JP7268547B2 - Information processing device and program - Google Patents

Description

The present invention relates to an information processing apparatus and program.

In Patent Literature 1, biometric information (biological reactions) of users during content reproduction is collected for a plurality of users, and users who exhibit biometric reactions similar to those of a target user are identified as similar users. Techniques for recommending highly rated content to target users have been disclosed.

JP 2009-134669 A

In Patent Document 1, the amount of hemoglobin contained in blood, the amount of blood flow, the amount of sweat, or the pulse rate is used as biological information, but these are physical quantities (waveform data) that easily change due to various factors. In Patent Literature 1, since similarity between users is determined by directly comparing such waveform data, similarity (similarity) between users cannot be obtained with high accuracy. For example, if the waveform data of two users are slightly different in shape or size, similarity cannot be determined with sufficient accuracy. For this reason, similar users cannot be specified with high accuracy, and as a result, the accuracy of content recommendation information may not be sufficient.

The present disclosure has been made in order to solve such a problem. The purpose is to provide a device and a program.

An information processing device according to an aspect of the present disclosure includes:
a storage unit that stores user's biological information measured during content reproduction in association with the user and the content;
a generation unit that generates category data including one or more reaction categories indicating the type of user reaction to content based on the biometric information;
a calculation unit that calculates a degree of similarity indicating either similarity between users or similarity between contents based on the category data and category related data indicating the degree of relatedness between reaction categories;
a creation unit that creates recommendation information based on the similarity;
Prepare.

A program according to an aspect of the present disclosure is
a process of storing the user's biometric information measured during content reproduction in association with the user and the content;
a process of generating category data including one or more reaction categories indicating the type of reaction of the user to the content based on the biometric information;
a process of calculating a degree of similarity indicating either similarity between users or similarity between contents based on the category data and category related data indicating the degree of relatedness between reaction categories;
a process of creating recommendation information based on the similarity;
run on the computer.

According to the present disclosure, it is possible to provide an information processing device and a recommendation program for accurately creating recommendation information based on a user's natural reaction during content reproduction.

1 is a block diagram showing the overall configuration of a system according to Example 1; FIG. 1 is a block diagram showing the configuration of a terminal device according to Example 1; FIG. 4 is a diagram illustrating an example of a user's head wearing headphones and a coordinate system according to the first embodiment; FIG. 4 is a diagram illustrating an example of output information of a biosensor according to Example 1; FIG. 1 is a block diagram showing the configuration of a recommendation device according to Example 1; FIG. 4 is a diagram illustrating an example of a data configuration of a biometric information DB according to Example 1; FIG. 6 is a flow chart showing the flow of recommendation information creation processing according to the first embodiment; 6 is a flow chart showing the flow of first category data generation processing according to the first embodiment; 4 is a diagram illustrating an example of a data configuration of a section information DB according to Example 1; FIG. FIG. 5 is a diagram showing an example of rule data according to the first embodiment; FIG. 4 is a diagram illustrating an example of a data configuration of a user reaction DB according to Example 1; FIG. 7 is a flow chart showing the flow of second category data generation processing according to the first embodiment; FIG. 5 is a diagram showing an example of user reaction vectors according to Example 1; 4 is a diagram showing an example of category-related data according to Example 1; FIG. FIG. 10 is a diagram showing an example of rule data customized by a user according to the first embodiment; FIG. FIG. 10 is a diagram illustrating an example of a screen displaying recommendation information according to the first embodiment; FIG. FIG. 7 is a diagram showing an example of a screen displaying recommendation information when a target user specifies content conditions according to the first embodiment; FIG. 11 is a diagram showing an example of a screen displaying non-recommended content together with recommended content according to modification 1-1; 10 is a flow chart showing the flow of recommendation information creation processing according to the second embodiment; FIG. 11 is a diagram showing an example of a content vector according to Example 2; FIG. FIG. 11 is a diagram showing an example of a screen displaying recommendation information according to the second embodiment; FIG. FIG. 11 is a diagram illustrating an example of a screen displaying recommendation information according to Example 3; FIG. 12 is a block diagram showing the configuration of a recommendation device according to Example 4; FIG. 14 is a diagram showing an example of the data configuration of a music information DB according to Example 4; FIG. 15 is a diagram illustrating an example of frequency components of a section to be processed according to Example 5; FIG. 13 is a diagram showing an example of rule data according to Example 5; FIG. 11 is a diagram showing an example of frequency components during reproduction and non-reproduction according to Example 6; FIG. 12 is a diagram showing an example of rule data according to Example 7;

Examples of specific embodiments of the present disclosure are described in detail below with reference to the drawings. In each drawing, the same elements are denoted by the same reference numerals, and redundant description will be omitted as necessary for clarity of description.

<Example 1>
FIG. 1 shows a configuration diagram of the entire system of the present disclosure. A recommendation device 5 , a content distribution server 4 , and base station devices 31 and 32 are connected via a network 80 . The recommendation device 5 is also called an information processing device. The base station devices 31 and 32 provide wireless communication by mobile phone networks such as 4G (Generation) and 5G to the terminal device 11 and other devices.

The terminal devices 11 to 14 are, for example, smart phones, tablet terminals, smart watches, personal computers, and the like. In the example of FIG. 1, the user u1 wears the biosensor 21 and operates the terminal device 11, the user u2 wears the biosensor 22 and operates the terminal device 12, and the user u3 has the biosensor 23 built-in. A case is shown in which the terminal device 13 is operated by the user u4 and the user u4 wears the biosensor 24 and operates the terminal device 14 . Also, the terminal devices 11 and 12 and the biosensor 22 are under the control of the base station device 31 and can communicate with each other, and the biosensor 21 can communicate with the terminal device 11 . Also, the terminal devices 13 and 14 are under the control of the base station device 32 and can communicate with each other, and the biosensor 24 can communicate with the terminal device 14 . However, the system is not limited to these configurations.

The content distribution server 4 provides one of a plurality of corresponding contents in response to a request from each of the terminal devices 11 and the like via the network 80 . The terminal device 11 or the like reproduces the content received from the content distribution server 4 or the content stored in the terminal device 11 or the like. Each of the users u1 to u4 of the terminal devices 11 to 14 views (views) the reproduced content. In the first embodiment, the content is typically music (songs), but may be video content such as movies, TV programs, music videos, and the like.

FIG. 2 is a diagram showing the functional configuration of the terminal device 11. As shown in FIG. The terminal devices 12 and 14 are the same as the terminal device 11, and the terminal device 13 is only partially different from the terminal device 11 in configuration, so illustration and detailed description thereof will be omitted.
The terminal device 11 includes a first communication section 110 , a second communication section 120 , a control section 130 , a storage section 140 , a display section 150 , an operation section 160 and an audio output section 170 . The first communication unit 110 is connected to a mobile phone network such as 4G or 5G, a wireless LAN (Local Area Network), or the like, and communicates with the content distribution server 4 and the recommendation device 5 . The second communication unit 120 has a short-range wireless communication function such as Bluetooth (registered trademark), and communicates with the biosensor 21 and headphones (not shown). The control unit 130 includes a CPU (Central Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), etc., and controls each unit. The display unit 150 is a liquid crystal display, an organic EL display, or the like. The display unit 150 displays information necessary for the user u1 to perform an operation, and also displays the content if the content is video. The operation unit 160 receives an operation (input) from the user. Note that the display unit 150 and the operation unit 160 may be configured integrally using a touch panel. The audio output unit 170 is a speaker or the like, and outputs music and audio of content. Note that when headphones or earphones are designated as the audio output destination, the audio output unit 170 does not output audio. The storage unit 140 stores data measured by the biosensor 21 and content data as needed. The storage unit 140 also stores programs for operating the terminal device 11 . Control unit 130 reads a program from storage unit 140 and executes it.

Returning to the description of FIG. The biosensors 21 to 24 are devices that detect the user's posture, movement, displacement, speed, acceleration, etc. during content reproduction. In addition, it may have a function of detecting blood flow, pulse, heart rate, respiration, electroencephalogram, and the like. For example, the biosensor 21 is an earphone or headphone incorporating an acceleration sensor or an angular velocity sensor. Alternatively, the biosensor 21 may be a smart watch, a ring-type device, a glasses-type display, or other wearable device that incorporates an acceleration sensor or an angular velocity sensor. The biosensor 21 or the like measures biometric information from the user and outputs the biometric information.

The terminal device 13 incorporates a biosensor 23 (for example, an acceleration sensor). In this way, the terminal device may incorporate the biosensor. The user u3 holds the terminal device 13 in his/her hand or puts it in a pocket or the like and uses it in close contact with the human body. In this case, the biosensor 23 is one component of the terminal device 13, and the biometric information of the user u3 measured by the biosensor 23 is sent to the base station device 32 and the network 80 by the first communication unit 110 in the terminal device 13. is sent to the recommendation device 5 via the The biosensors 21 and 24 each have an interface such as Bluetooth (registered trademark), and can communicate with the terminal devices 11 and 14, respectively. For example, the user's biological information measured by the biological sensor 21 is transmitted to the recommendation device 5 via the terminal device 11 , the base station device 31 and the network 80 . Moreover, the biosensor 22 has communication functions such as 4G and 5G, and can directly communicate with the base station device 31 . In this case, the biosensor 22 does not need communication via the terminal device 11 . Also, if the biosensor 22 or the like has a content reproduction function, for example, if it is a headphone or glasses-type display equipped with a communication function such as 4G or 5G, the terminal device can be omitted. That is, the terminal device and the biosensor may be integrated or separated.

In the following description, headphones with an acceleration sensor are used as biosensors unless otherwise specified. The headphones may also include an angular velocity sensor (gyro sensor). This headphone communicates with the terminal device via Bluetooth (registered trademark). In the following description, an example in which the user u1 wears the biosensor 21 and operates the terminal device 11 to reproduce content or make a request is used, but the same applies to combinations of other users, biosensors, and terminal devices. Of course it is. A user to whom the recommendation device 5 recommends content is called a “target user”. In the following examples, user u1 is the target user unless otherwise stated.

The user u1 operates the terminal device 11 so that the terminal device 11 sends a request to the content distribution server 4 and streams the desired content. The user u1 wears headphones and listens to the content, and the movement of the human body of the user u1 during content reproduction is output as biometric information data by the biometric sensor 21, and the terminal device 11 acquires it via the second communication unit 120. do. Further, the terminal device 11 transmits the biometric information data to the recommendation device 5 via the first communication unit 110 . Incidentally, biometric information data may be simply referred to as "biometric information".

FIG. 3 is a diagram illustrating the head of a user wearing headphones and a coordinate system. In FIG. 3, the direction connecting both ears is the x-axis, the direction from the midpoint connecting the ears to the tip of the nose is the y-axis, and the direction from the midpoint connecting the ears to the top of the head is the z-axis. This coordinate system is merely an example, and is not limited to this.

The headphone incorporates an acceleration sensor and can detect acceleration in three axes (x, y, z directions). Acceleration, velocity, and displacement can also be converted to each other by setting appropriate preconditions. These can be said to be data indicating the body movements of the user u1. In the first embodiment, to simplify the explanation, each biosensor outputs data related to the displacement of a predetermined part of the human body (for example, the tip of the nose or the top of the head).

FIG. 4 shows an output example of the biosensor 21. As shown in FIG. The horizontal axis t is the elapsed time from the beginning of a song (song A). The three graphs show the displacement of the human body in the x, y and z directions. In the following description, the amplitude (absolute value) of x-axis displacement is Ax, the amplitude (absolute value) of y-axis displacement is Ay (or Ay'), and the amplitude (absolute value) of z-axis displacement is Az. . In the example shown in FIG. 4, Ax is the largest, Ay is the second largest, and Az is the smallest in the section from the start of music A (t=0) to the passage of T1 time (for example, 30 seconds). Also, the period of output change is relatively long. From this output signal, it can be seen that the user mainly moves his head slowly on the xy plane. Such head movement is a so-called "horizontal glue" reaction. As used herein, the term "groove" refers to a reaction in which the user periodically moves his or her body to music.

On the other hand, when T1 has passed from the start of the music, Ax becomes small and Ay' and Az become very large. In addition, the period of output change is shortened. Therefore, it can be seen that the user is moving the head rapidly and violently, mainly in the yz plane. Such movement of the head is a so-called "vertical glue" reaction.

For example, song A has a slow intro part (part) from the beginning to T1, a sudden change in the tone of the song at T1, and a fast-paced "A melody" part (part) after T1. If so, it can be seen from the data shown in FIG. 4 that the user generally moved his/her body in accordance with the characteristics of each part of the music.

The terminal device 11 receives a user ID (user identifier), information on the content received from the content distribution server 4 and reproduced (content ID, content name, artist name, producer name, etc.), and biometric information that is sensor output information. are associated with each other and transmitted to the recommendation device 5 . Alternatively, the terminal device 11 transmits the user ID, the biological information, and the date and time when the biological information was measured to the recommendation device 5, and the content distribution server 4 transmits information such as the content ID, the user ID of the distribution destination, and the distribution date and time. It may be transmitted to the recommendation device 5 . In this case, the recommendation device 5 compares the data received from the terminal device 11 with the data received from the content distribution server 4, so that it is possible to grasp who viewed what content when and how they reacted. Note that, in the first embodiment, a description is given using a user ID that identifies a user, but a terminal ID that identifies a terminal device may also be used. For example, a terminal ID may be used in an environment where basically one terminal device is used by one user, such as a smartphone. In other words, both the user ID and the terminal ID identify a subject (user subject) who uses the content, and they are sometimes collectively called a user subject ID (user subject identifier). The biosensor or the terminal device may output the sensor output (waveform data) as shown in FIG. .

The configuration of the recommendation device 5 is shown in FIG. The recommendation device 5 is composed of a communication section 510 , a processing section 520 and a storage section 530 . The recommendation device 5 is connected to an input device 540 (keyboard, mouse, etc.) for operation by the administrator, and a display device 550 (liquid crystal display, etc.) for providing information to the administrator. The communication unit 510 is an interface circuit that communicates with the outside of the recommendation device 5 . The processing unit 520 is a processor that controls each component of the recommendation device 5, that is, a control device. The processing unit 520 includes a generation unit 521 that generates category information (category data), a calculation unit 522 that calculates similarity between users, and a creation unit 523 that creates recommendation information. For this reason, in the following description, when any one of the generating unit 521, the calculating unit 522, and the creating unit 523 executes a process, it may be collectively described as "performed by the processing unit 520". Also, the “generating unit” may be called the “category generating unit”, the “calculating unit” may be called the “similarity calculating unit”, and the “creating unit” may be called the “recommendation information creating unit”.

The storage unit 530 is a storage device that stores a biological information DB 531 (sensor information DB), a user reaction DB 532, a section information DB 533, category-related data 534, rule data 535, and programs 536. FIG. Then, the recommendation device 5 stores the sensor output information (biological information) received from the terminal device 11 or the like in the biological information DB 531 of the storage unit 530 in association with each of the plurality of contents. Note that the biometric information stored in the biometric information DB 531 in this embodiment is the biometric information of the user at the time of content reproduction. The biometric information at the time of content reproduction is also called "first biometric information".

An example of the data configuration of the biometric information DB 531 is shown in FIG. The user ID, content ID, playback start date and time, playback position, and x output, y output, and z output, which are displacement or acceleration in the three axial directions, are associated and stored. The playback position is the elapsed time from the beginning of the content, and is recorded (x output, y output, z output) at each biometric information sampling period (30 milliseconds in the example shown in this figure). As shown in the figure, sensor outputs are stored when each of a plurality of users reproduces content. The data stored in the biometric information DB 531 corresponds to waveform data of biometric information.

The displacement (acceleration) data (x output, y output, z output) in the three axial directions are independent data, and can be said to be three pieces of biological information, that is, three types of biological information. In this embodiment, an example in which recommendation information is created based on displacement data in three directions will be described, but displacement data in two directions may also be used. That is, recommendation information may be created based on multiple types (two or more types) of biometric information. Also, recommendation information may be created based on one type of biometric information (for example, x output only). However, by using multiple types of biometric information, it is possible to increase the types of reaction categories, which will be described later, so that the user's various reactions to the content can be reflected in the category data with higher accuracy. In the biometric information DB 531, a combination of three fields, user ID, content ID, and playback start date/time, is the main key of the table. That is, one data (one data unit) can be specified by specifying the user ID, content ID, and playback start date and time. As shown in the example of FIG. 6, when a certain user (User1) reproduces a certain content (C1) twice, two data are recorded.

Return to FIG. The program 536 is a computer program in which the recommendation information creation process (recommendation method) according to the first embodiment is implemented. Here, the processing unit 520 loads the program 536 into a memory (not shown) and executes the program 536 . Thereby, the processing unit 520 realizes the functions of the generation unit 521 , the calculation unit 522 and the creation unit 523 .

The user reaction DB 532 is an example of category data, and is information indicating a reaction category for each section of users and content. Category data includes one or more reaction categories. Here, the reaction category is a category (classification, type, type) related to user's reaction to the content. The reaction categories according to the first embodiment include, but are not limited to, combinations of types and strengths of user reactions. The section information DB 533 is information defining a plurality of sections obtained by dividing the content along the time axis. The category related data 534 is information indicating the degree of relatedness between reaction categories. The rule data 535 is information indicating rules for generating reaction categories from biometric information. User reaction DB 532, section information DB 533, category related data 534, and rule data 535 will be described later.

The generation unit 521 generates category data in which biometric information is associated with one of a plurality of reaction categories. That is, the generation unit 521 generates category data including one or more reaction categories based on biometric information. The generation unit 521 in this embodiment uses a predetermined reaction category that distinguishes data in two or more directions based on biological information that indicates the movement of the user's body in (related to) two or more directions, Generate categorical data. That is, category data is generated using various reaction categories based on multiple types of biological information. In particular, the generation unit 521 generates a reaction category for each section, and generates category data including the reaction categories arranged in order of section.

The calculation unit 522 calculates a degree of similarity indicating similarity between users based on the category data generated by the generation unit 521 and the category-related data 534 . In particular, the calculation unit 522 calculates such that when the types of reaction categories match, the degree of similarity is higher than when they do not match.

The creation unit 523 creates recommendation information based on the degree of similarity. The creating unit 523 according to the present embodiment selects recommended content based on the degree of similarity from among a plurality of pieces of content, creates recommended information including information about the selected recommended content, and sends it to the target user. Present recommendation information. For example, the creating unit 523 identifies, from among a plurality of users, a “similar user” that is another user having a relatively high degree of similarity to the target user, and selects a target user from the content that has already been reproduced by the similar user. The user selects unreproduced content as recommended content.

Alternatively, the creating unit 523 may select a plurality of recommended candidate contents based on the degree of similarity from among the plurality of contents. In this case, for example, the creation unit 523 selects, as recommended content, content that matches (has the same reaction category) the reaction category indicated by the target user in the past, from among the plurality of recommended content. Further, the creation unit 523 selects, as recommended content, content for which a similar user has indicated a predetermined reaction category, from among the plurality of recommended content. Also, the creation unit 523 selects, as recommended content, content matching the category specified by the target user from among the plurality of recommended content.

The flow of recommendation information creation processing in the recommendation device 5 is shown in the flowchart of FIG. This process is executed at a predetermined timing. Specifically, the recommendation device 5 executes this process in a batch process at a predetermined cycle (for example, every hour), or executes it at the timing when the user of the terminal device requests the recommendation information.

First, a request requesting recommendation information (also called a recommendation request or a recommendation information request) is transmitted from the terminal device 11 to the recommendation device 5 at an arbitrary timing. For example, when the user u1 logs in to a site by operating the terminal device 11, or when the user u1 performs an operation for displaying recommended information using the terminal device 11, the terminal device 11 operates as a recommendation device. Send a recommendation request to 5. The recommendation request includes the user ID or terminal ID of the user who uses the terminal device (the user who requests the recommendation information). A user who requests recommendation information is also called a target user. If specified by the target user, the recommendation request further includes conditions for the content desired by the target user (also called content conditions or content narrowing conditions).

Upon receiving the recommendation request, the recommendation device 5 determines whether recommendation information for the target user has already been created by batch processing or the like. Specifically, the processing unit 520 determines whether or not the storage unit 530 has recommendation information (content ID of recommended content, etc.) associated with the user ID of the target user. If the target user's recommendation information exists, the processing unit 520 reads it from the storage unit 530 and transmits the recommendation information to the terminal device 11 via the communication unit 510 . If the target user's recommendation information does not exist, the recommendation device 5 executes the process of the flowchart in FIG. 7 to create recommendation information.

In step S<b>100 , the generation unit 521 (processing unit 520 ) acquires biometric information data to be processed from the biometric information DB 531 of the storage unit 530 . Specifically, the generation unit 521 acquires unprocessed data as a processing target from the data stored in the biometric information DB 531 . Note that the biological information DB 531 of the storage unit 530 stores data received (acquired) from the terminal devices 11 to 14, and the generation unit 521 acquires data from the storage unit 530. It can also be said that biometric information data is acquired from 14.

In step S110, the generation unit 521 generates category data by one of the following methods based on biometric information data. It can be said that the generation unit 521 converts the biometric information data into category data. The generation unit 521 stores the generated category data in the user reaction DB 532 of the storage unit 530 .

(A1: First category data generation method)
FIG. 8 is a flow chart showing details of the first category data generation method in step S110. FIG. 8 shows the process of creating category data relating to one of the biometric information data processed in step S100 (data obtained when a certain user viewed certain content once). there is When a plurality of data are to be processed, these processes may be repeatedly executed.

In step S200, the generation unit 521 determines whether or not a section has been created for the content to be processed. Sections have already been created for content that has been processed in the past, and are registered in the section information DB 533 . A new content that has never been processed until now has no section created and is not registered in the section information DB 533 . That is, the generation unit 521 determines whether or not the content to be processed is registered in the section information DB 533 of the storage unit 530 . If the section has been created (S200: Yes), the process proceeds to step S220, and if the section has not been created (S200: No), the process proceeds to step S210.

In step S210, the generation unit 521 creates a section of content to be processed. A section is obtained by dividing one content into a plurality of sections in the time direction. For example, dividing a 200-second long content into sections every 4 seconds creates 50 sections. The generation unit 521 stores the created section information in the section information DB 533 of the storage unit 530 . FIG. 9 is a diagram showing an example of the data configuration of the section information DB 533. As shown in FIG. The section information DB 533 includes fields for content ID, section number (section ID), start point, and end point. The start point and end point are the elapsed time from the beginning of the content, and are in seconds in the example of FIG. For example, the section "2" of the content ID "C1" indicates that it is from "4" seconds to "8" seconds after the beginning of the content. Note that units smaller than seconds (such as milliseconds) may be used. In this embodiment, to simplify the explanation, each section has the same length, but each section may have a different length. Also, the length of the section may be different for each content. It is desirable to set the length of the section to a time during which similar user reactions are expected to continue. For example, if the content is a piece of music with a tempo of 120, the length of one beat corresponds to 0.5 seconds (500 milliseconds). When a user moves their body while listening to music, it is generally assumed that they move their body in the same pattern for about 4 to 8 beats. It is assumed that the reaction will continue. Of course, the tempo is different for each piece of music, but since users tend to repeat the same motions a certain number of times, there is a high possibility that the body motions will be the same for several seconds. Therefore, the length of one section should be set to about several seconds to ten seconds. The time (interval length) in which the user reaction is expected to continue to some extent is considerably longer than the sampling time (cycle) of the original biometric information. Therefore, one section contains a large number of biological information data (sample data, waveform data). In this embodiment, reaction categories are generated by a method described later based on a large number (that is, plural) of biometric information data included in one section. In the example shown in FIG. 9, the content "C1" has "M1" sections, and the content "C2" has "M2" sections. Generally, the number of sections differs for each content, but the number of sections may be the same for all contents.

In step S220, the generation unit 521 selects one unprocessed section as a section to be processed. Normally, the sections to be processed may be selected one by one in ascending order of section numbers.

In step S230, the generation unit 521 generates a category (reaction category) of the section based on the biometric information data of the section to be processed. Specifically, reaction categories are created by one of the following methods.

(A11) Rule-Based Method The first method is a method of creating reaction categories based on preset rules. Specifically, an administrator or the like of the recommendation device 5 sets rules and stores them in the rule data 535 of the storage unit 530 . FIG. 10 shows an example of rule data 535 in the first method. The rule data 535 is composed of fields of category ID, category name, and biological information condition. The rule data 535 in this first method may be called a "rule table".

In FIG. 10, 12 types of response categories with category IDs "0" to "11" are defined. The reaction categories shown in the example of this figure are composed of three large categories (large category, top category) of "horizontal glue", "vertical glue", and "rotary glue". These are reactions that differ in the directionality of the user's body movement, and are so-called "types of grooves to music." Further, three categories of "weak," "medium," and "strong" are defined for each of "horizontal lamination," "vertical lamination," and "rotational lamination," depending on the strength of the reaction. For example, when "horizontal glue" is called a major category or major category, the corresponding reaction categories with category IDs "1" to "3" may be called medium categories, medium categories, subcategories, and the like. Here, the "horizontal glue" is a response that causes the largest lateral movement of the body (the x-axis direction in FIG. 3). This is a reaction that appears relatively often when users listen to music in genres such as pops, rock, and jazz. "Vertical glue" is a response that involves a large amount of body movement in the vertical direction (the z-axis direction in FIG. 3) or the front-back direction (the y-axis direction in FIG. 3). It can be said that the lateral movement of the body (the x-axis direction in FIG. 3) is the smallest response. This reaction is relatively common when users listen to music in genres such as punk rock, heavy metal, and hardcore. "Rotating glue" is a reaction indicating a movement that draws an arc in the xy plane of FIG. "No reaction" of category ID "0" is a state in which body movement is very small. The category ID “10” of “strong paste” is a category indicating a reaction of the body moving violently, although the direction cannot be specified. "Others" of category ID "11" is a category indicating reactions that do not fall under any of the above categories.

In FIG. 10, Ax is the amplitude of displacement in the x-axis direction, Ay is the amplitude of displacement in the y-axis direction, and Az is the amplitude of displacement in the z-axis direction. θ1 to θ3 are predetermined threshold values and satisfy θ1<θ2<θ3. Fy is the frequency of vibration in the y-axis direction, and Fz is the frequency of vibration in the z-axis direction. Also, γ is a predetermined threshold. φx is the phase of vibration in the x-axis direction, and φy is the phase of vibration in the y-axis direction. Also, "∩" indicates an AND condition.

For example, the condition “Ax<θ1∩Ay<θ1∩Az<θ1” for the category ID “0” indicates that “Ax is less than θ1, Ay is less than θ1, and Az is less than θ1”. In addition, the condition “θ1≦Ax<θ2 ∩ Ax>Ay ∩ Ay>Az” for the category ID “1” indicates that “Ax is greater than or equal to θ1 and less than θ2, and Ax is greater than Ay and Ay is greater than Az”. show. That is, as a rule for judging "horizontal glue", the magnitude relation among Ax, Ay, and Az is used.

Further, for example, the condition “θ1≦(Ay, Az)<θ2∩Ax<Ay∩Ax<Az∩(Fy,Fz)>γ” for the category ID “4” is “Ay and Az are greater than or equal to θ1 and less than θ2, and Ax is less than Ay, Ax is less than Az, and Fy and Fz are greater than γ'. In other words, as a rule for judging "longitudinal glue", in addition to the magnitude relationship of Ax, Ay, and Az, the frequency of vibration in the y-axis direction and z-axis direction is higher than a threshold value.

Further, for example, the condition ``θ1≦(Ax, Ay)<θ2 ∩ Ay>Az ∩ |φx-φy| is smaller than Ay, and the difference (absolute value) between the phase of vibration in the x-axis direction and the phase of vibration in the y-axis direction is close to π/2 (90 degrees). For example, the phase difference may be within a predetermined range (for example, a range of 70 degrees to 110 degrees) around π/2 (90 degrees). That is, as a rule for judging "rotation glide", the difference between the phase of vibration in the x-axis direction and the phase of vibration in the y-axis direction is used in addition to the magnitude relationship of Ax, Ay, and Az.

While referring to the biometric information DB 531 and the section information DB 533, the generation unit 521 generates Ax, Ay, Az , Fy, Fz, φx, and φy are calculated. For example, the generation unit 521 calculates the maximum value or representative value (average value, median value, etc.) of the x output in the section to be processed, and sets it as Ax. The generation unit 521 also extracts the frequency components of the y output using a method such as Fourier analysis, and sets the fundamental frequency (the strongest frequency or the lowest peak frequency) among them to Fy. The generation unit 521 also sets the phase difference (φx−φy) when the x output and the y output are each approximated by a sine wave. Then, the generating unit 521 collates the calculated feature quantity with the condition part of the rule data 535 shown in FIG. Note that the number of category IDs (types of reaction categories) is not limited to 12 shown in FIG. A user can memorize (memorize) these types of response categories relatively easily. Therefore, as will be described later, the user can easily specify (input) a character string (symbol) indicating a reaction category when specifying a content condition or a user condition. Also, if there are several to several tens of reaction categories, it is possible to sufficiently reflect various reactions of users to content in the category data, and to accurately calculate the degree of similarity between users. As described above, in this embodiment, a moderate number (moderate type) of reaction categories are set so as to achieve both user operability when designating content conditions and user conditions and accuracy of similarity between users. I am using

Also, in the example shown in FIG. 10, the rule data 535 defines conditions related to three types of biometric information such as (x output, y output, and z output), but the number of types of biometric information used in the rule data 535 is arbitrary. and may be more or less than three types. For example, it is possible to use only x output (only one type) as the biometric information used in the rule data 535, and generate a reaction category using conditions relating to the amplitude Ax of the x output and the fundamental frequency Fx of the x output. In this case, it is not possible to generate reaction categories such as "horizontal glue" and "vertical glue" that distinguish the directionality of the user's body movement. It is possible to classify according to the strength of It is also possible to generate reaction categories that distinguish the period (frequency) of the user's body movement, such as "quick glide" and "slow glide". The frequency used for the condition of the biological information is not limited to the fundamental frequency, and information on other relatively dominant frequency components (harmonic components, etc.) may be used. In addition, when a certain movement pattern continues in the same section, it is regarded as "constant and sustained glue", and when various movement patterns are mixed in the same section, it is regarded as "rapid change". It is also possible to generate reaction categories that distinguish the continuity, regularity, intermittence, diversity, etc. of body movements. In addition, not only the same section but also neighboring sections (temporally adjacent sections, etc.) may be compared to determine the persistence of the body movement, and reaction categories may be generated accordingly. In this way, category data may be generated from one type of biometric information.

(A12) Method using machine learning The second method is a method using machine learning. When using the second method, learning biological information data (learning data) is stored in the storage unit 530 before category determination is performed, and in advance (at an arbitrary timing before step S110 is executed) , the processing unit 520 generates (learns) parameters (classifiers) for determining reaction categories using a known machine learning technique. Various techniques such as neural networks, deep learning, decision trees, support vector machines, and Bayesian classifiers can be used as machine learning techniques. For example, if the section length is 4 seconds and (x output, y output, z output) are collected every 30 milliseconds, one section contains about 133 (4000/30) sample data (waveform data). be Since one piece of sample data includes three kinds of data, when the sample data is directly input to the classifier, the input (input vector) of the classifier has 4000/30×3=400 dimensions. Also, the output (output vector) of the classifier is "12", which is the same as the number of reaction categories shown in the rule data 535 of FIG. In other words, it suffices to prepare the outputs of classifiers for the number of response categories to be generated, and to learn (train) the classifier so as to output the probability that the input data corresponds to the response category. For example, a total of 120,000 pieces of data for 12 kinds of reaction categories, 10,000 pieces for each reaction category, are prepared for learning, and each data is given a correct reaction category (teaching data), and then classified. Learn instruments. After completing the learning of the classifier, the processing unit 520 stores the learned classifier (learned model (not shown)) in the rule data 535 . Specifically, classifier parameters (for example, the network configuration and weighting coefficients of the neural network) are stored in the rule data 535 . Then, in step S230, the generation unit 521 reads the learned classifier from the rule data 535 and inputs the biometric information of the section to be processed to the classifier. Then, the generation unit 521 may determine the reaction category with the largest output value among the outputs of the classifiers corresponding to each reaction category as the reaction category of the section to be processed. Alternatively, classifier outputs corresponding to 11 reaction categories excluding "others (category ID "11")" are prepared and learned, and in step S230, the maximum value among the 11 outputs is less than a predetermined value. , the reaction category may be determined as "other".

Next, in step S240, the generation unit 521 determines whether or not the processing of all sections of the content has been completed. If the processing for all sections has been completed (S240: Yes), the process proceeds to step S250. If the processing of all sections has not been completed (S240: No), the process returns to step S220 and repeats the processing.

In step S<b>250 , the generation unit 521 associates the section number with the generated reaction category (category ID) and stores them in the user reaction DB 532 . For example, if the number of sections of a certain content is "50", 50 reaction categories are generated by the processing up to S240 and arranged in the order of sections (time order) (arranged in chronological order). (A1) The data generated by the first category data generation method is also called category time-series data.

An example of the data configuration of the user reaction DB 532 is shown in FIG. The user ID, content ID, playback start date/time, section number, and reaction category are associated and stored. The user ID, content ID, and reproduction start date and time are the same as in the biometric information DB 531 of FIG. In the user reaction DB 532, similarly to the biometric information DB 531, a combination of the three fields of the user ID, the content ID, and the reproduction start date and time serves as the primary key of the table. As described above, the section number is a number that identifies each section when one content is divided into a plurality of sections (sections on the time axis). The smaller the section number, the shorter the elapsed time from the beginning of the content. A reaction category is the category ID described above. That is, based on the biometric information, the user's reaction during content reproduction is associated with one of predetermined categories (types). As shown in the example of FIG. 11, when a certain user (User1) reproduces a certain content (C1) twice, two pieces of data are recorded. Even for the same user, the reaction of the user often differs each time the content is reproduced, so generally the category data will differ to some extent each time the content is reproduced.

Here, (A1) In the first category data generation method, a plurality of sections are created for one content, and category data is generated for each section. Therefore, it is possible to accurately incorporate (express) the user's sensibility and taste into the data. For example, even if the user's reaction changes significantly in the middle of the content, that information can be faithfully reflected in the data.

(A2: Second category data generation method)
In the second category data generation method, one section (representative section) is created for one content, and category data is generated for that one section. The processing of the second method is shown in the flow chart of FIG.

In step S400, the generation unit 521 creates one representative section for one content by one of the following methods.
(A21) Method of using the entire content as a representative section The first representative section creation method is a method of using the entire content as a representative section. In other words, if the content is 3 minutes long, the length of the representative section will be 3 minutes. This method is suitable when it is assumed that a substantially constant user response will persist throughout the content.
(A22) A method of setting a section where the user's reaction is most prominent as a representative section A second representative section creation method is a method of identifying a portion of the content where the user's reaction is most prominent (strong) and using it as a representative section. is. In this case, the generation unit 521 calculates, for each section of the content, an index that integrates a plurality of pieces of biometric information within the same section, selects a representative section from the plurality of sections in the content based on the index, and selects the representative section. Category data is generated based on the biometric information in the section. Specifically, the generation unit 521 first sets the length of the representative section. For example, the length of the representative section may be the length of time during which the user's body movement is assumed to continue during content reproduction. For example, assuming that the average tempo of a song is 120, and assuming that the user reacts in almost the same pattern during a period of about 8 to 16 beats, the length of the representative section is set to about 4 to 8 seconds. do it. An example in which the length of the representative section (representative section length) is eight seconds will be described below.

Next, the generating unit 521 divides the content into sections having representative section lengths, and calculates an index R indicating the magnitude of the user's reaction for each section. As a method of dividing the content into sections, it is sufficient to set 0 to 8 seconds from the beginning of the content as section 1, 8 to 16 seconds as section 2, and 16 to 24 seconds as section 3. Alternatively, the intervals may be overlapped so that 0 to 8 seconds is interval 1, 4 to 12 seconds is interval 2, and 8 to 16 seconds is interval 3. The index R is calculated according to Equation (1) or Equation (2) based on amplitudes Ax, Ay, and Az in the three axial directions. Ax, Ay, and Az may be the maximum value of the amplitude in that section or the average value of the amplitude. The max function in Equation (2) is a function that returns the maximum value of its arguments.

R=Ax+Ay+Az Formula (1)

R=max(Ax, Ay, Az) Formula (2)

The generation unit 521 then selects the section with the largest index R as the representative section. When a certain user reproduces certain content multiple times, the representative section may change for each reproduction. Also, for the same content, the representative section may be different for each user. (A22) The second representative segment creation method is suitable for cases where the user reacts to a portion of the content for a short period of time. In general, users do not always react to the entire content, but rather to a portion of the content (naturally). The second representative segment creation method can accurately reflect such a user's natural reactions in the category data. In addition, this method is based on the concept of emphasizing how the user reacts, even if it is for a short time, without emphasizing in which section (position) of the content the user reacts.

In step S410, the generation unit 521 generates a reaction category for the representative section in the same manner as in step S230. Specifically, when (A21) the first representative section creation method is used in step S400, (A11) the rule-based method may be used. In this case, the maximum amplitude or average amplitude in the representative section (whole content) is calculated for each axis, and these are set as (Ax, Ay, Az) and compared with the rule shown in FIG. Also, the most dominant frequency component in the representative section should be (Fx, Fy, Fz). Also, the phase difference between the axes when the biological information is approximated by the sine wave of the frequency component may be calculated and used. Alternatively, in step S400 (A21) when using the first representative interval creation method, aggregate (x output, y output, z output) such as (Ax, Ay, Az), (Fx, Fy, Fz) If a classifier is trained using the obtained values, (A12) a method using machine learning can also be applied. If (A22) the second representative interval creation method is used in step S400, either (A11) the rule-based method or (A12) the method using machine learning may be used.

In step S<b>420 , the generation unit 521 stores the reaction category generated in step S<b>410 in the user reaction DB 532 . The user reaction DB 532 is stored in the same format as in FIG. 11, but in this case, the number of sections of all content is "1". (A2) In the second category data generation method, since the number of sections of content is always "1", the amount of processing can be reduced and the storage capacity can be saved. The above is the description of the second category data generation method.

Returning to the description of the flowchart in FIG. In step S120, the generator 521 creates a user reaction vector for each user based on the user reaction DB 532, that is, category data. An example of a user reaction vector is shown in FIG. Assuming that the total number of contents stored in the user reaction DB 532 is N, data of contents 1 to N are arranged in order in the user reaction vector. The data of each content is further composed of reaction categories for each section.

FIG. 13 shows an example (category time-series data) in which category data is generated by the (A1) first method. Content 1 is composed of “M1” sections, and reaction categories in sections 1 to M1 are recorded. However, a value such as "-1" or "NULL" indicating "invalid" is recorded for a section in which the user has not reproduced or a section in which reaction data could not be obtained. (A1) When generated by the first method, the number of sections is generally different for each content. For example, the number of sections for content 2 is "M2" and the number of sections for content 3 is "M3". ing. (A2) When generated by the second method, the number of sections is all "1", so "M1=1", "M2=1", and "M3=1".

Let Q be the total number of sections for each of the contents 1 to N. In the above example, Q=M1+M2+M3...+MN. The user reaction vector becomes a Q-dimensional vector. That is, a user's reaction vector is expressed as V[j] (j=1 to Q). The generator 521 creates such a user reaction vector for each user stored in the user reaction DB 532 . Let P be the number of users stored in the user reaction DB 532 below. Since the generation unit 521 generates user reaction vectors for P persons, they are collectively referred to as a user reaction matrix A[i][j] (i=1 to P, j=1 to Q).

As described above, when a certain user reproduces the same content multiple times, a plurality of category data are generated for the number of times the content has been reproduced. In such a case, create a user reaction vector by one of the following methods.

(B1) Method of Using Data When the User Reacts Most Strongly The first method is a method of selecting one piece of data when the user reacts the strongest and making it a user reaction vector. (A2) When the second category data generation method is used (the number of sections is one), data with the largest index R according to formula (1) or formula (2) may be adopted. Also, in the case shown by "strong horizontal glue" in the first playback, "strong vertical glue" in the second playback, "strong rotary glue" in the third playback, and "strong glue" in the fourth playback, that is, When reaction categories of different types are mixed, the reaction categories of different types are prioritized according to a predetermined rule, and the reaction category with the highest priority is adopted as the user reaction vector. For example, in the rule data 535 shown in FIG. 10, the degree of reaction is ranked as "strong>medium>weak". Here, ">" indicates that the rank of the reaction category on the left side of the symbol is higher than the rank of the reaction category on the right side. For example, "horizontal glue strong" has a higher priority than "horizontal glue medium" and "vertical glue medium". Further, reaction categories having the same degree of reaction are ranked according to "strong glue>rotation glue>vertical glue>horizontal glue". As a result, the category IDs of the 12 types of reaction categories shown in FIG. >1>11>0>NULL”. In other words, from the data of a plurality of times of reproduction, the data at the time of reproduction indicating the reaction category with the highest priority is adopted as the user reaction vector. (A1) When the first category data generation method is used (when there are multiple sections), data containing more reaction categories with higher priority is adopted as the user reaction vector. For example, the higher the priority above, the higher the value (point) is set, the points corresponding to the reaction category of each section of the content are totaled, and the data with the highest total point value can be adopted as the user reaction vector. .

(B2) Method of Selecting Data for Each Section The second method is a method of selecting data for each section from data of a plurality of reproduction times. (A1) When the first category data generation method is used, for example, when "user A" reproduces "content A" for the first time, "no response: 0" is indicated in section 1, and the second reproduction is performed. If "weak horizontal glue: 1" is indicated in section 1 when the user reaction vector is "weak horizontal glue: 1", refer to the above-mentioned priority of the reaction category, and "weak horizontal glue: 1" with higher priority is displayed as the reaction of section 1 in the user reaction vector Category. The same processing is performed for section 2 and subsequent sections. That is, for each section, the reaction category with the highest priority is selected from among the user's multiple reaction categories, and that category is adopted as the user reaction vector. (A2) When the second category data generation method is used, the same result as the method of (B1) using the data when the user reacts most strongly is obtained.

ここでＦは、２つのカテゴリ（反応カテゴリ）間の類似度を算出する所定の関数である。関数Ｆは、２つの反応カテゴリを引数（入力）とし、その２つの反応カテゴリの類似度を出力する関数である。例えば、少なくとも一方の入力（引数）が無効を示す「－１」である場合は「０」を返す。また、２つの入力が共に有効である場合、あらかじめ定義された反応カテゴリ間の類似度を返す。数式（４）のＫは、２つのユーザ反応ベクトルの両方が有効な区間の数である。つまり、２人のユーザが共通に視聴し、反応カテゴリが生成された区間の数である。 Returning to the description of FIG. In step S130, the calculation unit 522 calculates the degree of similarity between users (the degree of similarity indicating similarity between users) based on the user reaction vector. For example, according to Equation (3) or Equation (4), the similarity S[u][v] (u=1 to P, v=1 to P, u≠v of any two users (u, v) ) is calculated.

Here, F is a predetermined function that calculates the degree of similarity between two categories (response categories). Function F is a function that takes two reaction categories as arguments (inputs) and outputs the degree of similarity between the two reaction categories. For example, if at least one input (argument) is "-1" indicating invalidity, "0" is returned. It also returns the similarity between predefined response categories if both inputs are valid. K in Equation (4) is the number of intervals in which both of the two user reaction vectors are valid. In other words, it is the number of sections in which two users viewed and viewed in common and reaction categories were generated.

An example of the function F is shown in FIG. Since the function F indicates the relationship between reaction categories, the function F is sometimes called "category related data" or "category related data". The category-related data 534 of the storage unit 530 stores the characteristics of the function F, that is, the information (definition) of the relationship between categories in a table format as shown in FIG. Here, the higher the relevance (similarity) between the two reaction categories, the higher the numerical value of the similarity. For example, if two users show reactions in the same reaction category in a certain section of certain content, the degree of similarity in that section is "1". Further, when the major classification (type) such as "horizontal lamination", "vertical lamination", and "rotational lamination" is the same, and the strength of reaction differs within the same type, the degree of similarity is relatively high. Also, if two users show significantly different reactions, the similarity will be a small value, and may even be a negative value. Also, if the types are different and the strength of the reaction is different, the degree of similarity will be a large negative value. For example, when viewing the same content, if one user reacts with "weak horizontal movement" and another user reacts with "strong vertical movement," it can be said that the sensibilities and senses of these two users are significantly different. Therefore, the degree of similarity between "weak horizontal glue" and "strong vertical glue" is "-0.9", which indicates that they are not very similar. In this way, the similarity between users can be accurately calculated by classifying in detail not only the type of glue (reaction) that the user showed, but also the type of glue and the strength of the reaction. can do. In the example shown in FIG. 14, the similarity between "no reaction" and between "others" is set to "1", but these similarities may be set to values such as "0.5". That is, the value may be lower than usual. The reason for this is that the "no reaction" and "other" reaction categories contain less information (higher ambiguity) regarding the user's reaction than the other reaction categories. In other words, even if two users indicate the same response category of "no reaction" or "others", the similarity between the two users is not necessarily high.

Also, the characteristics of the function F may be freely set by individual users. For example, information on the standard function F is transmitted from the recommendation device 5 to the terminal device 11 . Then, the characteristics of the function F, that is, the table shown in FIG. Note that the display is not limited to the tabular form, and may be displayed using a three-dimensional graph or the like. The user u1 of the terminal device 11 operates the operation unit 160 to modify the characteristics of the function F as appropriate, and stores them in the storage unit 140 of the terminal device 11. FIG. The terminal device 11 reads the information of the function F specified by the user u1 from the storage unit 140 at an appropriate timing, associates it with the user ID, and transmits it to the recommendation device 5 . The recommendation device 5 stores the user ID and the function F in the category related data 534 of the storage unit 530 in association with each other. That is, the recommendation device 5 updates the category-related data 534 . Then, when creating recommendation information for the target user, a function F specified by the target user is used to calculate the degree of similarity between users (the degree of similarity between the target user and other users). When such processing is performed, the user-to-user similarity can be calculated by more accurately reflecting the user's (target user's) sensibility and taste, so that the accuracy of the recommendation information can be increased.

An example of the function F specified by the user is shown in FIG. 15 differs from the standard function F shown in FIG. This user has a strong preference for "rotation glue", and sets the degree of similarity between "rotation glues" higher than the standard (to a large value). For example, he makes such a setting because he thinks that ``I often react with a spinning glide, and is very compatible with other users who show a spinning glide.'' For this reason, when this user and another user show a turning knot in a certain section, the similarity of that section becomes very high. As shown in FIG. 15, the degree of similarity between reaction categories may be set to a value greater than "1".
As described above, in this embodiment, the display unit 150 of the terminal device 11 displays the information of the category-related data (function F), and the target user can browse it. Therefore, the target user can more easily understand why the recommendation device 5 has determined the similar users and why the recommendation information has been created. Therefore, it is possible to increase the possibility that the target user will be satisfied with the recommendation information and accept it. Also, the target user can customize the category-related data to their liking. Since the recommendation device 5 creates recommendation information based on the customized category-related data, it is possible to further increase the possibility that the target user will accept the recommendation information. In other words, it is possible to increase the likelihood that the target user will be satisfied with the recommendation information and will reproduce or purchase the recommended content accordingly.

Returning to the description of FIG. In step S140, the creating unit 523 selects (identifies) a user who satisfies a predetermined condition from among the users (target user u1) to whom the recommendation information is provided. In the first embodiment, a user with a high degree of similarity S is selected (specified) as a user who satisfies a predetermined condition. A user having a relatively high degree of similarity selected here is called a "similar user", and a group of similar users is called a similar user group or a similar group. Specifically, the creating unit 523 selects a user v whose S[u1][v] (v=1 to P, v≠u1) is equal to or greater than a predetermined threshold value α, and selects a user v similar to the target user u1. do. Alternatively, a predetermined number (for example, 100) of users may be selected in descending order of S[u1][v] and used as similar users. In addition, you may select one user with the highest degree of similarity as a similar user. In other words, there may be one or more similar users. A similar user can be said to be a user who watches a lot of content that is common to the target user and who tends to show a similar response to the same section of the same content. Therefore, the similar users are highly likely to be users whose sensibilities and tastes are similar to those of the target user.

In step S150, the creation unit 523 creates recommendation information based on similar users. The main recommendation information is information related to recommended content. Specifically, first, the creation unit 523 extracts recommendation candidate content. For example, the creation unit 523 extracts content that the target user has not yet viewed from among the content that the similar user has already viewed as the recommended candidate content. For example, if there are 100 similar users, content viewed by at least one of the 100 users and not viewed by the target user is extracted. Basically, recommended candidate content is limited to content that has not been viewed by the target user, but it is not necessarily limited. That is, the content viewed by the target user may be included in the recommended content. In addition, content that has been viewed by the target user a predetermined number of times or less may be included in the recommendation candidate content. In addition, the recommended candidate content may be limited to only content whose existence the target user is not aware of. In a system (service) in which a user evaluates content by “like” or “dislike”, etc., content highly rated by similar users (e.g. “very like” or “like”) is recommended. It may be extracted as content. Next, the creation unit 523 selects recommended content from the recommended content. One of the following processes is performed according to what the target user specified as the content condition.

(C1) If the target user does not specify anything as a content condition, the recommended content is set as the recommended content. Further, when the number of recommended candidate contents is larger than a predetermined number, the contents to which the similar users reacted more strongly are selected from among them. For example, in the user reaction vector of similar users, the number of sections corresponding to a predetermined reaction category is counted for each content, and a predetermined number of contents are selected in descending order of the count. For example, of the reaction categories shown in FIG. 10, reaction categories corresponding to "strong reactions" such as "strong horizontal glue", "strong vertical glue", "strong rotary glue", "strong glue", etc. Use The content selected here becomes the recommended content. It can be said that there is a high possibility that the target user will also strongly respond to content that has been strongly responded to by a user whose sensibility and preferences are similar to those of the target user. In other words, it can be said that the content is highly likely to be liked by the target user.

Further, when the number of recommended candidate contents is larger than a predetermined number, the creation unit 523 may further select, as recommended contents, contents that match the reaction category indicated by the target user in the past. Specifically, the creating unit 523 first aggregates the division function or the number of contents for each reaction category with respect to the user reaction vector of the target user. Then, the creation unit 523 identifies a predetermined number (for example, three) of reaction categories in descending order of the aggregated number. That is, the creating unit 523 identifies reaction categories that the target user has indicated relatively often in the past. Next, the creating unit 523 refers to the user reaction DB 532 or the user reaction matrix, and selects, as recommended content, content having a section that matches the identified reaction category from among the recommended candidate content. Here, the processing may be performed on the data whose user ID matches the similar user stored in the user reaction DB 532, or the processing may be performed on the data of all users other than the target without limiting to similar users. you can go When there are many pieces of content that match the conditions, the creation unit 523 selects a predetermined number of pieces of content in descending order of the number of sections that match the specified reaction category.

(C2) When the target user specifies a reaction category condition as the content condition, the content that matches the content condition is selected (selected) from among the recommended candidate contents, and made the recommended content. A target user specifies, for example, a reaction category condition as a content condition.
The creation unit 523 refers to the user reaction DB 532 or the user reaction matrix, and calculates the number of sections that match the designated reaction category (designated category) for the recommended candidate content. Then, a predetermined number of pieces of content are selected in descending order of the number of matching sections, and are used as recommended content. For example, when the target user desires content with many reactions of "strong vertical glue" or "medium vertical glue" as recommended content, he/she designates a condition such as "strong vertical glue OR medium glue" as the content condition. The creation unit 523 targets the category data of the recommendation candidate contents and calculates the number of sections with "strong vertical glue" or "medium vertical glue" for each content. Then, a predetermined number (for example, 30) of contents are selected in descending order of the calculated number of sections.

The creating unit 523 associates the user ID of the target user with the content ID of the recommended content and stores them in the storage unit 530 . This data is used to determine whether or not it is necessary to create recommended information when the user requests recommended information at another timing. Note that the date and time when the recommendation information is created may be stored. Then, when the time difference between the date and time when the recommendation request is received and the date and time when the recommendation information is created is equal to or greater than a predetermined value, the recommendation information may be recreated (updated).

In addition, the creating unit 523 may include the following information in the recommendation information in addition to the selected recommended content. Therefore, when selecting recommended content, the creating unit 523 identifies a location (segment) to which the user strongly reacted, and stores the information (segment number, etc.) in the storage unit 530 . The target users here may be limited to similar users, or may include users other than similar users (for example, all users may be included). For example, from all the sections of the recommended content, a predetermined number (for example, one) of sections are specified in descending order of reaction from similar users or all users, and the section numbers are stored in the storage unit 530 . Alternatively, it is possible to limit the reaction categories to predetermined reaction categories (for example, "strong horizontal glue", "medium horizontal glue", "strong horizontal glue", and "medium rotary glue"), and determine the number of users showing the predetermined reaction category or the number of reactions. A predetermined number of sections may be identified in descending order. This specified section is used for digest reproduction of the content in the terminal device. Note that the segment to be specified may be changed according to the reaction category of the target user. For example, if the target user has a relatively high frequency of showing a reaction of "strong horizontal glue" compared to other users, it is possible to specify sections in the recommended content in descending order of the user's reaction of "strong horizontal glue". good. The section identified here may be called a "recommended section". That is, the creation unit 523 refers to the user reaction DB 532 or the user reaction matrix, calculates the number of users who indicated a predetermined reaction category or the number of reactions for each section of the content, and selects a recommended section according to the number. It is also possible to specify and create recommendation information including recommended sections. Furthermore, by using this recommended section, recommendation information may be created for displaying a reception column (operating means such as a GUI component) for receiving a reproduction instruction for reproducing the recommended content from the recommended section on the display unit 150 of the terminal device 11. good.

In addition, the creation unit 523 refers to the user reaction DB 532 or the user reaction matrix, and calculates the number of users who have responded in a predetermined reaction category for each recommended content. For example, with respect to content A, the number of users who indicate any of "strong horizontal glue", "strong vertical glue", "strong rotary glue", and "strong glue" among similar users or all users is calculated. calculate. That is, the number of users who showed a strong reaction to the content A is calculated. Alternatively, if the frequency of the target users showing a reaction of "weak horizontal movement" is high, the number of users who showed a reaction of "weak horizontal movement" to the content A may be calculated. In other words, the number of users who have responded in the same way as the target user's reaction category with relatively high frequency may be calculated.

The creation unit 523 also calculates (aggregates) a main reaction category for each recommended content. For example, for content A, the number of responses and the number of users are counted for each response category, and a predetermined number (for example, the top three) of response categories are selected in descending order of the number of totals, and set as the main response category (main response category). . The main response category can be said to be a response category in which more users have responded to the recommended content. Further, a predetermined number of reaction categories may be selected in descending order of the number of totals from response categories with a predetermined number of responses or a predetermined number or more of users, and may be used as main response categories. In this case, the main response category may be "not applicable" for content with a small number of responses or a small number of users. Further, the number of users who indicated the main reaction category may be calculated as the number of users who indicated the predetermined reaction category. For example, when the main reaction category is "weak horizontal glue", the number of users showing "weak horizontal glue" is calculated.

In this way, the creating unit 523 creates recommended content information, recommended section information, information on the number of users who indicated a predetermined reaction category, and information on a main reaction category as recommendation information. However, the creation of information other than the recommended content information may be omitted.

In step S160, the processing unit 520 controls the communication unit 510 to transmit (output) the recommendation information created in step S150 to the terminal device 11 used by the target user u1. For example, the processing unit 520 transmits information (content metadata) such as the content name, artist name, genre, production year, viewing link destination, etc., as recommended content information included in the recommendation information to the terminal device. At this time, the recommendation device 5 may acquire metadata from the content distribution server 4 using the content ID as a key and transmit it to the terminal device 11 . Alternatively, metadata may be stored in the storage unit 530 of the recommendation device 5 . In addition, the recommendation device 5 notifies the content distribution server 4 of the content ID of the recommended content and the user ID of the user u1 who uses the target terminal device 11, and the content distribution server 4 that has received this notifies the content distribution server 4 of the target terminal device. Metadata of the recommended content may be transmitted to the device 11 .

In addition, the processing unit 520 transmits to the terminal device 11 information about the recommended section of the recommended content identified in step S150, information about the number of users who indicated a predetermined reaction category, information about the main reaction category, and the like. After executing step S160, the recommendation device 5 ends the recommendation information creation process.

Upon receiving the recommendation information from the recommendation device 5 , the terminal device 11 displays it on the display unit 150 . FIG. 16 shows an example of a screen 61 displaying recommended information. This screen 61 is a display example of recommendation information when the target user u1 does not specify (C1) content conditions. As shown in FIG. 16, recommended content information 611 such as the content name, artist name, genre, production year, etc. of the recommended content is displayed for the target user u1 (“user A” in the display example). In addition, the number of users 612 who indicated the given reaction category described above and the dominant reaction category 613 are displayed. The target user u1 can see such information and estimate the properties (features) of the recommended content. The target user u1 can infer the properties of the content to some extent from information such as the genre and year of production, but this is rough information and it is difficult to estimate the detailed properties of the content. For example, if the content is a song, even if the information that the genre of the song is "rock" is obtained, the "rock" genre includes songs with a wide variety of musical characteristics. is difficult to predict. In addition, information such as genre is given by the provider (operator) of the content, and information already given to the content is often reused, and in many cases it does not represent the actual state of the content. . On the other hand, the information on the number of users 612 indicating a predetermined reaction category and the information on the main reaction category 613 are information compactly summarizing (consolidating) real reactions of users (consumers) who actually viewed the content. Therefore, it can be said that the actual state of the content from the user's point of view is more accurately represented. Therefore, the target user u1 can grasp (determine) the properties of the recommended content in more detail and easily.

The screen shown in FIG. 16 further displays two types of buttons for reproducing content. A “play from beginning” button 615 is a normal playback button, and the content is played back from the beginning. On the other hand, the “playback from recommended section” button 614 is a button for playing back the content from the above-described recommended section. When the target user u1 presses the "reproduce from recommended section" button 614, the information of the recommended section is transmitted from the terminal device 11 to the content distribution server 4, and the content is streamed and reproduced from that section. Here, the button 614 is an example of a field for receiving a reproduction instruction (operation means such as a GUI component) for reproducing from the recommended section. The section is a section where the feature of the content is well expressed, and is a section that is highly likely to be liked by the target user u1. By reproducing such a part, the target user u1 can efficiently determine in a short time whether the content is good or bad, that is, whether the content suits his taste.

FIG. 17 shows an example of the recommendation information display screen 62 when the target user u1 specifies (C2) the content condition. At the top of the screen 62, a text field 621, which is a GUI (Graphical User Interface) for the target user u1 ("user A" in the display example of FIG. 17) to specify desired content conditions, and a content condition transmission A button 622 for (update) is displayed. The example of FIG. 17 shows the state after user A designates a content condition related to the reaction category of “strong horizontal glue OR moderate horizontal glue OR spinning” and presses the send button 622 . When the target user u1 presses the user condition transmission button 622, a recommendation request is transmitted from the terminal device 11 to the recommendation device 5, and recommendation information is transmitted from the recommendation device 5 to the terminal device 11 as a response. In this case, the recommended content is limited to content that matches the content condition. In the example shown in FIG. 17, the content is limited to the main reaction categories including at least one of "strong horizontal movement", "moderate horizontal movement", and "during turning movement". Other elements on the screen are the same as those described in FIG.

According to this embodiment, instead of directly comparing biometric information (waveform data) between users to identify similar users, category data (reaction categories) is generated based on biometric information, and category data and category data are generated. Since the degree of similarity between users is calculated based on related data, similar users can be specified with high accuracy. In the prior art, for example, even if two users show substantially the same reaction, if the shape and size of the waveform data are slightly different, the similarity between the users is judged to be lower than it actually is. there was a case. For this reason, there were cases where similar users could not be specified with high accuracy. In other words, in the conventional technology, there are cases in which fine differences in waveform data are judged to be unnecessarily large differences even though they have little to do with the user's sensibilities and tastes with respect to content. On the other hand, in the present embodiment, since the waveform data corresponds to one of the predetermined reaction categories, even if the shape and size of the waveform data of two people are slightly different, they are less likely to be affected. Therefore, similar users can be specified with high accuracy. Therefore, recommendation information can be created with high accuracy. Moreover, according to the present embodiment, various reaction categories are generated based on multiple types of biometric information, so that the user's various reactions to the content can be accurately reflected in the category data. Therefore, similar users can be specified with high accuracy, and recommendation information can be created with high accuracy. In addition, according to the present embodiment, since a reaction category is generated for each section into which the content is divided along the time axis, the degree of similarity between users can be accurately calculated by reflecting where and how the user reacted to the content. well calculated. Therefore, similar users can be specified with high accuracy, and recommendation information can be created with high accuracy. In particular, in the present embodiment, the length of the section is set to a length of time that is assumed to be similar to the user's reaction and is longer than the sampling period of the waveform data. For this reason, even if two users show substantially the same reaction, but the reaction timing is slightly different, the similarity between the users can be accurately calculated without being affected by the difference.

Further, according to the present embodiment, the amount of data of category data can be reduced compared to the amount of data of original biometric information, so the degree of similarity between users can be efficiently calculated with a small amount of calculation. In particular, when there are a large number of users for whom similarity calculation is to be performed, the calculation can be performed very efficiently. Therefore, it is possible to accurately and efficiently recommend content that matches the user's sensibility and taste. Further, according to the present embodiment, the information of the category-related data can be presented to the user, so that the user can easily understand the reason why the recommended information and the information on similar users are created. This makes it possible to increase the possibility that the user will accept the recommended information. Further, according to this embodiment, since the user can change (adjust) the category-related data, the user can easily reflect his or her sensitivity and taste in the similarity calculation between users.

Moreover, according to this embodiment, since reaction categories are generated based on biometric information, the user can easily designate content conditions using reaction categories. For example, although it is difficult for the user to specify the characteristics of biological information (waveform data), since reaction categories are categorized (typified), character strings (symbols) can be used to facilitate the user's reaction to the content. can be specified. In particular, since a moderate number (moderate type) of response categories are generated, the user can easily specify content conditions. Therefore, the user can designate desired content from a point of view that has been impossible conventionally. Further, according to the present embodiment, since it is possible to specify a section (recommended section) of content in which another user has given a characteristic reaction, and cause the user to reproduce that section, the user can recognize the feature of the content for a short period of time. can be easily grasped. Further, according to the present embodiment, since the recommendation information includes information about the number of users who have given a predetermined reaction to the content and the main reaction categories indicated by many users, users are provided by the content provider. It is possible to easily comprehend the real features of the content based on not only the information (uniform information) but also the real reactions of other users. Moreover, according to the present embodiment, the user does not need to perform a special operation to obtain the recommended information, and can simply view the content normally (naturally), thereby reducing the burden on the user.

<Modification 1-1 of Embodiment 1>
In Example 1, a similar user corresponding to the target user is identified, and information to be provided to the target user is created based on the content viewed by the similar user. In Modified Example 1-1 of Example 1, a dissimilar user corresponding to the target user is specified, and information to be provided to the target user is created based on the content viewed by the dissimilar user. Here, a dissimilar user is a user whose degree of similarity to the target user is relatively low. In other words, the creation unit 523 in Modification 1-1 further identifies dissimilar users having a lower degree of similarity indicating similarity between users to the target user from among the plurality of users. Then, the creating unit 523 further creates the content that the dissimilar user has viewed many times as non-recommended information. After that, the creating unit 523 presents the recommended information and the non-recommended information together to the target user. Differences from the first embodiment will be described below.

The processing of the recommendation device 5 in this modified example 1-1 is partially different from the processing of the first embodiment shown in FIG. S100 to S130 are the same as in the first embodiment. In step S140A (not shown) corresponding to step S140, the creation unit 523 uses the condition "similarity is low" as the predetermined condition in addition to the condition "similarity is high". That is, in addition to selecting (identifying) similar users with high similarity, "dissimilar users" with low similarity are selected (identified). Specifically, the processing unit 520 selects a user v whose similarity S[u1][v] (v=1 to P, v≠u1) is less than a predetermined threshold value β with respect to the target user u1. and dissimilar users. The threshold value β is normally a value lower (smaller value) than the threshold value α described above, but the threshold value α and the threshold value β may be the same value. Alternatively, a predetermined number (for example, 50) of users may be selected in descending order of S[u1][v] to be dissimilar users. Note that one user with the lowest degree of similarity may be selected as the dissimilar user. In other words, there may be one or more dissimilar users.

In step S150A (not shown) corresponding to step S150, the creating unit 523 generates content (not shown) that is expected not to suit the taste of the target user u1 based on the dissimilar users, in addition to the recommendation information described in S150. (also called non-recommended content). For example, content that has not been viewed by the target user u1 and has been viewed by many dissimilar users is selected as non-recommended content. Then, recommended information to be provided to the target user u1 is created, including information about the non-recommended content.

If a certain user reacts (sounds) to the content viewed by the target user u1 very differently from that of the target user u1, that user is likely to be a dissimilar user. For example, if one user is "no reaction" and the other user is "strong vertical glue", it can be said that the sensibilities and preferences of these two users are significantly different. In such a case, the function As shown in the characteristic example of F, the similarity between the two persons in the section becomes a small value (negative value). In other words, the dissimilar user is highly likely to be a user who is significantly different from the target user in terms of sensibility and taste, and therefore the content viewed by the dissimilar user is highly likely not to be liked by the target user.

The reason why information on non-recommended content is provided as recommendation information in addition to information on recommended content is to inform the target user of the recommendation capability (information selection capability) of the recommendation device 5 (convince the target user). . The recommendation device 5 selects a very small number of contents (for example, 10) that are expected to match the sensibility and taste of the target user from a huge amount of contents (for example, 10 million types) and presents them as recommended contents. A target user generally recognizes only the presented recommended content and does not recognize (cannot feel) the vast amount of content that exists behind it. For this reason, it is difficult for the target user to evaluate the information selection ability of the recommendation device 5 and it may be difficult for the target user to realize the usefulness and value of the recommendation device 5 .

On the other hand, if both recommended content information and non-recommended content information are presented, the target user can easily evaluate the information selection capability of the recommendation device 5 by comparing the two types of information. For example, when the recommended content and the non-recommended content are greatly different from the target user's point of view, and most of the recommended content matches his/her taste and most of the non-recommended content does not match his/her taste, the target user may use the "recommending device It can be easily evaluated that the information selection ability of On the other hand, if it is difficult to distinguish between recommended content and non-recommended content, and the proportion of content that matches the user's taste is not significantly different between recommended content and non-recommended content, the target user may think that the recommendation device has excellent information selection ability. It can be easily evaluated as “not (inferior)”.

In step S160A (not shown) corresponding to step S160, the processing unit 520 controls the communication unit 510 to transmit (output) the recommendation information created in step S150A to the terminal device 11 used by the target user u1. . The terminal device 11 displays the received information on the display section 150 .

FIG. 18 shows an example of the display screen 63 of the recommended information in the modified example 1-1. As shown in FIG. 18, information (recommendation information 631) of content (recommended content) recommended to the target user is displayed in the upper portion of the screen 63 . This part is the same as the screen display shown in FIG. Information (non-recommended information 632) of content (non-recommended content) not recommended to the target user is displayed at the bottom of the screen 63 . The target user can easily evaluate the information selection capability of the recommendation device 5 by comparing the recommended content and the non-recommended content. Also, for non-recommended content, an operation button 633 is displayed for reproducing a recommended segment (a segment suitable for trial listening) based on reaction categories of other users. Here, a section in which many dissimilar users have responded is also referred to as a "recommended section" for the sake of convenience. As mentioned above, it is highly probable that the section where other users reacted a lot is the section where the nature (feature) of the content is well represented, so the target user can easily determine the nature of the content in a short time. can do.

In this modified example 1-1, the user can easily evaluate the information selection capability of the recommendation device 5 because not only the recommended content information but also the non-recommended content information is provided to the user. In particular, since recommended content and non-recommended content are compared to create recommended information to be displayed on the display unit of the terminal device, the user can easily grasp the information selection capability of the recommendation device 5 . If the user feels that the recommendation device 5 has a high information selection capability, the user will have a certain sense of trust in the recommendation device 5, and will easily accept the information presented by the recommendation device 5 thereafter. Therefore, the effect of promoting the use (viewing) of the content is obtained. The contents distribution service provider and the provider of the recommendation device 5 can expect an effect such as an increase in sales accompanying the promotion of use of the service.

<Modification 1-2 of Embodiment 1>
In the modified example 1-2 of the first embodiment, all contents are selected without being limited to the contents viewed or highly evaluated by the similar users, and the contents matching the contents condition specified by the target user are selected. Recommended content. In this modification, steps S130 and S140 are omitted after steps S100 to S120 are executed. That is, in this modified example, the degree of similarity between users is not calculated. That is, the calculator 522 of the recommendation device 5 can be omitted.

In S150B (not shown) corresponding to step S150, the creating unit 523 creates recommendation information about content (recommended content) to be recommended to the target user. In this modified example 1-2, the target user designates category conditions (reaction category conditions) as content conditions. The creation unit 523 performs the same processing as "(C2) When the target user specifies a reaction category condition as a content condition" for all contents instead of the contents viewed or highly evaluated by similar users. I do.

S160B (not shown) corresponding to step S160 is the same as step S160 of the first embodiment. That is, the processing unit 520 controls the communication unit 510 to transmit (output) the recommendation information created in step S150B to the terminal device 11 used by the target user u1. The terminal device 11 that has received the recommendation information displays a screen similar to that shown in FIG.

According to this modified example 1-2, it is possible to provide the target user with information on a wider range of content that matches the content conditions, not limited to the content that the similar user has viewed or highly rated. Further, in this modified example, as in the first embodiment, reaction categories are generated based on biometric information, so the user can easily specify content conditions using character strings (symbols).

<Example 2>
In Example 1, the degree of similarity between users is calculated, and recommendation information is created based thereon. In the second embodiment, the degree of similarity between contents is calculated, and recommendation information is created based thereon. Specifically, the calculation unit 522 calculates a degree of similarity indicating similarity between contents based on the category data generated by the generation unit 521 and the category-related data 534 . In addition, when the content specification is received from the target user, the creation unit 523 selects content having a higher degree of similarity indicating similarity between the specified contents as recommended content from among the plurality of contents. Differences from the first embodiment will be mainly described below.
The configuration of the recommendation device 5 in this embodiment is the same as that in the first embodiment shown in FIG.

When the user (target user) u1 performs an operation for displaying recommendation information, a recommendation request is transmitted from the terminal device 11 to the recommendation device 5 . The recommendation request of the second embodiment includes the content ID of the content (base content) that serves as a reference for recommendation. For example, when the terminal device 11 displays recommendation information such as "a user who viewed content A also viewed content B", content A is the base content. Also, the user ID of the user (target user) u1 who uses the terminal device 11 may be included in the recommendation request. In this case, it is possible to create recommendation information that is further specialized (personalized) for the target user. Also, as described in the first embodiment, the content condition may be included in the recommendation request. Upon receiving the recommendation request, the recommendation device 5 determines whether recommendation information for the base content has already been created by batch processing or the like. Specifically, the processing unit 520 determines whether or not the storage unit 530 has recommendation information (such as the content ID of the recommended content) associated with the content ID of the base content. If the base content recommendation information exists, the processing unit 520 reads it from the storage unit 530 and transmits it to the terminal device 11 via the communication unit 510 . If there is no recommendation information corresponding to the base content, the processing unit 520 executes the process shown in FIG. 19 to create recommendation information and transmit it to the terminal device 11. FIG. When newly creating recommendation information, the processing unit 520 stores it in the storage unit 530 in association with the content ID of the base content.

Step S300 is the same as step S100.
Step S310 is similar to step S110. However, in the second embodiment, when (A1) the first category data generation method is used, the number of sections (the number of sections) is the same for all contents. Specifically, when creating content sections in step S210, the number of sections Ma is made the same for all content. For example, when the number of sections Ma is set to "50", the length of one section is 4 seconds for content with a length of 200 seconds, and the length of one section is 8 seconds for content with a length of 400 seconds. Become. (A2) When the second category data generation method is used, the same as in the first embodiment, the number of sections Ma is "1".

In step S320, the generation unit 521 creates a content vector for each content based on the user reaction DB 532, that is, category data. An example of a content vector is shown in FIG. Assuming that the number of all users stored in the user reaction DB 532 is P, the data of users 1 to P are arranged in order in the content of one content. Each user's data is further composed of reaction categories for the section 1 to Ma. Instead of using data of all users stored in the user reaction DB 532, data of some users may be used to create a content vector. For example, a content vector may be generated targeting users who have viewed a predetermined number of times or more of content (ignoring users who have viewed less content). In that case, the number of users to be processed should be P.

In the example of FIG. 20, the user 1 portion of content 1 is composed of “Ma” sections, and reaction categories in sections 1 to Ma are recorded. However, a value such as "-1" or "NULL" indicating "invalid" is recorded for a section in which the user has not reproduced or a section in which reaction data could not be obtained. Similarly, the section for user 2 is composed of "Ma" sections, and the section for user 3 is the same. That is, the content vector of content 1 is composed of Q=P×Ma reaction categories. The content 2 and the other content are the same, and each consists of Q=P×Ma reaction categories.

Such a content vector is created for each content stored in the user reaction DB 532 . Assuming that the number of contents stored in the DB is N, the generation unit 521 creates N content vectors. 1 to Q).

ここで関数Ｆは、実施例１で説明した関数と同様である。数式（６）のＬは、コンテンツｕが有効であり、かつコンテンツｖが有効であるベクトル要素（区間）の数である。つまり、２つのコンテンツがともに有効である区間の数である。 In step S330, the calculation unit 522 calculates similarity between contents (similarity indicating similarity between contents) based on the content vector (content matrix B). For example, according to Equation (5) or Equation (6), the similarity S[u][v] (u=1 to N, v=1 to N, u≠v) of any two contents (u, v) Calculate

Here, the function F is the same as the function explained in the first embodiment. L in Expression (6) is the number of vector elements (sections) in which the content u is valid and the content v is valid. In other words, it is the number of sections in which both two contents are valid.

In step S340, the creation unit 523 selects (identifies) other content having a high degree of similarity S with respect to the content (base content Ca) serving as a reference for providing recommendation information. Contents having a relatively high degree of similarity identified here are referred to as similar contents, and a set of similar contents is referred to as a similar content set. Specifically, the creating unit 523 selects content v whose S[Ca][v] (v=1 to N, v≠Ca) is equal to or greater than a predetermined threshold α, and selects content v as similar content to the base content Ca. do. Alternatively, a predetermined number (for example, 30) of contents v may be selected in descending order of S[Ca][v] and made similar contents. In addition, you may select one content with the highest degree of similarity as similar content. In other words, there may be one or more similar contents. It can be said that similar content is content that many users who viewed the base content have viewed, and that the same users have shown similar reactions (nori). Therefore, from the user's point of view, similar content is highly likely to be content similar in nature and impression to the base content.

In step S350, the creation unit 523 creates recommendation information based on the similar content. The main recommendation information is information related to recommended content. The similar content may be used as the recommended content as it is. Alternatively, from among the similar contents, those that have been viewed by the user more often may be preferentially selected as recommended contents. Furthermore, when the user ID of the user u1 is included in the recommendation request, the content that the user u1 has not viewed may be preferentially selected from the similar content as the recommended content.

The creation unit 523 identifies a portion (section) of the recommended content that many users reacted to, and stores the information (section information) in the storage unit 530 . This specified section is used for digest reproduction of the content in the terminal device. The section specified here may be called a recommended section.

The creating unit 523 also calculates the number of users who have responded in a predetermined response category for each recommended content. For example, for content A, the number of users indicating any one of "strong horizontal glue", "strong vertical glue", "strong rotary glue", and "strong glue" is calculated. That is, the number of users who showed a strong reaction to the content A is calculated.

In addition, the creating unit 523 calculates (aggregates) a main reaction category for each recommended content. For example, for content A, the number of responses and the number of users are counted for each response category, and a predetermined number (for example, the top three) of response categories are selected in descending order of the number of totals, and set as the main response category (main response category). . Further, a predetermined number of reaction categories may be selected in descending order of the number of totals from response categories with a predetermined number of responses or a predetermined number or more of users, and may be used as main response categories. In this case, the main response category may be "not applicable" for content with a small number of responses or a small number of users. Further, the number of users who indicated the main reaction category may be calculated as the number of users who indicated the predetermined reaction category. For example, when the main reaction category is "weak horizontal glue", the number of users showing "weak horizontal glue" is calculated.

In this way, recommended content information, information on recommended sections, information on the number of users who indicated a predetermined reaction category, and information on main reaction categories are created as recommendation information. However, the creation of information other than the recommended content information may be omitted.

In step S360, the processing unit 520 controls the communication unit 510 to transmit (output) the recommendation information created in step S350 to the terminal device 11 that requests it.
Upon receiving the recommendation information from the recommendation device 5 , the terminal device displays it on the display unit 150 . FIG. 21 shows an example of a screen 64 displaying recommended information. FIG. 21 shows a case where a user using a terminal device displays information on content related to the content (related content) after viewing certain content (base content, "content A" in the example of this figure). is an example of A “reproduce from recommended section” button 641 is an example of a field for receiving a reproduction instruction for reproducing from a recommended section. Other information shown in FIG. 21 is the same as the information described in FIG.

In this embodiment, based on the category data, the degree of similarity between contents can be calculated with high accuracy, and based on this, recommendation information regarding other contents related to a specific content (base content) can be created with high accuracy. Therefore, it can be applied when the user wants to obtain information on other content related to specific content. Since the degree of similarity between contents can be accurately calculated based on the biometric information of the user when the contents are reproduced, the recommended contents can be selected with high accuracy. In addition, as in the first embodiment, information on the recommended section, information on the number of users who indicated a predetermined reaction category, information on the main reaction category, etc. are provided as recommendation information, so that the user can easily grasp the nature of the content. be able to. In addition, as in the first embodiment, the user can easily grasp the features of the content in a short time in order to provide the recommendation information for displaying the operation means (GUI component) for the user to reproduce the recommended section on the terminal device. can be done.

<Example 3>
In the examples so far, information about content is created as recommendation information. In Example 3, information about a user (other users) is created as recommendation information. That is, the creating unit 523 according to the third embodiment identifies similar users having relatively high similarity to the target user from among the plurality of users. Then, the creating unit 523 identifies a recommended user based on the similar users, creates information about the recommended user as recommended information, and presents the recommended information to the target user. Furthermore, when receiving a user condition related to a desired recommended user from a target user, the creating unit 523 preferably further considers the specified user condition to specify the recommended user. The target user specifies, as a user condition, a reaction category that a similar user has given to arbitrary content, or a condition related to a reaction category that a similar user has given to predetermined content (content specified by the target user). can be done. The creation unit 523 calculates an index that comprehensively indicates the degree of similarity between the target user and the other users and the degree to which the other users match the user conditions, and specifies the recommended users based on the index. good too. The creating unit 523 may include in the recommendation information information about the reaction category indicated by the recommending user with relatively high frequency. The creation unit 523 may include in the recommendation information information about the content and information about the section for which the recommending user has given a predetermined reaction (characteristic reaction). Furthermore, the creating unit 523 creates a reception field (operation means such as GUI components) for the target user to reproduce the content to which the recommending user has given a predetermined reaction or the section to which the recommending user has given a given reaction. Recommended information to be displayed on the display unit 150 of the device 11 may be created. Differences from the first embodiment will be mainly described below.

In the third embodiment, when a request (recommendation request) requesting recommendation information is transmitted from the terminal device 11 used by the target user u1 to the recommendation device 5, the recommendation device 5 corresponds to the flowchart of FIG. Execute the process. The recommendation request includes the user ID of the target user u1, and may further include user conditions desired by the target user u1 (also referred to as user conditions or user narrowing conditions). The recommendation device 5 executes steps S100 to S140 in the flowchart of FIG. 7 in the same manner as in the first embodiment.

In S150C (not shown) corresponding to step S150, the creating unit 523 creates recommendation information about users (recommended users) to be recommended to the target user u1. Here, the creation unit 523 creates recommendation information as follows, depending on whether or not a user condition is specified by the target user u1 and the specified content.
(D1) If the target user u1 does not specify anything as a user condition, the similar users selected in step S140 are set as recommended users for the target user u1. Alternatively, users with a high degree of similarity to the target user u1 may be further narrowed down from the similar users.
(D2) When the target user u1 specifies the number Ha of recommended users as a user condition, Ha users are selected as recommended users in descending order of similarity S[u1][v] with the target user u1.
(D3) If the target user u1 specifies a reaction category condition as a user condition, the similar users selected in step S140 are targeted, and users who match the specified user condition are selected as recommended users. Furthermore, the creating unit 523 may select recommended users as follows depending on whether content is specified in the user conditions.

(D3-A) When the target user u1 specifies a reaction category without specifying content, the creation unit 523 targets the user reaction vectors of similar users and matches the specified reaction category (specified category). Calculate the number of intervals. Then, a predetermined number of users are selected in descending order of the number of matching sections, and are designated as recommended users. For example, if the target user u1 wants to select a user who often responds to "strong vertical movement" or "medium vertical movement" as a recommended user, a condition such as "strong vertical movement OR moderate vertical movement" is specified as a user condition. The creation unit 523 calculates the number of sections with “strong vertical glue” or “moderate vertical glue” for each user, targeting the user reaction vectors of similar users. Then, a predetermined number (for example, 20) of users are selected in descending order of the calculated number of sections.

(D3-B) When the target user u1 specifies the content and the reaction vector, the creating unit 523 targets the user reaction vectors of the similar users to match the specified content and the specified reaction category (specified reaction category). category) is calculated. Then, a predetermined number of users are selected in descending order of the number of matching sections, and are designated as recommended users. For example, if the target user u1 wants to recommend a user who indicated "strong horizontal glue" in content A and "during spinning" in content B, the user condition is "content A = strong horizontal glue AND content A condition such as B=during rotation" is specified. The creation unit 523 calculates the number of sections that match user conditions specified for each user, targeting the user reaction vectors of similar users.

Moreover, the target user u1 may further designate a specific section of the content as a user condition. In this case, the recommendation device 5 notifies the terminal device 11 of information on each section of each content in advance, and then makes the target user u1 specify the content and the section. For example, the target user u1 indicates “strong horizontal glue” in all three sections 3 to 5 of content A, and indicates “during rolling” in any one of four sections 4 to 7 of content B. If the indicated user is desired to be a recommended user, a user condition such as "content A: ALL 3-5=strong lateral movement AND content B: ANY 4-7=medium rotation" is specified. The creation unit 523 calculates the number of sections that match user conditions specified for each user, targeting the user reaction vectors of similar users. Then, users with a large number of sections are preferentially selected.

Further, users may be selected by comprehensively considering the degree of similarity between users and the number of sections that match user conditions. For example, if only one section is specified in the user condition, such as "content A: 2 = strong horizontal glue" (user who indicated strong horizontal glue in section 2 of content A), this user condition is met. The number of sections is at most "1". If there are more than the required number of users (required number of recommended users) who match this user condition, the required number of users may be selected in descending order of similarity between users.

Also, Na[v], which is the number of sections or the number of contents that match the user condition of user v (v is a suffix indicating each similar user), and similarity S[u1][v] with target user u1. In addition, an index W (W[v]) may be calculated for each user according to Equation (7) or Equation (8), and users with a high (large) index W may be preferentially recommended users. For example, the index W may be calculated for all users (users other than the target user) or similar users, and a predetermined number of users may be selected in descending order of the index W to be recommended users. In Equation (7), η1 and η2 are positive predetermined values (weighting factors). Also, in Equation (8), λ1 and λ2 are positive predetermined numbers. According to formula (7) or formula (8), it is possible to calculate an index W that has a higher value as the number of sections (number of contents) that match the user condition increases, and a higher value as the degree of similarity between users increases. can.

In the third embodiment, user attribute information is stored in the storage unit 530 . Specifically, the storage unit 530 stores a user name (real name or nickname), age, gender, contact information (mail address, phone number, etc.), self-introduction message, etc. in association with the user ID. The creation unit 523 reads user attribute information from the storage unit 530 for the selected recommended user, and creates recommendation information using the user attribute information.

The creating unit 523 also refers to the user reaction vector of the recommended user, and aggregates the number of reactions (number of sections) for each reaction category for each user. Then, the information of the reaction category (main reaction category) frequently indicated by the user is included in the recommendation information. For example, information such as "user A often shows weak horizontal movement" and "user B often shows strong vertical movement and medium vertical movement" is included in the recommendation information.
The creating unit 523 also refers to the user reaction vector of the recommending user, and inserts information (main reaction content) about the content and the section of the content for which the user has given a characteristic reaction into the recommendation information. For example, "user A indicated strong horizontal glue in content A", "user B indicated medium vertical glue in section 10-11 of content B, and strong vertical glue in section 15-20 of content B". Such information is included in the recommendation information.

Step S160C (not shown) following step S150C is the same as step S160 of the first embodiment. That is, in step S160C, the processing unit 520 controls the communication unit 510 to transmit (output) the recommendation information created in step S150C to the terminal device 11 used by the target user u1.

As in the first embodiment, when receiving the recommendation information from the recommendation device 5 , the terminal device 11 displays it on the display unit 150 . FIG. 22 shows an example of a screen 65 displaying recommended information. At the top of the screen, a text field 651, which is a GUI for the target user (user A in the example of FIG. 22) to specify desired user conditions, and a button 652 for sending (updating) the user conditions are displayed. It is The example of FIG. 22 shows the state after the user A has specified the user condition of "strong horizontal glue OR medium horizontal glue OR rotating medium" and pressed the send button 652 . When the user presses the user condition transmission button 652 , the terminal device 11 transmits a recommendation request to the recommendation device 5 , and as a response thereto, recommendation information is transmitted from the recommendation device 5 to the terminal device 11 .

In FIG. 22, recommendation information for the target user ("user A" in the example of this figure) is displayed under the user condition. That is, the terminal device 11 displays the recommended information in association with the user condition specified by the user. Specifically, user attribute information 653 such as the recommended user's user name, age, gender, contact information, and self-introduction message is displayed. The user attribute information 653 displayed here is limited to information previously approved by the target user. "Private" is displayed for attribute information that the user does not want to disclose.

In addition, the key response category information 654 and key response content information 655 described above are displayed. In the example shown in FIG. 22, for example, the main reaction category of user 1 is "(1) strong horizontal glue (2) moderate horizontal glue", and user 1 most frequently indicates "strong horizontal glue", and the second is "strong horizontal glue". It indicates that "inside horizontal glue" is shown at a high frequency. Also, the main reaction content of user 1 is "(1) content 1 (2) content 2:3-9". This represents that "user 1 showed a reaction of the main reaction category in sections 3 to 9 of content 1 and content 2".

The target user can see such information and easily estimate the characteristics (sensitivity and taste) of the recommended user. The target user can roughly estimate the features of the recommended user from the user attribute information 653 such as age, gender, self-introduction message, etc., but these are not highly accurate information. For example, even users of the same age and gender have different sensitivities and tastes. Moreover, such information is information self-reported by the user, and may not be accurate. In other words, it is difficult to accurately know the user's sensibility and preferences from general user attribute information 653 . On the other hand, the main reaction category 654 and the main reaction content 655 are information created based on the user's natural reaction when viewing the content. Users often react to content unconsciously. Since the main reaction category 654 and the main reaction category 655 are information compactly summarizing (aggregating) real reactions of users (consumers) who actually viewed the content, they are actual characteristics of users (raw characteristics). It can be said that it expresses more accurately. As a result, the target user can easily grasp the characteristics (sensibility and taste) of the recommended user in more detail.

In the screen example shown in FIG. 22, a play button 656 for playing back main reaction content is also displayed for each recommended user. For example, when the play button 656 corresponding to user 1 is pressed, "content 1 and content 2", which are the main reaction contents of user 1, are played back in sequence. A fast-forward button for reproducing the next content and a fast-reverse button for reproducing the previous content may be displayed as appropriate. Further, when a specific section such as "Content 2: 3 to 9" is indicated, when the play button 656 is pressed, section 3 to section 9 of content 2 are reproduced, and then the next content is played. may be played. That is, a section (recommended section) in which another user reacts in a characteristic manner may be reproduced.

Also, in the screen example shown in FIG. 22, a button 657 for sending a message to a specific user is displayed. For example, when the target user presses the send button 657 corresponding to user 1, a text field or the like for entering a message is displayed, and a message for user 1 can be entered. The message is notified to User1. Also, if User 1 replies to the message, the target user is notified. Therefore, the target user can communicate with the recommending user even if the recommending user does not disclose the contact information. In other words, the user can easily communicate with other users who seem to have the same sensibilities and tastes without disclosing his/her own personal information. Message control between users (processing such as message transmission/reception and notification) may be performed by the recommendation device 5 or by another device.

In the above description, similar users with a relatively high degree of similarity to the target user are specified. Low dissimilar users may be identified. Then, in the same way as identifying recommended users based on similar users, non-recommended users may be identified based on dissimilar users. As with the screen 65 shown in FIG. 22, for non-recommended users, user attribute information, main reaction category information, main reaction content, a button for reproducing the main reaction content, a button for sending a message, and the like are displayed. It can be displayed on the screen. Also, similar to FIG. 18, the recommended user information and the non-recommended user information may be displayed on the screen in contrast. The target user can easily determine the information selection capability of the recommendation device 5 by, for example, comparing the main reaction content of the recommending user and the main reaction content of the non-recommending user.

According to this embodiment, since the degree of similarity between users is calculated based on category data and category-related data, it is possible to accurately and efficiently identify other users who match the sensitivity and taste of the user (target user). can be recommended to Further, according to this embodiment, since reaction categories are generated based on biometric information, the user can easily designate user conditions using the reaction categories. For example, although it is difficult for a user to specify characteristics of biological information (waveform data), since reaction categories are categorized (typed), character strings (symbols) can be used to specify user conditions for a desired user. can be easily specified. Further, according to this embodiment, the recommendation information can include the main reaction category and the main reaction category regarding the recommending user. Since this information is information that compactly summarizes (aggregates) the real reactions of users (consumers) who have actually viewed the content, compared to the user attribute information self-reported by the user, the user's actual It can be said that the feature (original feature) is represented more accurately. Therefore, the user (target user) can grasp the characteristics (sensitivity and preference) of the recommended user in more detail and easily.

<Modification 3-1 of Embodiment 3>
In the modified example 3-1 of the third embodiment, all users, not limited to similar users, are selected as recommended users by selecting users who match user conditions specified by the target user. In this modified example 3-1, steps S130 and S140 are omitted after steps S100 to S120 are executed. In other words, in Modification 3-1, the degree of similarity between users is not calculated. That is, the calculator 522 of the recommendation device 5 can be omitted.

In S150D (not shown) corresponding to step S150, the creating unit 523 creates recommendation information about users (recommended users) to be recommended to the target user. In this modified example 3-1, the target user designates (D3) a condition (category condition) related to the reaction category as the user condition. The creation unit 523 performs the same processing as (D3) of the third embodiment, targeting not similar users but all users (all users other than the target user).

S160D (not shown) corresponding to step S160 is the same as step S160C of the third embodiment. That is, the processing unit 520 controls the communication unit 510 to transmit (output) the recommendation information created in step S150D to the terminal device 11 used by the target user. The terminal device 11 that has received the recommendation information displays a screen similar to that shown in FIG. According to this modified example 3-1, it is possible to provide the target user with information on a wider range of users who match the user conditions, not limited to similar users. Further, according to this modification, as in the case of the third embodiment, reaction categories are generated based on biological information, so that the user can easily specify desired user conditions using the reaction categories.

<Example 4>
The content in Example 4 is music (songs), and in Example 4, category data is created with higher accuracy using properties unique to music. Differences from the first embodiment will be mainly described below. FIG. 23 shows the configuration of the recommendation device 5 of the fourth embodiment. In FIG. 23, a music information DB 537 (music information DB) is stored in the storage unit 530 as compared with the first embodiment.

An example of the data configuration of the music information DB 537 is shown in FIG. Music attribute information such as music name, artist name, genre, age, etc. is stored in association with the music ID. Furthermore, the start time of each measure (elapsed time from the beginning of the song) is recorded for each song. Also, the tempo (information indicating the speed of the music) and beat type are recorded for each measure. In the example shown in FIG. 24, the beat type is information indicating the strength of the beat of the music, such as "strong", "medium", and "weak". However, the information is not limited to the information indicating the strength of the beat, and may be information indicating a rhythm pattern such as "16 beats", "8 beats", "4 beats", "2 beats", and "triplets". . Note that beat information may be stored in the music information DB 537 instead of bar information shown in FIG. 24 . For example, the time for each beat (elapsed time from the beginning of the music) may be stored. The information stored in the music information DB 537 may be obtained by the recommendation device 5 from another device (music information database, etc.). Alternatively, the recommendation device 5 may analyze the sound signal of the music and create information such as the beat, tempo, measure, and beat type.

The recommendation device 5 executes recommendation information creation processing in a flow similar to that of the flow chart of FIG. However, it differs from the processing of the first embodiment in the following points. In step S110E (not shown) corresponding to step S110, the generation unit 521 generates category data based on the biometric information DB 531 and the music information DB 537. FIG. In this embodiment, the (A1) first category data generating method described in the first embodiment is used. That is, the generation unit 521 divides the content into a plurality of sections and generates a reaction category for each section.

In step S210E (not shown) corresponding to step S210, the generation unit 521 creates a section of the content to be processed. At this time, sections are generated using bar information or beat information stored in the music information DB 537 . When bar information is used, one or more bars are associated with one section. For example, measures 1 and 2 are defined as section 1, measures 3 and 4 are defined as section 2, and measures 5 and 6 are defined as section 3, and two measures are associated with one section. It is desirable that the number of bars corresponding to one section is about 1 to 4. This process is based on the knowledge that a normal song has a high probability of continuing similar beats and rhythm patterns for about 1 to 4 measures. When beat information is used, a plurality of beats are associated with one section. For example, beats 1 to 8 are section 1, beats 9 to 16 are section 2, beats 17 to 24 are section 3, and eight beats correspond to one section. It is desirable that the number of beats corresponding to one section is about 4 to 16. This is based on the knowledge that, in general, there are many songs in the quadruple time system, in which one measure is composed of four beats, and that there is a high possibility that similar beats and rhythm patterns will continue in measures 1 to 4. It is a process based on If the music information DB 537 contains beat information, the section may be created based on that information. For example, in the case of a piece of music in quadruple time, 4 beats, 8 beats, 16 beats, etc. are associated with one section. In the case of a piece of music in triple time, 3 beats, 6 beats, 9 beats, etc. are associated with one interval. In other words, the generation unit 521 preferably divides the content on the time axis by a predetermined multiple times the length of the beat or bar of the music to create a plurality of sections.

Also, the generation unit 521 may create a section based on tempo information. For example, when the music information DB 537 stores the tempo for each bar, a portion where the tempo changes more than a certain amount is assigned to a different section. For example, a bar whose tempo has changed by a predetermined value (for example, 10) or more compared to the tempo of the previous bar is assigned to a section different from that of the previous bar. Alternatively, a bar in which the tempo has changed by a predetermined ratio (eg, 5%) or more compared to the tempo of the previous bar may be set as a section different from the previous bar.

Alternatively, the section may be created based on the beat type. For example, the point where the beat type changes is taken as the boundary between two adjacent sections. In the example shown in FIG. 24, the beat type of bars 1 to 4 of the song ID "1" is "medium", and the beat type from bar 5 onwards is "strong". be a different segment.

After step S210E, the recommendation device 5 executes steps S220 to S250 in the same manner as in the first embodiment. Also, after step S110E, steps S120 to S160 are executed in the same manner as in the first embodiment. In this embodiment, since the sections are created based on the music information, it is possible to reduce the possibility that musically different elements are mixed in one section. In other words, it is possible to improve the possibility that a similar rhythm pattern or beat pattern will continue within one section. Therefore, it is possible to reduce the mixture of different types of user reactions (different types of body movements) within one section. Therefore, category data can be generated with higher accuracy, and recommendation information can be created with higher accuracy.

<Example 5>
In Example 5, similar to Example 4, the characteristic properties of music are used to create category data with higher accuracy. In Example 5, in addition to the process of Example 4, another process is performed. The configuration of the recommendation device 5 of the present embodiment is the configuration shown in FIG. 23, which is the same as that of the fourth embodiment. In other words, the storage section 530 stores the music information DB 537 . The generating unit 521 according to the fifth embodiment preferentially extracts frequency components having a predetermined relationship with frequency components based on the beat or speed of music among the frequency components analyzed from the biological information, and generates the extracted frequency components. Generate categorical data based on the components.

In step S110F (not shown) corresponding to step S110, the generation unit 521 generates category data based on the biometric information DB 531 and the music information DB 537. FIG. The flow of processing in step S110F is the same as in the flowchart of FIG. 8, and S200, S220, S240, and S250 are the same as in the first embodiment. Also, S210F corresponding to S210 is the same as S210E of the fourth embodiment.

In S230F corresponding to step S230, the generation unit 521 generates category data for the section to be processed using the beat or tempo information of the music. Specifically, the generation unit 521 uses a technique such as Fourier analysis to extract main frequency components of the biometric information data in the section to be processed. At this time, frequency components having a predetermined relationship with the tempo of the music are extracted, and other frequency components are not extracted. The user's biometric information data (the user's body movements) while listening to music is highly likely to be synchronized with the tempo of the music. However, the user does not always move his/her body every beat, and often moves his/her body every two or four beats. For example, in rock and pop music with a quadruple time signature, drums (such as a snare drum) are often played on the second and fourth beats, and users often move their bodies at the timing of the snare drum performance. . In this case, the biological information data contains many frequency components corresponding to two beats. In addition, drums (such as hi-hats) are often played with a period of 1/2 beat or 1/4 beat, and users often move their bodies (such as fingers and arms) at the timing of playing the hi-hat. In this case, the biological information data contains many frequency components corresponding to 1/2 beat and 1/4 beat. Therefore, a frequency having a predetermined relationship with the beat frequency is extracted from the biological information data. Here, the predetermined relationship is a relationship such as 1/4 times, 1/2 times, 1 time, 2 times, and 4 times. In the case of a piece of music that frequently uses triple time or triplets, it may be better to extract 1/3 or 3 times the frequency. That is, typically, frequencies that are a fraction of an integer and an integer multiple of the reference frequency (beat frequency) are extracted. However, in some cases, a frequency represented by a simple integer ratio, such as 2/3 times or 3/2 times the reference frequency, may be extracted. In other words, frequencies having a simple integer ratio relationship with the reference frequency may be preferentially extracted.

For example, if the tempo of a piece of music is "120" (120 beats per minute), the frequency of one beat corresponds to "2 Hz". The frequency corresponding to 2 beats is "1 Hz", the frequency corresponding to 1/2 beat is "4 Hz", and the frequency corresponding to 1/4 beat is "8 Hz". FIG. 25 shows an overview of the process of extracting frequency components related to beat frequencies. FIG. 25 is a diagram showing how frequency analysis is performed on the section to be processed and frequency components are extracted. Here, it is assumed that f0 is the reference frequency (beat frequency) and 1/2f0, f0, and 2f0 are extracted. In FIG. 25, there are five frequency peaks p1 to p5. Of these, the three peaks p1, p3 and p5 marked with "o" have a predetermined relationship with the reference frequency and are thus extracted. On the other hand, the two peaks p2 and p4 marked with "x" are not subject to extraction because they do not have a predetermined relationship with the reference frequency.

After extracting the frequency component to be extracted, the generation unit 521 generates category data by one of the following methods.
(E1) A method of synthesizing signal waveforms from frequency components The first method is a method of synthesizing signal waveforms (time domain data) based on extracted frequency components using a technique such as inverse Fourier transform. . If the biological information data is displacement data in three directions, the signal waveforms of x, y and z are synthesized. Then, using the synthesized signal waveforms, the same processing as in the first embodiment is performed to generate category data.
(E2) Method of Generating Category Data Directly from Frequency Components The second method is a method of directly generating category data from frequency components. Specifically, in (A11), a rule using the frequency component (intensity of the frequency component) to be extracted is defined. For example, let Dx be the frequency component of f0 on the x-axis, Dy be the frequency component of f0 on the y-axis, and Dz be the frequency component of f0 on the z-axis. If it does, it is determined as "strong horizontal glue". Of course, rules regarding frequencies other than the reference frequency may be used. Alternatively, in (A12), a classifier whose input is the intensity of each frequency component may be learned, and the reaction category may be determined using that classifier.

Further, the generating unit 521 may generate more diverse reaction categories than in the first embodiment based on beat information. Specifically, reaction categories are generated based on the time difference between the timing of bars and beats and the timing of peak displacement (acceleration) of the user's body. For example, with reference to the timing of the beat, (F1) the user's body movement (displacement peak) is earlier than a predetermined time (before), and (F2) the absolute value of the time difference between the two is less than the predetermined time. A reaction category is generated by distinguishing between (F3) the case where the user's body movement is slow (later) than a predetermined time. An example of such reaction categories is shown in FIG. FIG. 26 shows three reaction categories (small category, small category, and subcategory) obtained by further subdividing the reaction category (middle category) of "weak horizontal glue" shown in FIG. For the sake of brevity, other reaction categories are omitted. , subcategories, subcategories) are defined. As shown in FIG. 26, three response categories are generated according to the time difference between the timing of the beat and the peak timing of the user's biological information (body movement). To simplify the explanation, branch numbers such as "1-1", "1-2" and "1-3" are shown as category IDs, but of course "1", "2" and "3" A serial number or an arbitrary character string may be used as the category ID.

In the fifth embodiment, the frequency components related to the beats of the music are extracted from the biometric information data, and the frequency components unrelated to the beats are excluded. can be reduced. For example, even if the user moves his/her head unrelated to the music, the influence of the motion on the category data can be reduced. Therefore, it is possible to accurately calculate the degree of similarity between users or the degree of similarity between contents. Therefore, recommendation information can be created with high accuracy.

Further, according to this embodiment, by using the music information data in addition to the biological information data, it is possible to generate more diverse category data. Therefore, it is possible to express the difference in reaction for each user in more detail with the category data. Therefore, it is possible to accurately calculate the degree of similarity between users or the degree of similarity between contents. Also, the target user can specify desired contents and desired users in more detail.
It is also possible to implement this embodiment independently of the fourth embodiment. In other words, after creating a content section without using the music information DB in the same manner as in the first embodiment, the music information DB is used to preferentially select frequency components having a predetermined relationship with the beat and speed of the music from the biometric information. can be extracted to

In the above description, the category data is generated based on the bar and beat information. However, other music information (music information) may be used instead. Specifically, MIDI (Musical Instrument Digital Interface) data (musical instrument performance data) used for karaoke or the like may be used. For example, if you have drum performance data, you can calculate the degree of matching between the performance timing of each percussion instrument that makes up the drum (hi-hat, cymbal, snare drum, bass drum, etc.) and the timing of biological information data (peak timing). , the time difference may be calculated. Of course, it is not limited to drums, and performance data of any musical instrument such as guitar, bass, keyboard, vocals, or singing may be used. Alternatively, processing such as validating biometric information data that matches the timing of playing a musical instrument and invalidating (removing) biometric information data that does not match the timing of playing a musical instrument may be performed.

Also, the type of musical instrument may be reflected in the reaction category. Specifically, it is calculated which musical instrument has a high probability that the movement of the user's body matches the playing timing of the musical instrument, and the name of the musical instrument that matches is included in the reaction category. In addition to the type of musical instrument, the performance style and musical content may be set as response categories. For example, "Respond to hi-hat", "Respond to cymbals", "Respond to snare drum", "Respond to bass drum", "Respond to drum pounding", "Respond to guitar", "Respond to guitar bending", "Response to treble" "Respond to female chorus", "Respond to pitch shift of two octaves or more", and so on. According to the present embodiment, by using detailed music information, it is possible to increase the types of reaction categories, and reflect various user reactions in more detail and with high accuracy in the category data.

<Example 6>
In the sixth embodiment, the difference between the biometric information data before content reproduction and the biometric information data during content reproduction is used to create category data with higher accuracy. The configuration of the recommendation device 5 of the present embodiment is the configuration shown in FIG. 4, which is the same as that of the first embodiment.
In this embodiment, the biometric information data before or after the user reproduces the content, that is, the biometric information data when the content is not reproduced is stored in the biometric information DB 531 . This biometric information when content is not reproduced is also called "second biometric information". At this time, the first biometric information data when the content is reproduced and the second biometric information data when the content is not reproduced are stored in the biometric information DB 531 so as to be distinguishable. Then, the generation unit preferentially extracts frequency components different from the frequency components analyzed from the second biological information among the frequency components analyzed from the first biological information, and based on the extracted frequency components Generate categorical data. The details are described below.

In step S110G (not shown) corresponding to step S110, the generation unit 521 refers to the biometric information DB 531 and the music information DB 537 to generate category data. Specifically, first, using a technique such as Fourier analysis, main frequency components are extracted from the biological information data during non-reproduction. Next, in step S230G (not shown) corresponding to step S230, the generation unit 521 preferentially extracts frequency components other than the frequency components during non-reproduction from the biological information data during reproduction.

FIG. 27 shows an outline of processing for generating category data using frequency components during reproduction and frequency components during non-reproduction. In FIG. 27, the solid line is the frequency component of the biological information data during reproduction, and the broken line is the frequency component of the biological information data during non-reproduction. The frequency component at the time of reproduction has three peaks: peak p11 at frequency f1, peak p12 at frequency f2, and peak p13 at frequency f3. Also, the frequency component during non-playback has three peaks: peak p21 at frequency f4, peak p22 at frequency f5, and peak p23 at frequency f6. The generation unit 521 compares the peak frequency of the frequency component during reproduction with the nearest peak frequency of the frequency component during non-reproduction. Then, the difference between the peak frequency of the frequency component during reproduction and the peak frequency of the frequency component during non-reproduction is calculated. If the difference in frequency is equal to or greater than a predetermined value, the generation unit 521 extracts the frequency component during reproduction, and if the difference in frequency is less than the predetermined value, does not extract the frequency component during reproduction. In the example shown in FIG. 27, f1 and f4 are to be compared, and f1 is to be extracted because the difference between them is equal to or greater than the predetermined value. Also, f2 and f5 are to be compared, and since the difference between them is less than the predetermined value, f2 is not to be extracted. Also, f3 and f6 are to be compared, and since the difference between them is equal to or greater than the predetermined value, f3 is to be extracted. If there is no non-playback frequency peak near the playback frequency peak, or if the strength of the non-playback frequency component near the playback frequency peak is low, the playback frequency component is extracted. be targeted. After extracting the frequency component to be extracted, the generation unit 521 generates category data by the same method as described in the fifth embodiment.

In the present embodiment, category data is generated by removing components irrelevant to content viewing from biometric information data during playback based on biometric information data during non-playback. can generate For example, when content is played back while the user is walking, body displacement due to walking is included in biometric information data, and such components can be reduced. Therefore, category data can be created with higher accuracy, and recommendation information can be created with higher accuracy.

<Example 7>
In the embodiments so far, category data is generated using data indicating the movement of the user's body as biometric information data. In Example 7, category data is generated using biological information such as the user's pulse, pulse wave, heart rate, respiration, blood pressure, etc., in addition to the data indicating the movement of the user's body. In order to simplify the explanation, an example using pulse information will be described below, but similar processing can be performed using information on respiration such as respiration rate or information on blood pressure.
An example of the biosensor in this embodiment is a device capable of measuring a pulse, such as a smart watch. A single biosensor may measure both the user's body movement and pulse. Alternatively, biometric information of one user may be measured with a plurality of sensors, such as measuring the movement of the user's body with a headphone and measuring the user's pulse with a smart watch. The biometric information DB 531 stores pulse information as well as data indicating the movement of the user's body.

The flow of recommendation information creation processing in this embodiment is the same as the flow chart in FIG. In step S110H (not shown) corresponding to step S110, the generation unit 521 generates category data based on the data indicating body movement and the pulse data. Specifically, in step S230H corresponding to step S230, reaction categories are generated using the rule data 535 shown in FIG. FIG. 28 shows three reaction categories (“cool,” “normal,” and “hot”) that are further subdivided from the “weak horizontal glue” reaction category shown in FIG. For the sake of brevity, other reaction categories are omitted. , subcategories, subcategories) are defined. As shown in FIG. 28, three reaction categories are generated depending on whether the pulse is slower than a predetermined criterion, within a predetermined criterion range, or faster than a predetermined criterion. To simplify the explanation, branch numbers such as "1-1", "1-2" and "1-3" are shown as category IDs, but of course "1", "2" and "3" A serial number or an arbitrary character string may be used as the category ID.

Here, the predetermined criteria are any of the following.
(G1) Criteria Based on a Predetermined Threshold For example, if the pulse rate is within the range of "60 or more and less than 80", it is determined that it is within the predetermined criteria. If the pulse is less than "60", it is determined to be slow, and if the pulse is "80" or more, it is determined to be fast.
(G2) Reference based on user's normal value For example, the user's normal pulse rate (reference pulse) in daily life is calculated and used as a reference. Also, the reference pulse rate may be calculated for each time period or activity status. Further, similarly to the concept in the sixth embodiment, the pulse during non-playback (before content playback, etc.) may be used as the reference pulse. If the pulse is less than the reference pulse by a predetermined value (for example, 10) or more, it is determined to be "slow", and if it is greater than the reference pulse by a predetermined value or more, it is determined to be "fast". Alternatively, if the pulse rate is less than the reference pulse rate by a predetermined rate (for example, 10%) or more, it may be determined to be "slow", and if it is greater than the reference pulse rate by a predetermined rate or more, it may be determined to be "fast".

In the example shown in FIG. 28, if the pulse during content playback is slower than a predetermined standard, a response category of "cool (weakly cool)" is generated, and if the pulse during content playback is within the predetermined standard range, For example, a reaction category of "normal (horizontal movement is weak, normal)" is generated, and if the pulse during content reproduction is faster than a predetermined standard, a reaction category of "hot (horizontal movement is weak and hot)" is generated. When the pulse is fast, the user's situation is assumed to be "exercising the whole body", "exercising such as jogging", "extremely excited", and the like. On the other hand, when the pulse is slow, the user's situation is assumed to be "quiet", "lying down", "very calm", and the like. For example, there is a high possibility that user A, who is viewing the same content while jogging, and user B, who is lying on the bed and viewing the same content, have different sensibilities, hobbies, tastes, values, etc. . According to the present embodiment, category data is generated based on more types of biometric information (various biometric information) than in the previous embodiments, so that various reactions of the user during content reproduction can be achieved. can be reflected in the category data more finely (with high accuracy). Also, biometric information data can be made to correspond to more types of reaction categories. Therefore, based on the category data, the degree of similarity between users and the degree of similarity between contents can be calculated with higher accuracy. Therefore, it is possible to accurately reflect various user reactions to the content in the category data, and to create recommendation information with higher accuracy.

<Other embodiments>
It is of course possible to arbitrarily combine <Example 1> to <Example 7>. For example, <Example 2>, <Example 4>, <Example 5>, and <Example 6> may be combined.
As described above, the present disclosure has been described in accordance with the above embodiments, but the present disclosure is not limited only to the configurations of the above embodiments, and is applicable within the scope of the invention of the claims of the present application. Needless to say, it includes various modifications, modifications, and combinations that can be made by a trader.

Further, arbitrary processing of the recommendation device (information processing device) and the terminal device described above can be realized by causing a CPU (Central Processing Unit) to execute a computer program. In this case, the computer program can be stored and provided to the computer using various types of non-transitory computer readable medium. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible discs, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical discs), CD-ROMs (Read Only Memory), CD-Rs, Includes CD-R/W, semiconductor memory (eg mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)). The program may also be delivered to the computer on various types of transitory computer readable medium. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. Transitory computer-readable media can deliver the program to the computer via wired channels, such as wires and optical fibers, or wireless channels.

In addition to the case where the functions of the above-described embodiments are realized by a computer executing a program that realizes the functions of the above-described embodiments, this program can be executed by an OS (OS) running on the computer. Operating System) or application software to realize the functions of the above-described embodiments are also included in the embodiments of the present disclosure. Furthermore, even if all or part of the processing of this program is performed by a function expansion board inserted into the computer or a function expansion unit connected to the computer, and the functions of the above-described embodiments are realized, the present disclosure included in the embodiment of

u1 to u4 users 11 to 14 terminal devices 21 to 24 biosensors 31, 32 base station device 4 content distribution server 5 recommendation device 80 network 510 communication unit 520 processing unit 521 generation unit 522 calculation unit 523 creation unit 530 storage unit 531 biometric information database
532 User Response DB
533 section information DB
534 category-related data 535 rule data 536 program 540 input device 550 display device

Claims (10)

a storage unit that stores user's biological information measured during content reproduction in association with the user and the content;
a generation unit that generates category data including one or more reaction categories indicating the type of user reaction to content based on the biometric information;
a calculation unit that calculates a degree of similarity indicating either similarity between users or similarity between contents based on the category data and category related data indicating the degree of relatedness between reaction categories;
a creation unit that creates recommendation information based on the similarity;
Information processing device. The biometric information includes data indicating body movements of the user in two or more directions,
The generation unit generates the category data using a predetermined reaction category that distinguishes the data in two or more directions.
The information processing device according to claim 1 . The generating unit generates a reaction category for each section into which the content is divided along the time axis, and generates the category data including the reaction categories arranged in the order of the sections.
The information processing apparatus according to claim 1 or 2. the content is a song,
The generation unit divides the content on the time axis by a predetermined number of times the length of the beat or bar of the music piece, and creates a plurality of sections.
The information processing apparatus according to claim 3. The creation unit
calculating the number of users who indicated a predetermined reaction category or the number of reactions for each of a plurality of sections of content, identifying a recommended section according to the calculated number, and creating the recommendation information including the recommended section;
The information processing apparatus according to claim 3 or 4. the content is a song,
The generating unit preferentially extracts, from among the frequency components analyzed from the biological information, frequency components having a predetermined relationship with frequency components based on the beat or speed of the music, and based on the extracted frequency components generating the categorical data;
The information processing apparatus according to any one of claims 1 to 5. The creation unit
creating the recommendation information including at least one of the number of users who have responded to the content in a predetermined response category and the major response categories indicated by the users to the content;
The information processing apparatus according to any one of claims 1 to 6. The storage unit further stores the user's biological information measured while the content is not reproduced,
The generating unit preferentially extracts frequency components different from frequency components extracted from the biological information during non-reproduction of the content among the frequency components extracted from the biological information during content reproduction, and extracts the extracted frequency components. The information processing apparatus according to any one of claims 1 to 7, wherein the category data is generated based on. The creation unit
selecting a plurality of recommended candidate contents based on the similarity;
Among the plurality of recommended candidate contents, content for which another user having a relatively high degree of similarity with the target user indicated a predetermined reaction category, and content matching the reaction category indicated by the target user in the past. Alternatively, content that matches a reaction category specified by the target user is selected as recommended content, and information about the recommended content is included in the recommended information for the target user;
The information processing apparatus according to any one of claims 1 to 8. a process of storing the user's biometric information measured during content reproduction in association with the user and the content;
a process of generating category data including one or more reaction categories indicating the type of reaction of the user to the content based on the biometric information;
a process of calculating a degree of similarity indicating either similarity between users or similarity between contents based on the category data and category related data indicating the degree of relatedness between reaction categories;
a process of creating recommendation information based on the similarity;
A program that makes a computer run

Images