledge DB 111A and the other user's knowledge DB 111B, the probability value "8" written in the user's knowledge DB 111A is used as the integrated probability value. Similarly, the probability value for "Gunma → manga character" becomes "20."

[0198] The probability value of "Gunma → Iwasa-san" is written as "72" in one's knowledge DB 111A, but there is no corresponding probability value in the other person's knowledge DB 111B, so the probability value with "72" is used as the probability value after integration. The probability value of "Gunma → Celebrity A" is written as "8" in one's knowledge DB 111A, but there is no corresponding probability value in the other person's knowledge DB 111B, so the probability value with "8" is used as the probability value after integration.

[0199] In this way, the probability value stored in the user's own knowledge DB 111A may be used preferentially as the integrated probability value. In this case, the knowledge DB 111AB is constructed to have the user's preferences and lifestyle similar to the user's own preferences and lifestyle. In addition, the knowledge DB 111AB is constructed to have the user's preferences and lifestyle, which are not included in the user's own preferences and lifestyle, added to the user's own preferences and lifestyle.

[0200] In the example shown in Figure 20, if you add up the probability values ​​of "Gunma → Gunma," "Gunma → Manga character," "Gunma → Iwasa-san," and "Gunma → Celebrity A," the result is "108" (= 8 + 20 + 72 + 8). Each value may be adjusted so that the sum of these probability values ​​becomes 100 (%).

[0201] This technology can be applied when integrating two knowledge DBs 111, but can also be applied when integrating two or more knowledge DBs 111. When integrating multiple knowledge DBs 111, if a probability value associated with predetermined information is written in one's own knowledge DB 111, that probability value is used as the integrated probability value, and if no probability value is written in one's own knowledge DB 111, the probability value written in another user's knowledge DB 111 is used as the probability value.

[0202] Another method for integrating probability values ​​is to prioritize the other party's probability value. For certain information, if there is a probability value in the knowledge DB 111A constructed for the user (user A) and there is also a probability value in the knowledge DB 111B constructed for the other party (user B), the probability value written in the knowledge DB 111B constructed for the other party is used as the integrated probability value.

[0203] Furthermore, if there is a probability value for a given piece of information in the knowledge DB 111A constructed for the user (user A) and there is no probability value in the knowledge DB 111B constructed for the other person (user B), the probability value written in the knowledge DB 111A constructed for the user (user A) is used as the probability value after integration. Furthermore, if there is no probability value for a given piece of information in the knowledge DB 111A constructed for the user (user A) and there is a probability value in the knowledge DB 111B constructed for the other person (user B), the probability value written in the knowledge DB 111B constructed for the other person is used as the probability value after integration.

[0204] The values ​​integrated with priority given to the other person in the integrated knowledge DB 111AB shown in the right diagram of Figure 20 are referenced. First, for "Gunma → Gunma," since there is a probability value in both the user's knowledge DB 111A and the other person's knowledge DB 111B, the probability value "80" written in the other person's knowledge DB 111B is used as the integrated probability value. Similarly, the probability value for "Gunma → manga character" becomes "12."

[0205] The probability value of "Gunma → Iwasa-san" is written as "72" in one's knowledge DB 111A, but there is no corresponding probability value in the other person's knowledge DB 111B, so the probability value with "72" is used as the probability value after integration. The probability value of "Gunma → Celebrity A" is written as "8" in one's knowledge DB 111A, but there is no corresponding probability value in the other person's knowledge DB 111B, so the probability value with "8" is used as the probability value after integration.

[0206] In this way, the probability value stored in the other person's knowledge DB 111A may be used preferentially as the probability value after integration. In this case, a knowledge DB 111AB that is close to the other person's preferences and lifestyle is constructed. Also, a knowledge DB 111AB that retains the user's preferences and lifestyle that are not included in the preferences and lifestyle of the other person (user B) is constructed.

[0207] In the example shown in Figure 20, if you add up the probability values ​​of "Gunma → Gunma," "Gunma → Manga character," "Gunma → Iwasa-san," and "Gunma → Celebrity A," the result is "172" (= 80 + 12 + 72 + 8). Each value may be adjusted so that the sum of these probability values ​​becomes 100 (%).

[0208] This technology can be applied when integrating two knowledge DBs 111, but can also be applied when integrating two or more knowledge DBs 111. When integrating multiple knowledge DBs 111, if a probability value associated with predetermined information is written in the other person's knowledge DB 111, that probability value is used as the integrated probability value, and if no probability value is written in the other person's knowledge DB 111, the probability value written in one's own knowledge DB 111 is used as the integrated probability value.

[0209] As another method for integrating probability values, although not shown, the higher probability value may be selected. Alternatively, the lower probability value may be selected. Furthermore, information that is not recorded in the user's own knowledge DB 111, for example, information that has a probability value of "N / A" in FIG. 20, may not be recorded in the integrated knowledge DB 111.

[0210] Furthermore, the integration may be performed based on other calculation methods or rules not illustrated here.

[0211] <Configuration of information processing device according to the second embodiment> The configuration of the information processing device 11 according to the second embodiment will be described with reference to Fig. 21. Fig. 21 is a functional block diagram showing the configuration of the information processing device 11 according to the second embodiment. As shown in Fig. 21, the information processing device 11 includes a storage unit 118, a processing unit 128, an analysis unit 130, a generation unit 143, an output control unit 150, and a communication control unit 160. The storage unit 118, the processing unit 128, and the generation unit 143, which are different from the functional units included in the information processing device 10 (Fig. 2) according to the first embodiment, will be described below.

[0212] First, the storage unit 118 according to the second embodiment will be described with reference to Fig. 22. Fig. 22 is a functional block diagram showing the configuration of the storage unit 118 according to the second embodiment. The storage unit 118 according to the second embodiment includes a knowledge DB 116, a recommendation DB 117, an interaction DB 113, and a learning DB 114.

[0213] Moreover, the knowledge DB 116 according to the second embodiment does not store data that probabilistically represents the meaning of words, as in the knowledge DB 111 according to the first embodiment. Furthermore, the recommendation DB 117 according to the second embodiment does not store recommendation scores that are stored in the recommendation DB 112 according to the first embodiment.

[0214] In the second embodiment, analysis of the meaning of words or recommendation to a user is performed based on black-box parameters (hereinafter also simply referred to as "parameters") held by an analysis unit 131 and a generation unit 143, which will be described later. Therefore, in the second embodiment, the information recorded in the knowledge DB 116 and recommendation DB 117 according to the second embodiment is different from the information recorded in the knowledge DB 111 and recommendation DB 112 according to the first embodiment.

[0215] More specifically, the analysis unit 131 or the recommendation information generation unit 145 inputs input values ​​to a network in which an input layer consisting of multiple inputs and an output layer consisting of multiple outputs are connected by intermediate layers, and outputs output values ​​related to analysis results, recommendation information, etc. Hereinafter, parameters that define the weights of nodes in the network are referred to as "black-box parameters."

[0216] The learning DB 114 records input values ​​and output values ​​for events in which learning data is recorded. The input values ​​are various types of information required to obtain output values, such as user instructions, user status, and environmental information. The output values ​​can be, for example, indicators for estimating how appropriate the analysis results were (e.g., information such as the user's reaction). The learning data can be recorded based on a user instruction or automatically by the information processing device 11 in the background.

[0217] In this embodiment, the exchange DB 113 stores indexes of events as data. Furthermore, the exchange DB 113 stores label information (e.g., information indicating the time of occurrence of an event, the content of input information or output information, etc.) for extracting learning data required for relearning the algorithm recorded in the event. Therefore, learning data can be extracted from the learning DB 114 based on the exchange DB 113.

[0218] The information stored in storage unit 118 according to the second embodiment will be described, focusing on the differences from the information stored in storage unit 110 according to the first embodiment, with reference to Fig. 23. Fig. 23 is a diagram showing an example of information recorded in exchange DB 113, an update history of knowledge DB 116, and an update history of recommendation DB 117 according to the second embodiment.

[0219] As shown in Fig. 23, information on the probability of semantic content, recommendation scores, and the like are not recorded in the knowledge DB 116 and the recommendation DB 117. Therefore, as shown in Fig. 23, the update history of semantic content or the update history of recommendation scores, as shown in Fig. 11, are not stored in the storage unit 118 according to the second embodiment. Note that, like the exchange DB 113 according to the first embodiment, information on exchanges is recorded in the exchange DB 113 according to the second embodiment.

[0220] Next, the processing unit 128 according to the second embodiment will be described with reference to Fig. 24. Fig. 24 is a functional block diagram showing the configuration of the processing unit 128 according to the second embodiment. The processing unit 128 according to the second embodiment includes a learning unit 125 in addition to the functional units included in the processing unit 120 according to the first embodiment.

[0221] The learning unit 125 has a function of performing learning (for example, reinforcement learning) of various parameters possessed by the analysis unit 131 or the generation unit 143. More specifically, the learning unit 125 performs parameter learning based on the interaction information recorded in the interaction DB 113, for example, based on a technique such as reinforcement learning. As a result, the parameters are updated.

[0222] Here, parameter learning refers to optimizing black-box parameters in accordance with the accumulation of input values ​​and output values ​​(i.e., learning data). Note that the learning unit 125 may perform parameter learning when information related to an interaction recorded in the interaction DB 113 is added, deleted, or modified.

[0223] For learning, various machine learning techniques using neural networks such as RNN (Recurrent Neural Network) and CNN (Convolutional Neural Network) may be used.

[0224] When the information processing device 11 according to the second embodiment acquires input information, it generates output information based on various parameters held by the analysis unit 131 or the generation unit 143. Therefore, updating the parameters by the learning unit 125 corresponds to causing the information processing device 11 to re-learn the algorithm for generating the output information.

[0225] Next, a description will be given of the analysis unit 131. In the second embodiment, what correspondence relationship between input information and output information will result in a positive FB is learned as a black-box parameter (semantic analysis parameter).

[0226] Unlike the analysis unit 130 according to the first embodiment, the analysis unit 131 according to the second embodiment does not calculate the correspondence between input information and semantic content in the form of a probability, but instead uses the semantic analysis parameters to find the optimal correspondence from the preceding and following contexts / situations. More specifically, when voice information is input from the user, the analysis unit 131 uses the voice information as an input value, for example, and outputs the semantic analysis result of the voice information based on the semantic analysis parameters.

[0227] Furthermore, the analysis unit 131 may take as input values ​​various information (user situation, characteristic information, environmental information, content of user's instructions, etc.) in addition to the voice, and output the results of semantic analysis of the voice information. Here, the user's characteristic information may be information about the user's characteristics, such as age, sex, or address. Furthermore, the environmental information may be information about the space in which the user exists, such as information about time, location, or people who are with the user.

[0228] For example, if the user has been talking about friendships until just before, the analysis unit 131 may analyze "Gunma" to mean a person named "Iwasa." Also, if the user is traveling to Gunma prefecture, the analysis unit 131 may analyze "Gunma" to mean the name of the prefecture "Gunma." In this way, the content of the conversation until just before may be reflected in the semantic analysis parameters. Also, the content of the location may be reflected in the semantic analysis parameters. In this embodiment, the semantic analysis results output by the analysis unit 131 are used as input values ​​for the recommendation information generation unit 145, which will be described later, to generate recommendation information.

[0229] Next, the generation unit 143 will be described with reference to Fig. 25. Fig. 25 is a functional block diagram showing the configuration of the generation unit 143 according to the second embodiment. The generation unit 143 shown in Fig. 25 includes a confirmation information generation unit 144 and a recommendation information generation unit 145, similar to the generation unit 140 according to the first embodiment. As described above, the recommendation information generation unit 142 according to the first embodiment generates output information for recommending songs based on, for example, the recommendation score of the songs. On the other hand, in the second embodiment, the generation unit 143 learns, as black-box parameters (recommendation parameters), which song recommendation information should be generated to obtain positive feedback.

[0230] The recommendation information generation unit 145 generates optimal output information based on the recommendation parameters, for example, on the context or situation, and recommends, for example, music to the user. More specifically, the recommendation information generation unit 145 generates recommendation information as output based on the analysis results by the analysis unit 130, various information (user situation, characteristic information, environmental information, content of user instructions, etc.), and the recommendation parameters.

[0231] In this embodiment, the semantic analysis or generation of recommendation information uses voice information and the various types of information described above (user situation, characteristic information, environmental information, content of user instructions, etc.) For this reason, machine learning technology that can perform processing taking into account many conditions is suitable for processing that requires performing processing such as recommendation based on a variety of conditions, as in this embodiment.

[0232] For example, a user may prefer to play a song similar to a song that has appeared in the user's most recent conversation. In this case, the recommendation parameters may reflect the content of the song that has appeared in the user's most recent conversation, and the recommendation information generating unit 145 may generate output information based on the recommendation parameters for playing a song similar to the song that the user has been talking about up until that point.

[0233] In addition, if a certain song has been played for the user in the past, the user may prefer that the song be played again, for example, one week or more after the song was played. In this case, for example, information about the previously recommended song is reflected in the recommendation parameters, and the recommendation information generation unit 145 can generate output information that recommends the song if one week or more has passed since the previously recommended song was played.

[0234] <Processing example> First, a parameter update process in which the information processing device 11 updates black-box parameters (semantic analysis parameters and recommendation parameters) will be described with reference to Fig. 26. Fig. 26 is a flowchart illustrating an example of the parameter update process according to the second embodiment.

[0235] The information processing device 11 acquires input information (step S502). For example, the information processing device 11 receives information input to the user terminal 20 via the network 30. The received input information is transmitted to the analysis unit 131 via the communication control unit 160.

[0236] The analysis unit 131 analyzes the semantics of the input information (step S504). More specifically, the analysis unit 131 analyzes the semantics of the input information based on the semantic analysis parameters stored in the storage unit 118. The analysis result is transmitted to the generation unit 143.

[0237] The generation unit 143 generates output information (step S506). More specifically, the recommendation information generation unit 145 generates output information for making various recommendations to the user based on the analysis result and the recommendation parameters stored in the storage unit 118. The output information is transmitted to the output control unit 150.

[0238] The output control unit 150 outputs the output information (step S508). More specifically, the output control unit 150 transmits the output information to the communication control unit 160. The output information is transmitted to, for example, the user terminal 20 connected to the network 30. As a result, the user terminal 20 outputs the output information. For example, the user terminal 20 outputs a sound recommending a predetermined song to the user.

[0239] The information processing device 11 acquires FB (feedback) (step S510). For example, the information processing device 11 acquires a response to the output result from the user as FB. The acquired FB is transmitted to the processing unit 128.

[0240] Next, the processing unit 128 learns the semantic analysis parameters and the recommendation parameters (step S512). Specifically, the learning unit 125 learns the semantic analysis parameters and the recommendation parameters stored in the storage unit 118 based on the feedback from the user. This updates the semantic analysis parameters and the recommendation parameters. Once these parameters are updated, the parameter update process ends.

[0241] The parameter update process has been described above with reference to Fig. 26. In this way, various parameters used by the information processing device 11 to generate output information are updated based on, for example, an FB by a user, and thereby the information processing device 11 can generate output information more desired by the user.

[0242] As described above, the semantic analysis parameters and recommendation parameters are learned (updated) based on feedback from the user. Therefore, the knowledge DB 116 and recommendation DB 117 that store the semantic analysis parameters and recommendation parameters generated through interactions with the user become databases suited to the user. A database suited to the user is, for example, a database that reflects the user's preferences and lifestyle.

[0243] For example, assume the situation shown in Fig. 27. By performing the above-described processing, a knowledge DB 116A and a recommendation DB 117A are constructed for user A. Also, by performing the above-described processing, a knowledge DB 116B and a recommendation DB 117B are constructed for user B.

[0244] The knowledge DB 116A and recommendation DB 117A of user A are constructed by the learning unit 125A by being updated based on feedback from user A. The feedback is user A's reaction to information that is input from user A, for example, information input by voice, which is analyzed by the analysis unit 131A with reference to the knowledge DB 116A, and generated by the generation unit 143A with reference to the recommendation DB 117A.

[0245] The knowledge DB 116A and recommendation DB 117A of user A constructed in this way are databases that reflect the preferences and lifestyle of user A. Furthermore, information recommended using such databases can be information suitable for user A.

[0246] Similarly, knowledge DB 116B and recommendation DB 117B of user B are constructed by learning unit 125B by being updated based on feedback from user B. The feedback is user B's reaction to information that is input from user B, for example, information input by voice, which is analyzed by analysis unit 131B with reference to knowledge DB 116B, and generated by generation unit 143B with reference to recommendation DB 117B.

[0247] The knowledge DB 116B and recommendation DB 117B of user B thus constructed are databases that reflect the preferences and lifestyle of user B. Furthermore, information recommended using such databases can be information suited to user B. This is the same as in the case described with reference to FIG. 12 in the first embodiment.

[0248] As described with reference to FIG. 12, also in the second embodiment, when user B is someone whom user A admires or looks up to, user A may wish to obtain various information such as what such a celebrity is interested in, what the celebrity does, and at what times.

[0249] In such a case, it is assumed that a user would like to use the knowledge DB 116 and recommendation DB 117 learned by other users. Therefore, the following will explain a case where the knowledge DB 116 and recommendation DB 117 learned by other users can be used as the user's own knowledge DB 116 and recommendation DB 117.

[0250] In the following explanation, as explained with reference to FIG. 27, a case will be explained in which knowledge DB 116A and recommendation DB 117A are constructed as databases for user A, and knowledge DB 116B and recommendation DB 117B are constructed as databases for user B, and user A uses the database for user B (the database for user B is reused as the database for user A).

[0251] <The seventh method of using other users' databases> As a seventh method of using another user's database, a case will be described in which the user's own recommendation DB 117 is replaced with the recommendation DB 117 of another user, thereby using the other user's database as the user's own database.

[0252] 28 is a diagram illustrating the seventh reuse method. Recommendation DB 117A for user A is replaced with recommendation DB 117B for user B. This replacement allows user A to receive recommendations that reference recommendation DB 117B' for user B. Here, the replaced recommendation DB 117B is written with a dash to indicate that it has been replaced.

[0253] By performing such substitution, the information input by user A by voice input or the like is analyzed by analysis unit 131A by referring to knowledge DB 116A constructed for user A, and information to be recommended to user A is generated by generation unit 143A by referring to recommendation DB 117B' constructed for user B.

[0254] In this way, recommendations are made to user A with reference to recommendation DB 117B' constructed for user B. Therefore, recommendations are made to user A that match the preferences, lifestyle, etc. of user B.

[0255] In this way, after recommendation DB 117A constructed for user A is replaced with recommendation DB 117B' constructed for user B, if user A says, for example, "Play some music," the analysis unit 131A analyzes that an instruction to play music has been issued based on the semantic analysis parameters accumulated in knowledge DB 116A. The generation unit 143A (recommended information generation unit 145) refers to the recommendation parameters accumulated in recommendation DB 117B', selects a song to be recommended to user A based on the analysis result of the analysis unit 131A, and generates output information for playing the song.

[0256] In this case, the recommendation parameters stored in the recommendation DB 117B' are information generated based on the preferences of the user B, so that songs that match the preferences of the user B are presented to the user A.

[0257] Although no other examples are shown, as with the first reuse method, recommendation DB117A constructed for user A is referenced, semantic analysis is performed, and recommendation DB117B' constructed for user B is referenced to generate recommended information.

[0258] When user A provides feedback on the recommended song in this way, learning unit 125A learns (updates) the semantic analysis parameters and recommendation parameters based on the feedback.

[0259] As described above, the recommendation DB 117B' may become a database suited to the preferences of user A over time (as learning progresses) through the processing of the learning unit 125A. In other words, the recommendation DB 117B' may return to a state close to the recommendation DB 117A before replacement. Even though user A willingly replaced his / her own recommendation DB 117A with recommendation DB 117B' of user B, it may not be a desirable state for user A if the recommendation DB 117B' returns to the recommendation DB 117A for user A.

[0260] Therefore, when such a database replacement is performed, some kind of limitation may be placed on the learning (database update process) performed by the learning unit 125 so that the parameters stored in the database after replacement are not updated frequently. For example, limitations may be placed such that updates are performed only when instructed (permitted) by the user, or that updates are not performed for a predetermined period after replacement, for example, one week.

[0261] In this way, by replacing the recommendation DB 117 with the recommendation DB 117 of a user desired by the user, it becomes possible to receive recommendations using the recommendation DB 117 after replacement.

[0262] <The 8th method of using other users' databases> As an eighth method for reusing another user's database, a case where one's own knowledge DB 116 is replaced with another user's knowledge DB 116 will be described.

[0263] 29 is a diagram illustrating the eighth reuse method. Knowledge DB 116A of user A is replaced with knowledge DB 116B for user B. By this replacement, user A can receive recommendations that use the results of semantic analysis performed by referring to knowledge DB 116B' for user B. Here, the replaced knowledge DB 116B is written with a dash to indicate that it has been replaced.

[0264] By performing such substitution, the information input by user A by voice input or the like is analyzed by analysis unit 131A by referring to knowledge DB 116B' constructed for user B, and information to be recommended to user A is generated by generation unit 143A by referring to recommendation DB 117A constructed for user A.

[0265] In this way, semantic analysis is performed for user A with reference to knowledge DB 116B' constructed for user B. Therefore, semantic analysis is performed for user A based on semantic analysis parameters obtained from the life and friendships of user B, and recommendations are made using the results of the semantic analysis.

[0266] For example, if user A says, "Play a song about Gunma," the analysis unit 131A performs semantic analysis by referring to the knowledge DB 116B'. Because the knowledge DB 116A before the replacement recorded parameters that would result in an output indicating that the input "Gunma" is a friend of user A, the instruction would be analyzed as an instruction to play a song that "Gunma" is a friend of user A. However, because such parameters are not recorded in the knowledge DB 116B' after the replacement, a different analysis result would be output. For example, the analysis result would be a result that there is a contradiction in the instruction content, or that the instruction is to play a song about Gunma because Gunma is a place name.

[0267] The generation unit 143A refers to the recommendation DB 117A and generates information about songs to be recommended to the user A based on the analysis results and recommendation parameters from the analysis unit 131A. For example, if it is determined that there is a contradiction in the instruction content, a message such as "What is Gunma?" may be generated, a song from Gunma may be selected based on the analysis result of playing a song from Gunma, or a song that Gunma likes may be selected based on the assumption that "Gunma = Gunma."

[0268] Although no other examples are shown, as with the second reuse method, the knowledge DB 116B' constructed for user B is referenced, semantic analysis is performed, and the recommendation DB 117A constructed for user A is referenced to generate recommended information.

[0269] When user A provides feedback on the recommended song in this way, learning unit 125A learns (updates) the semantic analysis parameters and recommendation parameters based on the feedback.

[0270] As described above, the knowledge DB 116B' may become a database suited to the preferences of user A over time (as learning progresses) through the processing of the learning unit 125A. In other words, there is a possibility that the knowledge DB 116B' may return to a state close to the knowledge DB 116A before replacement. Even though user A willingly replaced his / her own knowledge DB 116A with user B's knowledge DB 116B', it may not be a desirable state for user A if the knowledge DB 116B' returns to the knowledge DB 116A for user A.

[0271] Therefore, when such a database replacement is performed, some kind of limitation may be placed on the learning (database update process) performed by the learning unit 125 so that the parameters stored in the database after replacement are not updated frequently. For example, limitations may be placed such that updates are performed only when instructed (permitted) by the user, or that updates are not performed for a predetermined period after replacement, for example, one week.

[0272] In this way, by replacing the knowledge DB 116 with the knowledge DB 116 of the user desired by the user, it is possible to provide recommendations using the replaced knowledge DB 116.

[0273] <The ninth method of misusing other users' databases> As a ninth method of using other users' databases, a case where one's own recommendation DB 117 and other users' recommendation DB 117 are used in combination will be described.

[0274] 30 is a diagram illustrating the ninth diversion method. A recommendation DB 117B' for user B is added to the storage unit 118 (FIG. 22) of the information processing device 11 of user A. As a result, the storage unit 118 of user A stores a recommendation DB 117A constructed for user A and a recommendation DB 117B' constructed for user B. A DB switching unit 303 is added so that either one of the two recommendation DBs 117 can be referenced.

[0275] The DB switching unit 303 may be provided as a part of the function of the generating unit 143, or may be provided between the storage unit 118 and the generating unit 143 in the configuration of the information processing device 11 shown in FIG.

[0276] User A can receive recommendations that refer to recommendation DB 117B' for user B. When DB switching unit 303 switches the database to be referred to to recommendation DB 117B', the processing is performed as described in the seventh diversion method. Therefore, information input by user A by voice input or the like is analyzed by analysis unit 131A by referring to knowledge DB 116A constructed for user A, and information to be recommended to user A can be generated by generation unit 143A by referring to recommendation DB 117B' constructed for user B.

[0277] Furthermore, when DB switching unit 303 switches the database to be referred to to recommendation DB 117A, the process described with reference to the flowchart in Fig. 26 is performed. Therefore, information input by user A by voice input or the like is analyzed by analysis unit 131A by referring to knowledge DB 116A constructed for user A, and information to be recommended to user A can be generated by generation unit 143A by referring to recommendation DB 117A constructed for user A.

[0278] In this way, recommendations are made to user A with reference to recommendation DB 117A constructed for user A or recommendation DB 117B' constructed for user B. Therefore, recommendations are made to user A that match the preferences and lifestyles of users A and B, respectively.

[0279] The DB switching unit 303 can be configured to switch when the location of user A changes significantly, for example, when the user A goes on a trip or moves. For example, when user A is traveling in area B, the DB is switched to recommendation DB 117B' constructed for user B who lives in area B. Then, recommended information is generated by referring to recommendation DB 117B'. In this case, it becomes possible to recommend to user A information closely related to area B, such as restaurants that user B regularly uses or places to play that user B frequently visits.

[0280] Although no other examples are shown, the DB switching unit 303 may switch databases at the timing explained in the third diversion method. Of course, the switching timing of the DB switching unit 303 may be at a timing other than that described above, and the above example is merely an example and is not intended to be limiting.

[0281] <10th method of using other users' databases> As a tenth method of using other users' databases, a case where one's own knowledge DB 116 and other users' knowledge DB 116 are used in combination will be described.

[0282] 31 is a diagram for explaining the fourth reuse method. A knowledge DB 116B' for user B is added to the storage unit 118 (FIG. 22) of the information processing device 11 of user A. As a result, the storage unit 118 of user A stores a knowledge DB 116A constructed for user A and a knowledge DB 116B' constructed for user B. A DB switching unit 304 is added so that either one of the two knowledge DBs 116 can be switched and referenced.

[0283] The DB switching unit 304 may be provided as a part of the function of the generating unit 143, or may be provided between the storage unit 118 and the generating unit 143 in the configuration of the information processing device 11 shown in FIG.

[0284] User A can receive recommendations based on the analysis results obtained by referring to the knowledge DB 116B' for user B. When the DB switching unit 304 switches the database to be referred to to the knowledge DB 116B', the processing is performed as described in the above-mentioned eighth diversion method. Therefore, information input by user A by voice input or the like is analyzed by the analysis unit 131A by referring to the knowledge DB 116B' constructed for user B, and information to be recommended to user A can be generated by the generation unit 143A by referring to the recommendation DB 117A constructed for user A.

[0285] Furthermore, when the DB switching unit 304 switches the database to be referred to to the knowledge DB 116A, the processing is as described with reference to the flowchart in Fig. 26. Therefore, information input by the user A by voice input or the like is analyzed by the analysis unit 131A by referring to the knowledge DB 116A constructed for the user A, and the generation unit 143A refers to the recommendation DB 117A constructed for the user A, thereby generating information to be recommended to the user A.

[0286] In this way, semantic analysis is performed with reference to the knowledge DB 116A constructed for user A or the knowledge DB 116B' constructed for recommended user B, and the result of the semantic analysis is used to make a recommendation to user A. Therefore, recommendations are made to user A that match the preferences and lifestyles of users A and B, respectively.

[0287] The DB switching unit 304 can be configured to switch when the location of user A changes significantly, for example, when the user A goes on a trip or moves. For example, when user A is traveling in area B, the DB switching unit 304 switches to knowledge DB 116B' constructed for user B who lives in area B. Then, semantic analysis may be performed with reference to knowledge DB 116B', and recommended information may be generated based on the analysis results.

[0288] In this case, information such as restaurants that user B, who lives in area B, regularly uses and places to play that he or she frequently visits is recorded in knowledge DB 116B' (for example, parameters are recorded so that when "meal" is input, "Store A" is output), so it becomes possible to perform semantic analysis that is appropriate for information that is specific to area B, and make recommendations to user A based on that semantic analysis.

[0289] Although no other examples are shown, the DB switching unit 304 may switch databases at the timing explained in the fourth repurposing method. Of course, the switching timing of the DB switching unit 304 may be other than the above, and the above example is merely an example and is not intended to be limiting.

[0290] <11th method of using other users' databases> As an eleventh method for utilizing other users' databases, a case where one's own recommendation DB 117 is integrated with another user's recommendation DB 117 will be described.

[0291] Consider a case where recommendation DB 117AB is constructed by integrating recommendation DB 117A constructed for user A and recommendation DB 117B constructed for user B. As described above, recommendation DB 117 does not record the recommendation scores recorded in recommendation DB 112 according to the first embodiment, but does record recommendation parameters.

[0292] For example, in the case of recommendation scores, the integrated recommendation score can be calculated by calculating the average value of the recommendation scores as described with reference to Fig. 18, but recommendation parameters are recorded in each of recommendation DB 117A and recommendation DB 117B, and it is difficult to use the recorded recommendation parameters as they are to calculate, for example, an average value and use it as the integrated parameter. The learning unit 125 learns (updates) the parameters based on the interaction information recorded in interaction DB 113, for example, using a technique such as reinforcement learning.

[0293] For this reason, as shown in Fig. 32, when recommendation DB 117A constructed for user A and recommendation DB 117B constructed for user B are integrated to construct recommendation DB 117AB, first, interaction DB 113A constructed for user A and interaction DB 113B' constructed for user B are integrated. Then, learning unit 125A performs re-learning based on the interaction information recorded in the integrated interaction DB 113AB, thereby constructing recommendation DB 117AB. The integration of interaction DB 113 will be described later.

[0294] As described above, the interaction DB 113A stores data on input information from the user A and output information based on an algorithm for the input information. The learning unit 124A learns based on the data stored in the interaction DB 113A, and the recommendation DB 117A is constructed. This algorithm is trained to be suitable for the user A.

[0295] Furthermore, the interaction DB 113B stores data relating to input information from the user B and output information based on an algorithm for the input information. The learning unit 124B learns based on the data stored in the interaction DB 113B, thereby constructing the recommendation DB 117B. This algorithm is learned to be an algorithm suitable for the user B (different from the algorithm suitable for the user A).

[0296] Such interaction DB113A and interaction DB113B are integrated, and re-learning is performed using the integrated interaction DB113AB. In the example shown in Fig. 32, the learning unit 125A performs re-learning based on the integrated interaction DB113AB, and the recommendation DB117AB is constructed. As a result, the recommendation DB117AB can be a database in which the recommendation DB117A constructed for user A and the recommendation DB117B constructed for user B are integrated.

[0297] In the above and following explanations, the explanation will be continued assuming that the re-learning is performed by the learning unit 125, in other words, that the re-learning is performed in the device on the user A side. However, the re-learning may be performed in a device other than the device on the user A side (information processing device 11). In other words, the re-learning may be performed in a device other than the information processing device 11, and the learning model after the re-learning may be supplied to the information processing device 11.

[0298] Furthermore, when relearning is performed in another information processing device, the exchange DB 113A and the exchange DB 113B are supplied (sent) to the other information processing device that performs relearning.

[0299] By integrating and relearning the interaction DB113, information input by user A by voice input or the like is analyzed by analysis unit 131A by referring to knowledge DB116A constructed for user A, and information to be recommended to user A is generated by generation unit 143A by referring to recommendation DB117AB, which is an integration of recommendation DB117A constructed for user A and recommendation DB117B constructed for user B.

[0300] In this way, recommendations are made to user A by referencing recommendation DB 117AB, which is an integration of different databases. Therefore, recommendations are made to user A that match the preferences and lifestyles of user A and user B.

[0301] This technology can be applied when integrating two recommendation DBs 117, but it can also be applied when integrating two or more recommendation DBs 117. When integrating multiple recommendation DBs 117, multiple interaction DBs 113 are integrated using an integration method described below, and then re-learning is performed by the learning unit 125, thereby integrating the multiple recommendation DBs 117.

[0302] <12th method of using other users' databases> As a twelfth method of using other users' databases, a case where one's own knowledge DB 116 is integrated with another user's knowledge DB 116 will be described.

[0303] Fig. 33 is a diagram for explaining the twelfth reuse method. A knowledge DB 116AB for user A is constructed by integrating a knowledge DB 116A for user A and a knowledge DB 116B for user B. As with the eleventh reuse method described with reference to Fig. 32, first, an interaction DB 113A constructed for user A and an interaction DB 113B' constructed for user B are integrated. Then, the learning unit 125A re-learns based on the interaction information recorded in the integrated interaction DB 113AB, thereby constructing the knowledge DB 116AB. The integration of the interaction DB 113 will be described later.

[0304] By performing such integration, information input by user A by voice input or the like is analyzed by analysis unit 131A by referring to knowledge DB 116AB, which is an integration of knowledge DB 116A constructed for user A and knowledge DB 116B constructed for user B, and information to be recommended to user A is generated by generation unit 143A by referring to knowledge DB 116A constructed for user A.

[0305] In this way, semantic analysis is performed with reference to knowledge DB 116AB, which is an integration of different databases, and recommendations based on the analysis results are made to user A. Therefore, recommendations are made to user A that match the preferences and lifestyles of user A and user B.

[0306] This technology can be applied when integrating two knowledge DBs 116, but it can also be applied when integrating two or more knowledge DBs 116. When integrating multiple knowledge DBs 116, multiple interaction DBs 113 are integrated using an integration method described below, and then re-learning is performed by the learning unit 125, thereby integrating the multiple knowledge DBs 116.

[0307] <13th method of misappropriating other users' databases> As a thirteenth method of reusing other users' databases, a case will be described in which one's own knowledge DB 116 is integrated with other users' knowledge DB 116 and one's own recommendation DB 117 is integrated with other users' recommendation DB 117.

[0308] 34 is a diagram for explaining the thirteenth diversion method. As with the eleventh and twelfth diversion methods, first, the exchange DB 113A constructed for user A and the exchange DB 113B' constructed for user B are integrated. The integration of the exchange DB 113 will be described later.

[0309] The learning unit 125A re-learns based on the interaction information recorded in the integrated interaction DB 113AB, thereby constructing a knowledge DB 116AB. Also, the learning unit 125A re-learns based on the interaction information recorded in the integrated interaction DB 113AB, thereby constructing a recommendation DB 117AB.

[0310] By performing such integration, information input by user A by voice input or the like is analyzed by analysis unit 131A by referring to knowledge DB 116AB, which is an integration of knowledge DB 116A constructed for user A and knowledge DB 116B constructed for user B, and information to be recommended to user A is generated by generation unit 143A by referring to recommendation DB 117AB, which is an integration of recommendation DB 117A constructed for user A and recommendation DB 117B constructed for user B.

[0311] In this way, semantic analysis is performed with reference to knowledge DB 116AB, which is an integration of different databases, and recommendations based on the analysis results are made to user A by referring to recommendation DB 117AB, which is an integration of different databases. Therefore, recommendations are made to user A that match the preferences and lifestyles of user A and user B.

[0312] <How to integrate the transaction database> In the eleventh to thirteenth appropriation methods, when the knowledge DB 116 and the recommendation DB 117 are integrated, the exchange DB 113 is also integrated. The integration of the exchange DB 113 will now be described.

[0313] As described above, when integrating interaction DB113A for user A and interaction DB113B for user B to construct interaction DB113AB, all interaction information recorded in interaction DB113B for user B may be used, or interaction information that meets specified conditions may be extracted and the extracted interaction information may be used.

[0314] The predetermined condition for extracting the interaction information may be, for example, a predetermined period of time, a predetermined word, etc. For example, if user B has passed the entrance exam for the school that user A is taking, and information for passing the exam is to be extracted from user B's interaction DB 113B, the period during which user B studied for the exam can be set as the predetermined period for extracting the information.

[0315] A case where information that meets predetermined conditions is extracted from exchange DB 113 will be described with reference to Fig. 35 and Fig. 36. Here, an example will be described where information that meets predetermined conditions is extracted from exchange DB 113B for user B and exchange DB 113B' is created.

[0316] The upper part of Fig. 35 shows some of the interactions recorded in interaction DB 113B, with the horizontal axis representing the time axis and showing the history of interactions between user B and information processing device 11. The lower part of Fig. 35 shows some of the interactions extracted from interaction DB 113B.

[0317] In the interaction DB 113B shown in the upper part of Fig. 35, input information from user B is input to the information processing device 11, and the times (t1 to t15) at which output information corresponding to the input information is output are arranged in chronological order on the time axis. A diagonally shaded triangular marker is shown at the time when input and output of input information and output information occurred. The information processing device 11 trains an algorithm based on these interactions.

[0318] The records of interactions recorded in the interaction DB 113B are extracted. For example, information about interactions at times t3, t4, t5, t9, t11, and t12, which correspond to the triangular marks with grids, is extracted from the interaction DB 113B and recorded in the interaction DB 113B′.

[0319] The processing unit 120 searches for interactions to be extracted based on the interaction DB 113B and extracts the interactions. More specifically, the extraction unit 122 searches for interactions (label information) recorded in the interaction DB 113B, for example, based on an input from a user. For example, the extraction unit 122 may search for label information from the interaction DB 113B based on a keyword such as the above-mentioned "Gunma," and extract training data based on the label information. The correction unit 124 extracts the searched training data from the training DB 114. The correction unit 124 may also extract the searched label information, etc. from the interaction DB 113B.

[0320] The extraction unit 122 may also extract learning data that is derivatively influenced by information about the exchange to be extracted from the learning DB 114. For example, when learning data is extracted based on the keyword "Gunma," the extraction unit 122 may extract learning data corresponding to an event whose output information changes before and after the extraction. The correction unit 124 may extract learning data related to the event.

[0321] The events extracted in this way correspond to the exchanges at times t3, t4, t5, t9, t11, and t12 shown in FIG.

[0322] A case where learning data that is derivatively influenced by information about the exchange to be extracted is extracted from the learning DB 114 will now be described.

[0323] The information processing device 11 extracts keywords from the exchange DB 113, and extracts information contained in the exchange DB 113 based on the extraction results.

[0324] More specifically, the extraction unit 122 acquires keywords to be extracted and extracts the keywords from the exchange DB 113. Note that the keywords may be, for example, input to the user terminal 20 by a user operation, transmitted to the information processing device 11, and transmitted to the extraction unit 122. Here, it is assumed that the word "Gunma" is transmitted to the extraction unit 122 as a keyword. The extraction unit 122 searches the exchange DB 113 for information containing "Gunma" and extracts learning data from the learning DB 114.

[0325] Next, the correction unit 124 extracts learning data containing "Gunma" recorded in the learning DB 114. For example, the correction unit 124 extracts learning data containing "Gunma" recorded in the learning DB 114, such as "Gunma's favorite song is XX," "Gunma is my friend Iwasa's nickname," and "Play the song Gunma."

[0326] In this way, information about "Gunma" is extracted from the exchange DB 113. However, there is a possibility that not all of the information about "Gunma" is extracted using this alone.

[0327] For example, information that does not include the word "Gunma" cannot be found by extraction using the keyword "Gunma" as described above. For example, assume that the input information "Play a song with a similar taste to the song I listened to yesterday" is input to the information processing device 11. If the song listened to yesterday was a song about Gunma, the recommendation parameters for a taste similar to the song about Gunma are updated. In this case, the input information that does not include the word "Gunma" directly affects the information in the recommendation DB 112.

[0328] Furthermore, suppose that input information such as "Play a song completely different from the song I listened to yesterday" is input to the information processing device 11. If the song listened to yesterday was a song about Gunma, the recommendation parameters for songs with a completely different taste from the song about Gunma are updated in the recommendation DB 117. In other words, the input information that does not include the word "Gunma" indirectly influences the information in the recommendation DB 117. In this way, interactions that influence the information in the recommendation DB 117 are also extracted from the interaction DB 113.

[0329] In this way, it is considered that the user may want to extract data related to input information that does not include the information "Gunma" from the storage unit 118. Therefore, one possible method for extracting this related data is to store information related to various data in advance, for example, in the storage unit 118. For example, one possible method is to store input information such as "Play a song with a similar style to the song I listened to yesterday" and "Play a song completely different from the song I listened to yesterday" in the interaction DB 113 as information related to the information "Play a song about Gunma." In this way, when the interaction DB 113 is searched using the keyword "Gunma," the above information "Play a song with a similar style to the song I listened to yesterday" and "Play a song completely different from the song I listened to yesterday" is searched from the interaction DB 113. Based on the searched information, learning data can be extracted from the learning DB 114.

[0330] Although this method is acceptable, it requires that information representing the relationships between the various pieces of information stored in the interaction DB 113 be stored in the interaction DB 113, for example, which could result in the amount of information recorded in the interaction DB 113 becoming enormous.

[0331] <Methods for extracting other related information> Therefore, a method for determining whether or not predetermined information is related information when an inconsistency occurs in the output information after extraction of the predetermined information will be described. More specifically, when predetermined information is extracted, it is assumed that the predetermined information is deleted. If the output information before the deletion differs from the output information after the deletion, the extracted information is determined to be related to the keyword. According to this method, it is not necessary to record information indicating the relationship between various pieces of information stored in the exchange DB 113.

[0332] Referring to Figure 36, we will explain a method for determining whether information about an interaction containing a keyword is related to a keyword based on changes in the output information output based on an algorithm when the information about the interaction is extracted.

[0333] Fig. 36 is a diagram showing output information generated before and after deletion of information related to interactions, and processing details based on changes in the output information before and after that. Three examples are shown in Fig. 36. The three examples shown in Fig. 36 will be explained below. From the left, the song the user listened to yesterday, the output (before and after deletion), the change in output, and the processing details are shown.

[0334] The output (before deletion) is the content of the output information in response to the input information from the user, "Play a song with a similar taste to the song I listened to yesterday." In the following three examples, the extraction unit 122 extracts information containing the keyword "Gunma" from the storage unit 118. The content of the output information in response to the above input information after the extraction is performed is shown as the output (after deletion).

[0335] In the first example, the interaction information records that the user listened only to "Gunma songs" yesterday. Therefore, the output information for the input information is information for playing only "Gunma songs." On the other hand, if the information containing "Gunma" is deleted, the history of playing "Gunma songs" will disappear, and the algorithm will no longer be able to understand the user's intentions. Therefore, the output information will change to, for example, "Did I play that song yesterday?"

[0336] In this way, if it is determined that the output information will change due to the hypothetical deletion of information containing the keyword "Gunma," the determination unit 123 determines that the input information is data related to the keyword "Gunma" (related data). In this way, the information processing device 11 according to this embodiment extracts information containing a keyword, and if the extracted keyword is deleted, it can determine the relevance between the input information and the keyword based on the change in the output information before and after the deletion. This eliminates the need for the information processing device 11 to store what keywords each piece of input information is related to.

[0337] Next, in the second example, yesterday the user listened to a song about Gunma and a song other than Gunma. That is, the output (before deletion) is an output that plays "a song about Gunma" and "a song other than Gunma." On the other hand, the output (after deletion) after various information including "Gunma" is temporarily deleted is an output that plays only "a song other than Gunma." In this case as well, a change occurs between the output (before deletion) and the output (after deletion). For this reason, the determination unit 123 determines that the input information is data related to the keyword "Gunma."

[0338] Next, in the third example, the user listened only to "songs other than Gunma" yesterday. Therefore, the output (before deletion) is an output that plays only "songs other than Gunma." On the other hand, the output (before deletion) is an output that is not related to the keyword. Therefore, the output (after deletion) does not change from the output (before deletion). In this case, the determination unit 123 determines that the input information is data that is not related to the keyword.

[0339] In this way, by determining the relevance between a keyword and input information, it becomes possible to extract input information related to the keyword based on the keyword. That is, not only the input information "Play a Gunma song," but also input information (communication) related to the keyword "Gunma," such as "Play a song with a similar taste to the (Gunma) song I listened to yesterday," or "Play a song that is completely different from the (Gunma) song I listened to yesterday," can be extracted.

[0340] Furthermore, according to this method, the relevance between keywords and input information is determined based on changes in output information, which eliminates the need for the information processing device 11 to store the relevance between various keywords and input information, making it possible to extract information related to interactions based on less information.

[0341] <Additional methods for extracting related information> Another related data extraction method will be described with reference to Fig. 37. Four examples are shown in Fig. 37. Four examples in which the output is changed will be described below. In each example, the extraction unit 122 searches the exchange DB 113 for the keyword "Gunma" and extracts learning data from the learning DB 114 based on the search results.

[0342] In the first example, the user listened only to "Gunma's Song" yesterday. In this case, as in the case described with reference to Fig. 36, the change in output before and after deletion is large. Therefore, in this case, the determination unit 123 determines that the change in output is large and determines that the data is related data.

[0343] Next, in the second and third examples, the user listened to "Gunma songs" and "non-Gunma songs" yesterday. However, in the second and third examples, the number of "Gunma songs" and the number of "non-Gunma songs" that the user listened to are different. Specifically, in the second example, the user terminal 20 plays nine "Gunma songs" and one "non-Gunma song." On the other hand, in the third example, the user terminal 20 plays one "Gunma song" and nine "non-Gunma songs."

[0344] In the second example, before and after the extraction of learning data related to the keyword "Gunma," the output information changes from information to play nine "Gunma songs" and one "non-Gunma song" to information to play one "Gunma song." In this case, the determination unit 123 determines that the change in output before and after deletion is large, and determines that the data is related data.

[0345] On the other hand, in the third example, before and after the extraction of learning data related to the keyword "Gunma," the output information changes from information to play one "Gunma song" and nine "non-Gunma songs" to nine "non-Gunma songs." In this case, the determination unit 123 determines that the change in output before and after deletion is small, and determines that the data is not related data.

[0346] In the fourth example, the user listened only to "Gunma no uta" yesterday. In this case, as in the case described with reference to Fig. 36, the determination unit 123 determines that the amount of change in output before and after deletion is small, and determines that the data is not related data.

[0347] An example of the information extraction process by the information processing device 11 according to the present embodiment has been outlined above with reference to Fig. 37. Next, the extraction process of the exchange DB 113 by the information processing device 11 according to the present embodiment will be described with reference to Fig. 38.

[0348] First, the information processing device 11 acquires input information (step S402). More specifically, the information processing device 11 receives, as input information, a keyword input by a user to the user terminal 20 and information requesting extraction of learning data related to the keyword (hereinafter also simply referred to as "request information") via the network 30. Here, the keyword is, for example, the word "Gunma." The information processing device 11 receives the keyword and request information as input information and transmits the received input information to the extraction unit 122 included in the processing unit 120 via the communication control unit 160.

[0349] Next, the extraction unit 122 extracts related information based on the transmitted input information (step S404). Specifically, the extraction unit 122 extracts information related to the keyword "Gunma." More specifically, the extraction unit 122 extracts information related to the keyword "Gunma" that is stored in the knowledge DB 111, the recommendation DB 112, or the interaction DB 113. Note that the extraction unit 122 does not extract output information recorded in the interaction DB 113.

[0350] Next, the generation unit 140 generates output information (step S406). At this time, the generation unit 140 generates the output information assuming that the related information extracted in step S404 (excluding the input information used in step S406) has been extracted. In other words, the generation unit 140 generates the output information assuming that there has been no interaction related to the keyword "Gunma." At this time, the output information generated by the generation unit 140 may differ from the output information recorded in the interaction DB 113.

[0351] Next, the determination unit 123 determines the magnitude of the change in the output information (step S408). More specifically, the determination unit 123 determines the magnitude of the difference between the output information generated in step S406 and the output information recorded in the learning DB 114 that corresponds to the input information used to generate the output information, as the magnitude of the change in the output information.

[0352] Next, the correction unit 124 extracts the output information recorded in the learning DB 114 according to the determination result by the determination unit 123 in step S408 (step S410). For example, if it is determined in step S408 that the change in the output information is large, the correction unit 124 extracts the output information that was the subject of the determination in step S408 and the input information corresponding to it, which are recorded in the learning DB 114.

[0353] Furthermore, if it is determined in step S408 that there is no change in the output information, the correction unit 124 maintains the output information that was recorded in the exchange DB 113 and that was the subject of the determination in step S408.

[0354] Next, if the extraction unit 122 determines based on the exchange DB 113 that there is undetermined output information (step S412: Yes), the process returns to step S402. On the other hand, if the extraction unit 122 determines based on the exchange DB 113 that there is no undetermined output information (step S412: No), the extraction process shown in FIG.

[0355] In this manner, the information processing device 11 according to the present embodiment extracts the exchange information recorded in the exchange DB 113.

[0356] Referring again to the extraction example shown in FIG. 35, for example, interactions at times t3, t4, and t5 are extracted as interactions within a predetermined period. As related data related to these interactions at times t3, t4, and t5, interactions at times t9, t11, and t12 are extracted. In this case, if only interactions within a predetermined period are extracted, only the interactions at times t3, t4, and t5 will be extracted. However, by also extracting related data, interactions at times t9, t11, and t12 will also be extracted.

[0357] For example, if user B is a celebrity and interactions during the recording period of a specific program are extracted from interaction DB 113B, interactions during the recording period are extracted first. For example, interactions at times t3, t4, and t5 are extracted as interactions during the recording period. User B may also be interacting with information processing system 1 about topics related to the content of the recording even when not recording.

[0358] Such exchanges such as conversations outside of recording are also extracted as exchanges related to exchanges during the recording period. For example, exchanges at times t9, t11, and t12 are extracted as related data.

[0359] In this way, predetermined information is extracted from the interaction DB 113B constructed for user B, and an interaction DB 113B' is constructed. This interaction DB 113B' is provided to user A and integrated with the interaction DB 113A constructed for user A, thereby constructing an interaction DB 113AB. This integration will be described with reference to FIG. 39.

[0360] The upper part of FIG. 39 shows the exchange history of the exchange DB 113B', the middle part of FIG. 39 shows the exchange history of the exchange DB 113A, and the lower part of FIG. 39 shows the exchange history of the exchange DB 113AB.

[0361] Referring to the interaction history of interaction DB113B' shown in the upper part of Figure 39, by the above-mentioned process, interactions at times t3, t4, t5, t9, t11, and t12 are extracted from interaction DB113B for user B and recorded in interaction DB113B'. Referring to the interaction history of interaction DB113A shown in the middle part of Figure 39, interactions at times t1' to t10' are recorded in interaction DB113A. Interactions marked with a dash indicate interactions in interaction DB113A.

[0362] When integrating the interaction DB113B' and the interaction DB113A, the interactions recorded in each database are integrated by arranging them in chronological order. In the example shown in Fig. 39, the interactions are rearranged in the order of time t1', time t2', time t3', time t3, time t4', time t4, time t5', time t5, time t6', time t7', time t9, time t8', time t11, time t12, time t9', and time t10'.

[0363] In this way, re-learning is performed using the interaction DB 113AB in which the interaction information rearranged in chronological order is recorded, whereby an integrated knowledge DB 116AB and / or recommendation DB 117AB is constructed.

[0364] Incidentally, as described with reference to FIG. 39, if the exchange information is arranged in chronological order and the exchange DB 113B' and the exchange DB 113A are integrated, the contents of the exchange may change suddenly.

[0365] For example, the interactions at time t3, time t4, and time t5 are interactions of user B, and the interactions at time t3', time t4', and time t5' are interactions of user A. After integration, the interactions are rearranged in the order of time t3', time t3, time t4', time t4, time t5', and time t5.

[0366] Assume that user B interacts at time t3, time t4, and time t5 in location B, and user A interacts at time t3', time t4', and time t5' in location A. After integration, the data is rearranged in the order of time t3', time t3, time t4', time t4, time t5', and time t5, resulting in a mixture of interactions at location A and interactions at location B. In this way, if the location where the interactions took place suddenly changes, inconsistencies will occur in the interactions, and if relearning is performed using data with such inconsistencies, the relearning may not be performed correctly.

[0367] Therefore, integration may be performed as shown in Fig. 40. The diagrams shown in the upper and middle sections of Fig. 40 are the same as the diagrams shown in the upper and middle sections of Fig. 39, and respectively show the exchange history of exchange DB113B' and the exchange history of exchange DB113A.

[0368] The lower part of Fig. 40 shows the transaction history of the transaction DB113AB. In the transaction DB113AB shown in the lower part of Fig. 40, transactions at times t1' to t10' are listed, followed by transactions at times t3, t4, t5, t9, t11, and t12. The database to be integrated, in this case, the data of the transaction DB113B', is shifted and arranged to a time when no inconsistency occurs.

[0369] As in the above example, assume that user A interacts at time t1' to t10' in location A, and user B interacts at time t3, t4, t5, t9, t11, and t12 in location B. In this case, interaction DB 113AB shown in the lower part of Fig. 40 has the interactions that user A interacts at time t1' to t10' in location A, followed by the interactions that user B interacts at time t3, t4, t5, t9, t11, and t12 in location B, so it is considered that the location where the interactions are taking place will not suddenly change, and the content of the interactions will not suddenly change either.

[0370] This makes it possible to prevent integration that would cause inconsistencies due to sudden changes in the content of interactions. Also, it is possible to perform correct relearning by referring to the interaction DB 113AB where integration that prevents inconsistencies from occurring has been performed.

[0371] 41, the strength of the interaction information recorded in interaction DB 113AB that has been recorded in interaction DB 113B' may be set to be weaker. Interaction DB 113B' records interaction information extracted from interaction DB 113B for user B, and by weakening the strength of this interaction information, it is possible to re-learn user B's personality with a weakened character.

[0372] For example, weakening the FB in the interaction DB to be integrated can act to maintain the personality before integration. Here, an example has been described in which weakening the FB information contained in User B's interaction DB acts to maintain User A's personality, but it is also possible to weaken the FB in User A's interaction DB to maintain User B's personality (to strongly reflect User B's personality).

[0373] Here, an example has been described in which two exchange DBs 113 are integrated, but any number of databases may be integrated.

[0374] Incidentally, there is a possibility that highly confidential information such as personal information is recorded in the exchange DB 113. When the exchange information is extracted from the exchange DB 113 described with reference to Figs. 35 to 38, information that is determined to be highly confidential, such as personal information, may not be extracted.

[0375] Furthermore, as shown in A of Fig. 42, the extracted exchange information may be encrypted and provided to other users. Exchange information that meets predetermined conditions is extracted from exchange DB 113B using the method described above. This extracted exchange information is encrypted using a predetermined encryption method. Then, a database in which the encrypted exchange information is recorded is provided to user A as exchange DB 113B'.

[0376] The learning unit 125A included in the information processing device 11 on the user A side holds a decryption key 331. When using the exchange information recorded in the exchange DB 113B′, the learning unit 125A decrypts the exchange information using the held key 331, and performs re-learning using the decrypted exchange information.

[0377] In this way, by providing encrypted exchange information to other users (in this case, user A), it is possible to prevent the exchange information itself from being leaked to other users, and it is also possible to prevent highly confidential information such as personal information from being leaked.

[0378] Alternatively, as shown in B of FIG. 42, the extracted interaction information may be converted into features, and the features may be provided to other users. From the interaction DB 113B, interaction information that meets predetermined conditions is extracted using the method described above. Features are extracted from this extracted interaction information. Then, an interaction feature DB 113B" is generated in which the extracted features are recorded as interaction information, and the DB is provided to user A.

[0379] The learning unit 125A included in the information processing device 11 on the user A side performs re-learning using the feature amount when using the interaction information recorded in the interaction DB 113B'.

[0380] As described above, the learning unit 125 performs parameter learning based on a technique such as reinforcement learning, for example, based on the interaction information recorded in the interaction DB 113. As described above, parameter learning means optimizing black-box parameters in accordance with the accumulation of learning data.

[0381] The analysis unit 131 or the recommendation information generation unit 145 inputs input values ​​to a network in which an input layer consisting of multiple inputs and an output layer consisting of multiple outputs are connected by intermediate layers, and outputs output values ​​related to the analysis results or recommendation information, etc. The parameters that define the weights of the nodes in this network are called black-box parameters.

[0382] For this reason, the feature can be, for example, the output of an intermediate layer. Even if the feature were to be leaked, it would be difficult for a user to understand its meaning. By providing the interaction information that has been made into the feature to another user (in this case, user A), it is possible to prevent the interaction information itself from being leaked to other users, and to prevent highly confidential information such as personal information from being leaked.

[0383] In the above-described embodiment, for example, referring again to Figure 34, predetermined interaction information is extracted from interaction DB113B on the user B side, and interaction DB113B' in which the extracted interaction information is recorded is provided to user A side, and interaction DB113AB is constructed.

[0384] In this way, instead of providing the interaction DB 113B' to the user A side, the interaction DB 113A of the user A may be provided to the user B side, and the interaction DB 113AB may be constructed on the user B side. Then, the interaction DB 113AB constructed on the user B side may be provided (returned) to the user A side.

[0385] In this way, by having the integration process performed on the user B side, the interaction information recorded in user B's interaction DB 113B is not provided to user A as is, thereby preventing the leakage of highly confidential information such as user B's personal information.

[0386] Incidentally, the interaction DB 113 records a history of interactions with users. Furthermore, interactions with users are performed via user terminals 20 (FIG. 1). The user terminals 20 may be terminals that perform interactions using only voice, or may be terminals that perform interactions using voice and video.

[0387] For example, if the user terminal 20 is a device that interacts with the user mainly through voice and determines the user's reaction by filming the user and analyzing the video, the audio and video are input to the user terminal 20, and the audio and video are recorded in the interaction DB 113.

[0388] If the user terminal 20 does not have a function for capturing an image of the user, voice is input to the user terminal 20, and the voice is recorded in the exchange DB 113.

[0389] In this way, the types (formats) of data recorded in interaction DB 113 may differ. For example, assume that audio data is recorded in interaction DB 113A of user A, and audio data and video data are recorded in interaction DB 113B of user B. In such a case, the interaction information extracted from interaction DB 113B of user B will be audio data and video data, and this audio data and video data will be integrated with the audio data in interaction DB 113A of user A.

[0390] In this way, when the data being handled is different, the difference can be ignored and re-learning can be performed. Video data is used, for example, to determine a user's reaction, but re-learning can be performed assuming that such data serving as a basis for determination, in other words, the user's biometric information, is not present. Re-learning assuming that the biometric information is not present may result in an erroneous learning result. Therefore, feature quantities may be used.

[0391] 43 is a diagram illustrating a case where integration is performed using features. The learning unit 125A on the user A side includes a feature conversion unit 351A and an update parameter calculation unit 352A. The feature conversion unit 351A converts input voice data into features and outputs them to the update parameter calculation unit 352A. The update parameter calculation unit 352A calculates knowledge parameters to be recorded in the knowledge DB 116A and / or recommendation parameters to be recorded in the recommendation database 117A, and records them in the corresponding databases.

[0392] Similarly, learning unit 125B on the user B side includes feature conversion unit 351B and update parameter calculation unit 352B. Feature conversion unit 351B converts input audio data and video data into features and outputs them to update parameter calculation unit 352B. Update parameter calculation unit 352B calculates knowledge parameters to be recorded in knowledge DB 116B and / or recommendation parameters to be recorded in recommendation database 117B, and records them in the corresponding databases.

[0393] In Figure 43, audio data and video data are used as examples of input data, but this technology can also be applied to cases where various other data, such as position data and temperature data, are input.

[0394] As the feature, the feature described with reference to Fig. 42, i.e., the black box parameter, can be used. As in the case described with reference to Fig. 42, interaction information is extracted from interaction DB 113B. When the technology described with reference to Fig. 43 is used, this extracted interaction information becomes a feature. Interaction DB 113B' in which this feature is recorded is provided to user A and integrated with user A's interaction DB 113A.

[0395] During re-learning, as shown in the lower left of Fig. 43, the feature values ​​are supplied to the update parameter calculation unit 352A, and re-learning is performed. By performing learning using such feature values, it is possible to absorb differences in database formats and integrate the databases. Furthermore, by using feature values, it is possible to prevent information leakage during integration.

[0396] <Application example> An application example of the information processing system 1 according to this embodiment will be described below.

[0397] <First application example> The information processing system 1 of the present disclosure can also be applied to technologies such as autonomous driving or driving navigation. For example, a case where the information processing system 1 is applied to autonomous driving that assists a user's driving will be described. Note that autonomous driving means that the vehicle uses information from various sensors installed in the vehicle to travel to a destination set autonomously by the vehicle itself, even without the user performing any driving operation. However, this also includes a case where the system assists a part of the driving operation when the user is driving.

[0398] When the information processing system 1 is applied to autonomous driving, data related to past driving is recorded in the interaction DB 113. From this past driving history, driving data preferred by the user (assumed to be Driver A) is extracted. For example, driving data that has generated positive feedback is extracted. The database in which the extracted driving data is recorded is referred to as an extracted driving DB.

[0399] Consider a database (referred to as a local DB) included in an information processing system 1 installed in a vehicle known as a shared car, where multiple users share one vehicle, or in a taxi. The local DB is optimized for the area where the vehicle is used. By integrating such a local DB with the extracted driving DB described above, a database that reflects driver A's preferred driving habits can be constructed.

[0400] By using this database that reflects Driver A's preferred driving preferences to perform relearning, it becomes possible to perform automated driving that reflects Driver A's preferences. For example, it is possible to perform automated driving that realizes Driver A's preferred driving course selection, acceleration, steering, etc.

[0401] The local DB and the extracted driving DB may be switched for use. In this case, too, driving optimized for the area and driving preferred by Driver A can be realized by switching between the local DB and the extracted driving DB.

[0402] <Second application example> Another application example in which the information processing system 1 is applied to automatic driving will be described.

[0403] Driving data is acquired from vehicles traveling in a specific area (say, Area A), and a driving DB for Area A is created. Driving data that yields positive feedback is extracted from this driving DB. The extracted driving data is, for example, data from when the driver or passengers felt the driving was comfortable.

[0404] When Driver A drives within Area A, the driving data in the driving DB for Area A is reflected in the database in which Driver A's driving data is recorded. The database in which Driver A's driving data is recorded is optimized for Driver A, and the driving DB for Area A is optimized for driving in Area A. By integrating such databases, it is possible to build a database in which the driving preferences of Driver A are reflected in the driving that is optimal for Area A.

[0405] By performing re-learning using a database that reflects optimal driving for this area A, it becomes possible to perform automated driving that reflects the preferences of driver A and the optimal driving for area A. For example, it is possible to perform automated driving that realizes the selection of the optimal driving course, acceleration, steering, etc. for driving within area A based on driver A's preferences.

[0406] In addition, the database in which the driving data of driver A is recorded may be switched to use the driving data in the driving DB for area A. In this case, driving optimized for the area and driving preferred by driver A can be realized by switching between the database in which the driving data of driver A is recorded and the driving data in the driving DB for area A.

[0407] <Third application example> An application example in which the information processing system 1 is applied to an AI agent will be described.

[0408] An interaction DB 113 is generated for characters that do not exist in the real world, such as characters from animations, cartoons, etc., or AI characters. In this case, since the interaction DB 113 is for users that do not exist in the real world, it is not a history of actual interactions, but rather an interaction DB 113 created assuming virtual interactions based on the preferences, actions, etc. of a character (assumed to be character A) set by the creator.

[0409] Alternatively, the interaction DB 113 may be created from the interactions of character A within the animation. In this case, since audio data and video data are obtained, the interactions of character A can be extracted to create character A's interaction DB 113. Furthermore, in the case of a cartoon, since text data and image data are obtained, the interactions of character A can be extracted from these data to create character A's interaction DB 113 (hereinafter referred to as character DB).

[0410] Such a character DB is reflected in a DB (assumed to be exchange DB 113A) within user A's information processing system 1. By integrating such databases, a database can be constructed that reflects user A's preferences for characters that he or she likes.

[0411] By relearning using a database that reflects this character DB, user A's AI (information processing system 1) will respond in a manner that resembles character A. For example, the way it speaks and the content of its recommendations will become more like character A's. Thus, user A can virtually experience interacting with his or her favorite character A.

[0412] Furthermore, the character DB and the database of user A may be switched for use. By switching to the character DB, it becomes possible to have a simulated conversation with character A and receive recommendations from character A.

[0413] <Fourth application example> Another application example in which the information processing system 1 is applied to an AI agent will be described.

[0414] For example, a database of an AI agent (information processing system 1) used by a person whom user A aspires to be or admires (user B), such as a celebrity or idol, is reflected in user A's (self) database.

[0415] By relearning using a database that reflects user B's database, user A's AI (information processing system 1) can receive recommendations that reflect user B's interests and preferences.

[0416] Furthermore, the database for user A and the database for user B may be switched for use. For example, the database for user B may be switched to during a specific situation, such as when traveling. In this case, user A can receive recommendations that reflect user B's hobbies and preferences while traveling, and can feel as if he or she is traveling with user B.

[0417] <Fifth application example> Another application example in which the information processing system 1 is applied to an AI agent will be described.

[0418] When user A acts as an avatar in a virtual space, an interaction DB 113 for that avatar is generated. For example, user A may wish to act as a different personality when active in the virtual space, and may set the hobbies, preferences, behavior, etc. of the avatar with the different personality, and interaction information based on the set conditions is recorded in interaction DB 113. The avatar interaction DB 113 (hereinafter referred to as avatar DB) may be created by another user instead of by user A.

[0419] Such an avatar DB is reflected in the DB in the information processing system 1 of user A. By integrating such databases, it is possible to construct a database that reflects user A's preferences for avatars.

[0420] By using a database that reflects this avatar DB to perform re-learning, the avatar can perform activities that resemble those of the avatar set by user A. For example, while maintaining the avatar's behavior before the avatar DB was reflected, it is possible to perform processing such as issuing a warning when the avatar behaves in an unnatural manner compared to the expected avatar, or recommending behaviors that are determined to be appropriate for the avatar.

[0421] User A can become the avatar he or she envisions (desired) and act in the virtual space.

[0422] <Sixth application example> An application example in which the information processing system 1 is applied to a chatbot will be described.

[0423] A chatbot is an automatic conversation program that utilizes artificial intelligence, and is a computer incorporating artificial intelligence that converses on behalf of humans. The information processing system 1 can be applied to the computer side of the chatbot.

[0424] Conversation history from a specific cultural sphere (say, cultural sphere B) is acquired, and a database (referred to as cultural sphere BDB) related to interactions in cultural sphere B is created. For example, there may be databases for multiple cultural spheres, and interactions in cultural sphere B may be extracted from these databases to create the cultural sphere BDB. User A lives in cultural sphere A, which is different from cultural sphere B, and a database (referred to as cultural sphere ADB) related to appropriate interactions has been created in cultural sphere A.

[0425] Let's imagine that user A moves from cultural sphere A to cultural sphere B or goes on a trip. In such a case, the cultural sphere BDB of cultural sphere B is reflected in user A's cultural sphere ADB. By reflecting the database in this way, a database that reflects the culture of cultural sphere B can be constructed.

[0426] By using this database that reflects the culture of cultural sphere B to perform relearning, the AI ​​(information processing system 1) of a chat pod that has already been trained for user A (personal use) can be made into a chat pod that reflects the culture of cultural sphere B. With such a chat pod, for example, behavior unique to cultural sphere B can be added, making it possible to provide user A with interactions unique to cultural sphere B.

[0427] <Seventh application example> An application example in which the information processing system 1 is applied to robot control will be described.

[0428] A database (referred to as Worker ADB) has been created for Worker A, who performs work using a robot. Worker ADB is an AI optimized for Worker A.

[0429] Meanwhile, in workplace B, which is located in a different place from where worker A is working, a database optimized for workplace B (hereinafter referred to as workplace BDB) is constructed. The workplace BDB is AI optimized for workplace B, obtained from the work history of workers who worked in workplace B. For example, if workplace B is a small factory, information optimized for work in such a small space is recorded in the workplace BDB.

[0430] For example, when worker A moves to workplace B and performs work using a robot at that new location, the worker ADB and workplace BDB are integrated. In other words, the AI ​​optimized for worker A is reflected in the AI ​​optimized for workplace B. A database is constructed in which the worker ADB and workplace BDB are integrated, and re-learning is performed using this database.

[0431] Although the robot with the re-trained AI behaves the same as usual from the perspective of User A, it operates in a way that is suited to the environment of the new workplace, Workplace B. Therefore, User A can perform the same tasks at the new workplace as before the move, using the same familiar operations.

[0432] <8th application example> Another application example in which the information processing system 1 is applied to robot control will be described.

[0433] A database (referred to as the "specific task DB") optimized for a specific task has been constructed. A specific task is, for example, control corresponding to a new process. In addition, a database (referred to as the "worker ADB") has been constructed for worker A, who performs work using a robot. Worker ADB is an AI optimized for worker A.

[0434] For example, when worker A controls a robot in a new process, the worker A DB and the specific task DB are integrated. In other words, the specific task DB is reflected in the AI ​​optimized for worker A. A database is constructed in which the worker A DB and the specific task DB are integrated, and re-learning is performed using this database.

[0435] The robot with the re-learned database (AI) applied will behave in a way that is appropriate for the new process, even though it appears to be behaving the same as usual from the perspective of User A. This means that User A will be able to work in the new process with familiar operations, without any sense of discomfort.

[0436] <9th Application Example> A further application example in which the information processing system 1 is applied to robot control will be described.

[0437] A database (referred to as "user DB") optimized for a specific user is constructed. A specific user is, for example, a female user, and the user DB is a database optimized for use by women. In addition, a database (referred to as "worker ADB") is constructed for worker A, who performs work using a robot. Worker ADB is an AI optimized for worker A.

[0438] For example, if worker A is a man, the worker ADB is a database suitable for men operating robots. When such a robot is used by a woman, the worker ADB and user DB are integrated. In other words, the user DB is reflected in the AI ​​optimized for worker A. A database in which the worker ADB and user DB are integrated is constructed, and re-learning is performed using this database.

[0439] Even if the robot before re-learning was optimized for men and difficult for women to use, the robot after re-learning will be easy for women to use as well. Therefore, even if the user of the robot changes, it can be optimized so that it is easy to use for the new user.

[0440] The seventh to ninth application examples are cases in which the information processing system 1 is applied to robot control, but this robot may be an industrial robot or a robot for personal use. For example, the seventh to ninth application examples can be applied by applying the information processing system 1 to a general-purpose helper robot for personal use. Furthermore, like the first and second application examples, the seventh to ninth application examples can also be applied to vehicle control.

[0441] <10th application example> An application example in which the information processing system 1 is applied to information management will be described.

[0442] For example, the following describes a case where the information processing system 1 is applied to an AI that predicts stock prices. For example, when Company A acquires Business X of Company B, data related to Business X is extracted from the database of the AI ​​that predicts Company B's own stock. This extracted data related to Business X is reflected in the DB of the AI ​​(information processing system 1) that predicts stock prices.

[0443] By reflecting data related to Business X, it becomes possible to more accurately predict the stock price after Company A acquires Business X of Company B.

[0444] While the information management described here has been given as an example of managing economic information, it can also be applied to managing information about factories, making predictions to improve work efficiency, formulating work plans, and predicting product shipments. It can also be applied to managing agricultural information, planning pesticide spraying times and amounts, and predicting harvest yields.

[0445] <11th Application Example> An application example in which the information processing system 1 is applied to behavior recognition, authentication, a monitoring system, a security-related system, etc. will be described.

[0446] For example, the following describes an example in which the information processing system 1 is applied to a system that performs authentication processing with a specific person as the authentication target. For example, consider a case in which person A is the authentication target, but person A breaks a bone. In such a case, there will be differences in person A's behavior before and after the fracture, and therefore, there is a possibility that processing using a database that was optimized to authenticate person A before the fracture will not be able to authenticate person A after the fracture.

[0447] A database (hereinafter referred to as a general behavior DB) is constructed to authenticate general behavior. The general behavior DB also records the behavior patterns of people who have broken bones. When it is determined that person A has broken a bone, data on the behavior pattern of the person who has broken a bone is extracted from the general behavior DB, and the extracted data is reflected in the database of the information processing system 1.

[0448] By incorporating a database of the behavioral patterns of people with broken bones into AI optimized for authenticating a specific person (Person A in this case), even if Person A, the person being authenticated, breaks a bone and their behavioral patterns suddenly change, they can be authenticated by comparing them with the general patterns of people with broken bones. Therefore, even if the person being authenticated changes, authentication can be performed without any problems.

[0449] <Information processing system composed of multiple devices> In the first and second embodiments described above, an information processing system is configured by the information processing device 10 or the information processing device 11. However, the information processing system is not limited to this and may be configured by a plurality of devices. FIG. 44 is a diagram showing an example of an information processing system 2 configured by a plurality of devices. As shown in FIG. 44, the information processing system 2 is configured by an information processing device 12 and a data server 15. Furthermore, the information processing device 12 and the data server 15 are connected via a network 30.

[0450] The configuration of the information processing device 12 will be described with reference to FIG. 44. FIG. 44 is a functional block diagram showing the configuration of the information processing device 12. Unlike the information processing devices 10 and 11 according to the first and second embodiments, the information processing device 12 shown in FIG. 44 does not include databases corresponding to a knowledge DB, a recommendation DB, an interaction DB, a learning DB, or the like. These databases are recorded in the data server 15. In this case, the information processing device 12 can obtain information as needed from a data server connected to the network and perform re-learning of the algorithm, etc. Although not shown in FIG. 44, the information processing device 12 is assumed to have a storage unit that stores information necessary for various processes.

[0451] <Hardware configuration> Next, an example of a hardware configuration of the information processing devices 10, 11, and 12 or the user terminal 20 constituting the information processing system 1 according to an embodiment of the present disclosure will be described in detail with reference to Fig. 45. Fig. 45 is a functional block diagram showing an example of a hardware configuration of the user terminal 20 or the information processing devices 10, 11, and 12 constituting the information processing system 1 according to an embodiment of the present disclosure.

[0452] The information processing device 10 constituting the information processing system 1 according to this embodiment mainly includes a CPU 601, a ROM 602, and a RAM 603. The information processing device 10 further includes a host bus 604, a bridge 605, an external bus 606, an interface 607, an input device 608, an output device 609, a storage device 610, a drive 612, a connection port 614, and a communication device 616.

[0453] The CPU 601 functions as an arithmetic processing device and control device, and controls all or part of the operations within the information processing device 10 in accordance with various programs recorded in the ROM 602, RAM 603, storage device 610, or removable recording medium 613. The ROM 602 stores programs used by the CPU 601, calculation parameters, etc. The RAM 603 temporarily stores programs used by the CPU 601, parameters that change as appropriate during program execution, etc. These are connected to each other by a host bus 604 constituted by an internal bus such as a CPU bus. For example, the processing unit 120, analysis unit 130, generation unit 140, output control unit 150, and communication control unit 160 shown in FIG. 3 can be constituted by the CPU 601.

[0454] The host bus 604 is connected to an external bus 606 such as a PCI (Peripheral Component Interconnect / Interface) bus via a bridge 605. In addition, an input device 608, an output device 609, a storage device 610, a drive 612, a connection port 614, and a communication device 616 are connected to the external bus 606 via an interface 607.

[0455] The input device 608 is an operation means operated by a user, such as a mouse, keyboard, touch panel, button, switch, lever, pedal, etc. The input device 608 may be, for example, a remote control means (so-called remote control) using infrared or other radio waves, or an externally connected device 615 such as a mobile phone or PDA that is compatible with the operation of the information processing device 10. The input device 608 is further composed of, for example, an input control circuit that generates an input signal based on information input by the user using the above operation means and outputs the signal to the CPU 601. A user of the information processing device 10, 11, 12 or the user terminal 20 can input various data and instruct processing operations to the information processing device 10, 11, 12 or the user terminal 20 by operating the input device 608.

[0456] The output device 609 is configured with a device capable of visually or audibly notifying the user of acquired information. Such devices include display devices such as CRT display devices, liquid crystal display devices, plasma display devices, EL display devices, and lamps, audio output devices such as speakers and headphones, and printer devices. The output device 609 outputs, for example, results obtained from various processes performed by the information processing devices 10, 11, and 12 or the user terminal 20. Specifically, the display device displays, as text or images, the results obtained from various processes performed by the information processing devices 10, 11, and 12 or the user terminal 20. On the other hand, the audio output device converts audio signals consisting of reproduced voice data, acoustic data, etc. into analog signals and outputs them.

[0457] The storage device 610 is a data storage device configured as an example of a storage unit of the information processing device 10. The storage device 610 is configured, for example, by a magnetic storage device such as an HDD (Hard Disk Drive), a semiconductor storage device, an optical storage device, or a magneto-optical storage device. The storage device 610 stores programs executed by the CPU 601, various data, and the like. For example, the storage unit 110 shown in FIG. 3 can be configured by the storage device 610.

[0458] The drive 612 is a reader / writer for a recording medium, and is built into or externally attached to the information processing device 10. The drive 612 reads information recorded on a removable recording medium 613, such as an attached magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, and outputs the information to the RAM 603. The drive 612 can also write information to the attached removable recording medium 613, such as an attached magnetic disk, optical disk, magneto-optical disk, or semiconductor memory. The removable recording medium 613 may be, for example, a DVD medium, an HD-DVD medium, or a Blu-ray (registered trademark) medium. The removable recording medium 613 may also be, for example, a CompactFlash (registered trademark) (CF), a flash memory, or an SD memory card (Secure Digital memory card). The removable recording medium 613 may also be, for example, an IC card (Integrated Circuit card) equipped with a contactless IC chip, or an electronic device.

[0459] The connection port 614 is a port for directly connecting to the information processing devices 10, 11, 12 or the user terminal 20. Examples of the connection port 614 include a USB (Universal Serial Bus) port, an IEEE 1394 port, and a SCSI (Small Computer System Interface) port. Other examples of the connection port 614 include an RS-232C port, an optical audio terminal, and an HDMI (registered trademark) (High-Definition Multimedia Interface) port. By connecting an external device 615 to this connection port 614, the information processing device 10 can directly obtain various types of data from the external device 615 or provide various types of data to the external device 615.

[0460] The communication device 616 is, for example, a communication interface configured with a communication device or the like for connecting to a communication network 917. The communication device 616 is, for example, a communication card for a wired or wireless LAN (Local Area Network), Bluetooth (registered trademark), or WUSB (Wireless USB). The communication device 616 may also be a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), or a modem for various types of communication. The communication device 616 can transmit and receive signals, for example, between the Internet and other communication devices in accordance with a predetermined protocol such as TCP / IP. The communication network 617 connected to the communication device 616 is configured with a network connected by wire or wirelessly, and may be, for example, the Internet, a home LAN, infrared communication, radio wave communication, satellite communication, or the like.

[0461] The above describes an example of a hardware configuration capable of implementing the functions of the information processing devices 10, 11, and 12 that constitute the user terminal 20 or the information processing system 1 according to an embodiment of the present disclosure. Each of the above components may be configured using general-purpose components, or may be configured using hardware specialized for the function of each component. Therefore, the hardware configuration used can be changed as appropriate depending on the technical level at the time of implementing the embodiment. Note that, although not shown in FIG. 21 , various components corresponding to the information processing devices 10, 11, and 12 that constitute the user terminal 20 or the information processing system 1 are naturally provided.

[0462] It is possible to create a computer program for implementing each function of the information processing devices 10, 11, and 12 constituting the information processing system 1 according to the present embodiment as described above and install it on a personal computer or the like. It is also possible to provide a computer-readable recording medium storing such a computer program. Examples of the recording medium include a magnetic disk, an optical disk, a magneto-optical disk, and a flash memory. The computer program may also be distributed, for example, via a network without using a recording medium. The number of computers that execute the computer program is not particularly limited. For example, the computer program may be executed by multiple computers (e.g., multiple servers) working together.

[0463] The program executed by the computer may be a program that processes in chronological order according to the order described in this specification, or may be a program that processes in parallel or at the required timing, such as when called.

[0464] In this specification, a system refers to an entire device made up of multiple devices.

[0465] The effects described in this specification are merely examples and are not limiting, and other effects may also be present.

[0466] It should be noted that the embodiments of the present technology are not limited to the above-described embodiments, and various modifications are possible within the scope of the present technology.

[0467] The present technology can also be configured as follows. (1) For the algorithm that changes based on the accumulation of first learning data, Re-learning is performed based on the first learning data and specific learning data from the second learning data that forms another algorithm that changes based on the accumulation of learning data. Information processing device. (2) The first training data includes data regarding output information from the algorithm based on input information to the algorithm. The information processing device according to (1) above. (3) The first learning data is based on data accumulated in an environment in which the algorithm is used. The information processing device according to (1) or (2). (4) The specific learning data is learning data obtained by extracting specific learning histories that meet predetermined conditions from the learning histories of the algorithms based on a database that records data related to input information for the other algorithms. The information processing device according to any one of (1) to (3). (5) the first training data is training data based on input information in a first environment using the algorithm; The second training data is training data based on input information in a second environment using the other algorithm. The information processing device according to any one of (1) to (4). (6) A portion of the first training data is replaced with at least a portion of the specific training data, and the re-training is performed. The information processing device according to any one of (1) to (5). (7) The specific learning data is converted into a predetermined data format and then re-learned. The information processing device according to any one of (1) to (6). (8) The influence of the first learning data and the specific learning data is adjusted, and the re-learning is performed. The information processing device according to any one of (1) to (7). (9) The first learning data and the second learning data are learning data that output predetermined recommendation information for input information. The information processing device according to any one of (1) to (8). (10) The specific training data is encrypted data. The information processing device according to any one of (1) to (9). (11) The specific learning data is data converted into a predetermined feature amount. The information processing device according to any one of (1) to (10). (12) The specific learning data is data received from another information processing device. The information processing device according to any one of (1) to (11). (13) The re-learning is performed by another information processing device. The information processing device according to any one of (1) to (12). (14) Sending the first learning data to the other information processing device; The other information processing device is caused to perform the relearning. The information processing device according to (13) above. (15) The information processing device For the algorithm that changes based on the accumulation of first learning data, Re-learning is performed based on the first learning data and specific learning data from the second learning data that forms another algorithm that changes based on the accumulation of learning data. Information processing methods. [Explanation of symbols]

[0468] 1,2 Information processing system, 10,11,12 Information processing device, 15 Data server, 20 User terminal, 30 Network, 40 Learning history data, 110 Memory unit, 117 Recommendation database, 118 Memory unit, 120 Processing unit, 121 Update unit, 122 Extraction unit, 123 Determination unit, 124 Correction unit, 125 Learning unit, 128 Processing unit, 130 Analysis unit, 131 Analysis unit, 140 Generation unit, 141 Confirmation information generation unit, 142 Recommendation information generation unit, 143 Generation unit, 144 Confirmation information generation unit, 145 Recommendation information generation unit, 150 Output control unit, 160 Communication control unit, 210 Communication control unit, 220 Output control unit, 301 to 304 DB switching unit, 331 Key, 351 Feature Transformation Unit, 352 Update Parameter Calculation Unit

Claims

1. First learning data relating to first input information and first output information based on a first algorithm suitable for a first user for the first input information is integrated with second learning data relating to second input information and second output information based on a second algorithm suitable for a second user for the second input information, and re-learning is performed using the integrated learning data as new first learning data to update the first algorithm. Information processing device.

2. The second learning data to be integrated is learning data obtained by extracting specific learning histories that meet predetermined conditions. The information processing device according to claim 1 .

3. the first learning data is learning data based on input information input by the first user, and the first algorithm is an algorithm generated based on accumulation of the first learning data; the second learning data is learning data based on input information input by the second user, and the second algorithm is an algorithm generated based on accumulation of the second learning data; The information processing device according to claim 1 .

4. A portion of the first training data is replaced with at least a portion of the second training data, and the re-training is performed. The information processing device according to claim 1 .

5. The second learning data is converted into a predetermined data format and then re-learned. The information processing device according to claim 1 .

6. The first learning data and the second learning data are learning data that output predetermined recommendation information for input information. The information processing device according to claim 1 .

7. The encrypted second learning data is supplied, and the decrypted second learning data and the first learning data are integrated. The information processing device according to claim 1 .

8. The second learning data is data converted into predetermined features. The information processing device according to claim 1 .

9. The re-learning is performed by another information processing device. The information processing device according to claim 1 .

10. The information processing device First learning data relating to first input information and first output information based on a first algorithm suitable for a first user for the first input information is integrated with second learning data relating to second input information and second output information based on a second algorithm suitable for a second user for the second input information, and re-learning is performed using the integrated learning data as new first learning data to update the first algorithm. Information processing methods.

Images