Viewer rating survey method

Abstract

To provide a viewing rating research method which can appropriately research viewing ratings of television broadcasting in accordance with attributes of individuals in a research object area by using viewing log information not including viewing data of individual members of viewers of respective television devices.SOLUTION: The viewing rating research method comprises a first step of selecting a plurality of target households 20c from per-device households 20b in a research object area, a second step of acquiring viewing log information of the target households 20c, a third step of generating viewing information classified by attributes of members of the respective target households 20c from the viewing log information, and a fourth step of researching viewing ratings based on the viewing information classified by attributes. In the first step, attributes of members of the respective per-device households 20b are estimated from viewing log information of television devices 23b of a plurality of per-device households 20b in the research object area, and the target households 20c are selected so that a distribution of whole attributes of the target households 20c matches a prescribed reference distribution.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a method for investigating television broadcast ratings. [Background technology]

[0002] When investigating television broadcast audience ratings for each household or for each individual classified by age, sex, etc. in a target area for a survey of television broadcast audience ratings, it has been common practice to install an audience information detection device in each of a number of sample households selected in advance in the target area to detect when, who was watching which television broadcast, and to collect audience data for each sample household via the audience information detection device, and to use the audience data to conduct a television broadcast audience rating survey in the target area (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2-283191 [Patent Document 2] Patent No. 6433615 Summary of the Invention [Problem to be solved by the invention]

[0004] The method of collecting viewing data by installing dedicated viewing information detection devices in sample households as described above tends to make viewing rate surveys expensive because it requires selecting sample households, installing viewing information detection devices in the sample households, and paying compensation to the sample households.

[0005] Meanwhile, in recent years, televisions and recording devices equipped with television broadcast receivers have become commonplace, equipped with a function to automatically acquire viewing log information indicating which channel (or broadcast station affiliate) of a television broadcast was viewed and when, and to transmit the viewing log information to a server or the like of the manufacturer of the television or recording device via an external network such as the Internet.In the following description, a television having the above function, or a combination of a recording device having the above function and a television to which the recording device is connected, may be referred to as a television device (or simply a device).

[0006] By utilizing the above-described functions of this type of television device, it is possible to collect viewing log information for television broadcasts. However, the viewing log information that can be acquired from this type of television device is viewing log information for each television device, and does not include viewing data for each of the constituent members of the television device's viewers (more specifically, viewers who can watch television broadcasts via the television device).

[0007] For example, in Patent Document 2, the applicant of the present application has proposed a technology for estimating the attributes of members of each household and whether or not each of those attributes has viewed content from the viewing log information of each household using a model created by machine learning processing. By applying the technology disclosed in Patent Document 2 to a household equipped with a television device having the function of transmitting viewing log information as described above, it is possible to estimate the attributes of members of viewers of the television device and whether or not each of those attributes has viewed content.

[0008] However, Patent Document 2 does not sufficiently consider what viewing log information from television devices of which households in the survey area should be used to properly estimate the viewing rate for each individual attribute in the survey area.

[0009] The present invention has been made in view of the above background, and aims to provide an audience rating survey method that enables surveys of television broadcast audience ratings according to the attributes of individuals in a surveyed area to be conducted appropriately using viewing log information that does not include viewing data for each constituent member of the audience of each television device. [Means for solving the problem]

[0010] In order to achieve the above object, the audience rating survey method of the present invention comprises: 1. A viewer rating survey method executed in a viewer rating survey device that executes processing related to a viewer rating survey of television broadcasting, comprising: A first step of selecting a plurality of target devices from a plurality of television devices each including a television broadcast receiver and a viewing data output device capable of outputting viewing log information indicating which channel or broadcast station affiliated television broadcast was viewed and at what time via the receiver, and belonging to a predetermined survey target area; a second step of acquiring viewing log information output from the viewing data output device of each of the plurality of target devices selected in the first step; a third step of generating attribute-specific viewing information indicating which channel or broadcasting station affiliated television broadcast was viewed and at what time from the viewing log information for each target device acquired in the second step, using a first model A created in advance by machine learning processing for each attribute of the constituent members of the viewers of each target device; and a fourth step of conducting an audience rating survey for each of the plurality of target devices based on the attribute-specific viewing information obtained in the third step, The first step is characterized by comprising step 1a of collecting viewing log information output from the viewing data output device of each of the plurality of television devices; step 1b of estimating the attributes of the constituent members of the viewers of each television device from the viewing log information collected in step 1a using model B created in advance by machine learning processing; and step 1c of selecting the plurality of target devices from the plurality of television devices using the attributes of the constituent members of the viewers of each television device estimated in step 1b so that the distribution of the attributes of the constituent members of the entire audience of the plurality of target devices matches a predetermined standard distribution (first invention).

[0011] In this invention, "viewers of a television device" specifically refers to one or more viewers who can watch television broadcasts via the television device. Furthermore, "matching the reference distribution" does not necessarily mean an exact match with the reference distribution, but also includes a near match (approximation). This also applies to "matching" in the following description of the embodiments.

[0012] According to the first invention, in step 1b, by appropriately creating model B, even if the viewing log information of each television device in the survey area does not include viewing data for each constituent member of the viewers of that television device, the attributes of the constituent members of the viewers of each television device in the survey area can be appropriately estimated from the viewing log information.

[0013] Furthermore, in the third step, by appropriately creating Model A, even if the viewing log information of each television device in the survey area does not include viewing data for each constituent member of the viewers of that television device, attribute-specific viewing information can be generated from the viewing log information for each attribute of the constituent members of the viewers of each television device in the survey area (for each attribute estimated in step 1b).

[0014] In the present invention, in step 1c, the plurality of target devices are selected from the plurality of television devices in the survey area so that the distribution of the attributes of the constituent members of the total audience of the plurality of television devices in the survey area matches (including the case where it almost matches) a predetermined standard distribution. In this case, according to various studies by the inventors of the present application, by appropriately setting the standard distribution, it is possible to make the viewing situation data estimated based on the attribute-specific viewing information related to the target devices in the survey area (attribute-specific viewing information generated from viewing log information of the target devices) match the actual audience rating in the survey area as closely as possible.

[0015] Therefore, according to the first invention, it is possible to properly conduct a survey of television broadcast viewership ratings according to the attributes of individuals in the survey area using viewing log information obtained from each television device, which does not include viewing data for each constituent member of the viewers of each television device.

[0016] In one aspect of the present invention, the reference distribution in step 1c may be an actual distribution according to the attributes in the target area of ​​the audience rating survey (a second aspect of the present invention). This allows a known distribution to be used as the reference distribution.

[0017] Furthermore, the first or second invention may further include a fifth step of creating the A model and the B model by machine learning processing using as learning data viewing data obtained regarding television broadcast viewing in a plurality of sample households belonging to the same area as the survey target area (third invention). According to this, after properly creating Model A and Model B and selecting target devices, it becomes possible to conduct an audience rating survey without using viewing data from sample households that belong to the same area as the survey target area, thereby reducing the cost of the audience rating survey.

[0018] Furthermore, in the first or second invention, it is also possible to adopt an aspect further comprising a fifth step of creating the A model and the B model by machine learning processing using as learning data viewing data obtained regarding television broadcast viewing in a plurality of sample households belonging to an area different from the target area of ​​the audience rating survey (fourth invention). This makes it possible to select target devices based on the reference distribution even if no sample households are selected in the survey area, thereby reducing the cost of audience rating surveys. [Brief explanation of the drawings]

[0019] [Figure 1]1 is a diagram showing an overall system according to an embodiment of the present invention. [Figure 2] 2 is a block diagram for explaining a first model created by a first model creation unit of the audience rating survey device shown in FIG. 1. [Figure 3] 2 is a block diagram for explaining a second model created by a second model creation unit of the audience rating survey device shown in FIG. 1. [Figure 4] 2 is a block diagram for explaining the processing of a target household selection unit of the audience rating survey device shown in FIG. 1. [Figure 5] 2 is a block diagram for explaining the processing of an attribute-specific audience information generation unit of the audience rating survey device shown in FIG. 1. [Figure 6] 2 is a flowchart showing the processing of an audience rating survey performed by the audience rating survey device shown in FIG. DETAILED DESCRIPTION OF THE INVENTION

[0020] An embodiment of the present invention will be described below with reference to FIGS. 1 to 6. Te The system has an audience rating survey device 1 that executes processing related to audience rating surveys of television broadcasts, etc. The audience rating survey device 1 is configured, for example, with one or more computers. The computer includes a processor such as a CPU, a memory (storage device), an interface circuit, a communication device, etc. (not shown). The audience rating survey device 1 can acquire viewing data related to television broadcast viewing by each household 20 from multiple households 20 in one or multiple regions (for example, a region on a prefecture-by-prefecture basis, or a region combining multiple prefectures such as the Kanto region or the Kinki region). The viewing data includes viewing log information that indicates which channel (or which broadcasting station affiliate) of television broadcast was viewed and when.

[0021] In this embodiment, each "household 20" is roughly divided into a sample household 20a and a device-specific household 20b. Each device-specific household 20b is a household that includes a television device 23b that includes a television broadcast receiver 21 and a viewing data output device 22b capable of outputting viewing data related to viewing of the television broadcast, and that includes one or more viewers of the television device 23b (more specifically, one or more viewers who can watch the television broadcast received by the receiver 21 of the television device 23b) as its members. In other words, a device-specific household 20b is a combination of a television device 23b and a viewer of the television device 23b.

[0022] In this case, one device-specific household 20b is associated with each individual television device 23b. Therefore, even if a dwelling unit is equipped with multiple television devices 23b, one device-specific household 20b is associated with each of the multiple television devices 23b. In this case, each device-specific household 20b corresponding to each of the multiple television devices 23b may include the same members as viewers. In other words, one or more members of one device-specific household 20b may be members of another device-specific household 20b.

[0023] The television device 23b of each device-specific household 20b may be composed of a television or a television and a recording device connected thereto. The viewing data output device 22b of the television device 23b corresponds to the viewing data output device of the present invention and is composed of, for example, a processor such as a microcomputer, memory, an interface circuit, a communication device, etc. (not shown). The viewing data output device 22b can detect the channel of the television broadcast viewed via the television device 23b including it and can detect the date and time when the television broadcast of that channel was viewed (more specifically, the date and time when viewing of the television broadcast of that channel began and ended). From this detected information, viewing log information of the device-specific household 20b (hereinafter referred to as device-specific viewing log information) can be generated. In other words, the device-specific viewing log information is the viewing log information of all members included in the device-specific household 20b. Furthermore, the device-specific viewing log information corresponds to the viewing log information of the present invention.

[0024] The viewing data output device 22b can then periodically (or in response to a request from) transmit viewing data including the generated device-specific viewing log information to a server (not shown) of the manufacturer of the television or recording device that constitutes the television device 23b via an external network NW formed by the Internet, a telephone line network, or the like. In this case, the viewing data transmitted from the viewing data output device 22b includes, in addition to the device-specific viewing log information, location information (e.g., information indicating the first three digits of the postal code) that has been pre-registered in the television or recording device that constitutes the television device 23b, and identification information for the television device 23b. Note that the identification information for the television device 23b can also be used as identification information for the device-specific household 20b that owns the television device 23b.

[0025] Each sample household 20a is equipped with one or more television broadcast receivers 21 and a viewing data output device 22a capable of outputting viewing data related to viewing of television broadcasts, and includes one or more viewers who can watch the television broadcasts received by the receiver 21 as members, and the viewing data that can be output from the viewing data output device 22a includes viewing log information (hereinafter, personal viewing log information) for each of the members (viewers). The receiver 21 is installed in a television or recording device (not shown) provided in the sample household 20a, and the viewing data output device 22a is connected to the television or recording device.

[0026] Unlike the viewing data output device 22b of the device-specific household 20b, the viewing data output device 22a is a device that functions as a so-called people meter and is composed of, for example, a processor such as a microcomputer (not shown), memory, an interface circuit, a communication device, etc. This viewing data output device 22a can detect the channel of the television broadcast watched by each member of the sample household 20a to which it belongs, as well as the date and time when the member watched the television broadcast on that channel (more specifically, the date and time when the member started and finished watching the television broadcast on that channel), and can generate personal viewing log information from this detected information. In this case, the member watching the television broadcast can be identified, for example, through the operation of a specific terminal device by each member or through human authentication processing based on images captured by a camera.

[0027] Furthermore, the viewing data output device 22a can also generate viewing log information for all members of the sample household 20a (hereinafter referred to as household viewing log information) by integrating the personal viewing log information for each member.The viewing data output device 22a can then periodically (or in response to a request from the audience rating survey device 1) transmit viewing data including the generated personal viewing log information and household viewing log information to the audience rating survey device 1 via the external network NW.In this case, the viewing data transmitted from the viewing data output device 22a can include, in addition to the personal viewing log information and household viewing log information, identification information for the sample household 20a, identification information for the members corresponding to each personal viewing log information, the location of the sample household 20a, and information such as the attributes of each member of the sample household 20a (age, gender, etc.).

[0028] The location of the sample household 20a and information about each member may be pre-registered in the audience rating survey device 1. In that case, the viewing data transmitted from the viewing data output device 22a does not need to include the location of the sample household 20a and information about each member.

[0029] Furthermore, sample household 20a may include, in addition to viewing data output device 22a, a viewing data output device 22b similar to device-specific household 20b. In this case, viewing data output device 22a may acquire household viewing log information via viewing data output device 22b. Furthermore, household viewing log information for sample household 20a can also be generated from individual viewing log information by audience rating survey device 1. In this case, the viewing data transmitted from viewing data output device 22a does not need to include household viewing log information.

[0030] Additionally, the household viewing log information for sample household 20a can also be used as viewing log information equivalent to device-specific viewing log information. For this reason, in the following description, for convenience, the household viewing log information for sample household 20a may be referred to as device-specific viewing log information.

[0031] The audience rating survey device 1 has the following functions realized by the implemented hardware configuration and program (software configuration): a first model learning processing unit 11, a second model learning processing unit 12, a viewing data acquisition unit 13, a target household selection unit 14, an attribute-specific viewing information generation unit 15, and an audience rating estimation unit 16.

[0032] The viewing data acquisition unit 13 is capable of communicating with each viewing data output device 22a of a plurality of sample households 20a via an external network NW, and by performing this communication, is able to acquire the viewing data transmitted from the viewing data output device 22a. Note that the viewing data acquisition unit 13 may acquire the viewing data of each sample household 20a via an appropriate storage device.

[0033] Furthermore, the viewing data acquisition unit 13 can communicate with a server (not shown) of the manufacturer of each television device 23b in the device-specific household 20b, and by performing this communication, it is possible to acquire viewing data of each television device 23b of that manufacturer. Note that, if the audience rating survey device 1 can communicate with the viewing data output device 22b of the device-specific household 20b, the viewing data acquisition unit 13 may acquire the viewing data of the television device 23b in the device-specific household 20b directly from the viewing data output device 22b of that television device 23b. Furthermore, the audience rating survey device 1 may acquire the viewing data of the television device 23b from each manufacturer via an appropriate storage device.

[0034] The first model learning processing unit 11 is a processing unit that creates a first model by machine learning processing to estimate which members of each device-specific household 20b belong to the target area of ​​the audience rating survey (hereinafter referred to as the survey target area) based on device-specific viewing log information for a predetermined period (e.g., one month, several months, one year, etc.) of the device-specific household 20b.

[0035] Here, the attributes of the members of each device-specific household 20b are classified into multiple types of attributes, for example, according to the age and gender of the members. For example, the attributes are classified into multiple types such as children under y1, men aged y1 or older but under y2, women aged y1 or older but under y2, men aged y2 or older, and women aged y2 or older. Hereinafter, the number of types of attributes will be designated as N, and each of the N types of attributes will be appropriately represented as At(i) (i = 1, 2, ..., N). Note that the attributes of the members of each device-specific household 20b can be classified according to various parameters, such as not only age and gender, but also occupation, educational background, etc.

[0036] The first model is created so as to determine, for example, as shown in Figure 2, a response variable indicating the presence or absence of each of the constituent members of N types of attributes At(i) (i = 1, 2, ..., N) in each device-specific household 20b (device-specific household 20b with identification information ID x1 in Figure 2) in the survey area from device-specific viewing log information for a predetermined period of time. Note that in Figure 2 and the block diagrams shown in Figures 3 to 5 described below, processing execution units are indicated by thick lines, and data output units or data are indicated by thin lines.

[0037] The explanatory variables to be input to the first model may be, for example, index values ​​that represent the degree of suitability for each of a plurality of viewing patterns (patterns that classify viewing tendencies) classified according to the channel (or broadcasting station series) of the viewed television broadcast, the characteristics (genre, etc.) of the content of the viewed television broadcast, date and time data (data indicating the month, day of the week, time slot, etc.) relating to the date and time when the television broadcast was viewed, etc. (More specifically, index values ​​that represent the degree to which the household viewing log information for a predetermined period of each device-specific household 20b matches each of a plurality of viewing patterns.) In other words, the larger the index value, the higher the tendency for viewing to be performed in the viewing pattern corresponding to the index value.

[0038] The index value may be, for example, the total viewing time duration during which viewing was performed in a pattern corresponding to each of multiple viewing patterns in device-specific viewing log information for a specified period, or the average viewing time duration during which viewing was performed in the corresponding pattern, or the ratio of these viewing time durations to a standard time duration.

[0039] As an example, suppose that a news program is broadcast on Channel n from 18:00 to 19:00 on Wednesdays of each week in July, and there are four weeks worth of Wednesdays in July (four days), and it is estimated from the device-specific viewing log information of device-specific household 20b that the time spent watching the news program on Channel n from 18:00 to 19:00 on each Wednesday in July was 30 minutes, 40 minutes, 30 minutes, and 50 minutes on the first, second, third, and fourth Wednesdays, respectively.

[0040] In this case, the index value representing the degree of suitability for the viewing pattern of watching a news program on Channel n between 18:00 and 19:00 on Wednesdays in July could be the total viewing time duration of the news program (= 30 minutes + 40 minutes + 30 minutes + 50 minutes = 150 minutes), or the average viewing time duration of the news program (= 150 minutes / 4 = 37.5 minutes), or the ratio (= 0.625) of the total viewing time duration (= 150 minutes) to the total time duration (= 4 hours) for the above time period on Wednesdays in July, or the ratio (= 0.625) of the average viewing time duration (= 37.5 minutes) to the time duration (= 1 hour) for the above time period on Wednesdays in July.

[0041] The first model learning processing unit 11 then creates a first model through machine learning processing that uses, as learning data, viewing log information by device (household viewing log information) for a predetermined period of each of a plurality of sample households 20a belonging to a pre-selected sample region and the actual attributes (any of At(1) to At(N)) of each member of the plurality of sample households 20a. In this case, the viewing log information by device (household viewing log information) for each sample household 20a is acquired via the viewing data acquisition unit 13. The actual attributes of the members of each sample household 20a are identified from known information about each member of the sample household 20a. A known algorithm may be used as the algorithm for the machine learning processing of the first model learning processing unit 11.

[0042] The sample households 20a from which learning data is obtained are not limited to all sample households 20a belonging to the sample area, but may be a portion of all sample households 20a belonging to the sample area. The sample area may be, for example, the survey area. However, a region different from the survey area may also be selected as the sample area. For example, if the survey area is Aomori Prefecture, the sample area is not limited to Aomori Prefecture, but may be a region in the Tohoku region relatively close to Aomori Prefecture (e.g., Sendai City) or the Kanto region. The first model may be, for example, a model similar to the mathematical model for determining household composition in Patent Document 2.

[0043] Additionally, whether the created first model is an appropriate model for estimating the attributes of members of each device-specific household 20b in the survey area can be verified, for example, by confirming whether the attributes of members estimated by the first model from device-specific viewing log information (household viewing log information) obtained from each sample household 20a in the survey area match the attributes of the actual members of the sample household 20a. Furthermore, to create an appropriate first model, the type and number of explanatory variables of the first model may be changed, or the set of sample households 20a from which training data is obtained within the sample area or the sample area may be changed as appropriate. By performing such verification and modification as appropriate, an appropriate first model can be created.

[0044] The second model learning processing unit 12 is a processing unit that creates a second model by machine learning processing to estimate which channel (or broadcasting station network) of television broadcast was watched and at what time for each attribute of the members of each device-specific household 20b in the survey area.

[0045] The second model is configured to estimate, for example, for each combination of a television broadcast channel (or broadcast station network) and date and time data (such as month, day of the week, and time slot) related to the viewing date and time of the television broadcast, which member of the device-specific household 20b viewed the television broadcast corresponding to each combination, from the device-specific viewing log information for the device-specific household 20b for a predetermined period. In this case, a separate second model is created for each type of member attribute. That is, N second models are created corresponding to N types of attributes At(i) (i = 1, 2, ..., N).

[0046] For example, if the channel in the combination is Channel 1 and the date and time data is a time slot from 18:00 to 19:00 on Wednesday in July, the television broadcast corresponding to the combination means a television broadcast on Channel 1 from 18:00 to 19:00 on Wednesday in July. The same applies to other combinations of channel (or broadcasting station) and date and time data. The time slot in the date and time data is a time slot in a predetermined time span, such as one hour or 30 minutes.

[0047] More specifically, the second model corresponding to each attribute At(i) is constructed so as to determine, for example, as shown in Figure 3, a response variable indicating an estimated result of whether or not members of the device-specific household 20b with the attribute At(i) of the device-specific household 20b in the survey area (in Figure 3, device-specific households 20b with members with the attribute At(n)) watched a television broadcast corresponding to each combination of channel (or broadcast station series) and date and time data, from household viewing log information for a predetermined period of time for each device-specific household 20b with members with the attribute At(i), for example. In this case, as with the explanatory variables of the first model, index values ​​indicating the degree of suitability for each of a plurality of viewing patterns can be used as explanatory variables input to the second model corresponding to each attribute At(i).

[0048] In addition, Figure 3 shows four example combinations of two broadcast station series X and Y and date and time data, and also shows examples of the determination results of whether or not the members of attribute At(n) watched the television broadcasts corresponding to each of the four combinations.

[0049] Then, the second model learning processing unit 12 creates a second model corresponding to each attribute At(i) by machine learning processing that uses the device-specific viewing log information (household viewing log information) and personal viewing log information for a predetermined period of each of the multiple sample households 20a belonging to the sample area as viewing data regarding television broadcast viewing in each of the multiple sample households 20a.

[0050] In this case, household viewing log information and personal viewing log information for each sample household 20a are acquired via the viewing data acquisition unit 13. A known algorithm may be used as the machine learning processing algorithm of the first model learning processing unit 11. The sample households 20a from which learning data is obtained are not limited to all sample households 20a belonging to the sample area, but may be some of the sample households 20a among all sample households 20a belonging to the sample area. The second model may be, for example, a model similar to the mathematical model for determining personal viewing in Patent Document 2.

[0051] Additionally, whether the created second model is an appropriate model for estimating whether or not a television broadcast corresponding to each combination of channel (or broadcast station) and date / time data was viewed for each attribute of each device household 20b in the survey area can be verified by confirming whether or not the estimated viewing status obtained by the second model from device-specific viewing log information (household viewing log information) obtained from each sample household 20a in the survey area matches the actual viewing status identified from the individual viewing log information of the sample household 20a. Furthermore, to create an appropriate second model, the type and number of explanatory variables of the second model may be changed, or the set of sample households 20a in the sample area from which training data is obtained, or the sample area itself, may be changed as appropriate. By performing such verification and modification as appropriate, an appropriate second model can be created.

[0052] The target household selection unit 14 is a processing unit that, when conducting an audience rating survey in the survey area, selects a plurality of target device-specific households 20c (hereinafter simply referred to as target households 20c) to be the targets of the audience rating survey from a plurality of device-specific households 20b belonging to the survey area. Note that, since each device-specific household 20b is associated with an individual television device 23b, selecting a target household 20c is equivalent to selecting a television device 23b corresponding to the target household 20c. The television device 23b corresponding to the target household 20c corresponds to the target device in the present invention.

[0053] As shown in Figure 4, the target household selection unit 14 first estimates the attributes of the members of each device-specific household 20b in the survey area from device-specific viewing log information for a predetermined period acquired for each device-specific household 20b via the viewing data acquisition unit 13, using a first model created by the first model learning processing unit 11 (a first model that has been verified to be able to properly estimate the attributes of the members of each device-specific household 20b in the survey area).

[0054] Then, the target household selection unit 14 selects multiple target households 20c from all of the device-specific households 20b (device-specific households 20b from which the attributes of the constituent members have been estimated) so that the distribution of the attributes of all constituent members of the multiple target households 20c to be selected (the distribution of the number of constituent members for each attribute At(1) to At(N)) matches a predetermined standard distribution. In this case, the standard distribution can be determined by reflecting the results of an audience rating survey conducted in the survey area (for example, an audience rating survey based on viewing data from sample households 20a in the survey area) so that an appropriate (highly reliable) audience rating survey can be conducted from the device-specific viewing log information of the target households 20c selected based on the standard distribution.

[0055] Specifically, device-specific viewing log information for device-specific households 20b in the survey area is obtained for a specified period including the time of the conducted audience rating survey, and target households 20c are selected based on an appropriate standard distribution using the attributes of the members of each household estimated from the device-specific viewing log information using the first model.

[0056] Furthermore, by using the viewing log information by device of the target household 20c, the attribute-specific viewing information generation unit 15 and the audience rating estimation unit 16 (described later) are processed to estimate viewing situation data (data equivalent to audience ratings) for each attribute, and it is verified whether the viewing situation data matches the actual audience ratings obtained from the audience rating survey that was conducted. Then, depending on the verification results, it is possible to determine an appropriate standard distribution by appropriately changing the standard distribution.

[0057] An example of the reference distribution may be the actual distribution of the attributes of residents in the survey area. In this case, the target household selection unit 14 selects multiple target households 20c so that the ratio of members with each attribute At(i) (i=1, 2, ..., N) to the total number of members of the multiple target households 20c matches the actual ratio of the number of residents with the attribute A(i) to the total number of residents in the survey area.

[0058] The total number of target households 20c can be set arbitrarily, for example, above a predetermined lower limit. Furthermore, the target households 20c may include all or part of the sample households 20a in the survey area in addition to the device-specific households 20b. This also applies to the target households 20c selected in the process of determining an appropriate standard distribution. In this case, the attributes of the members of each sample household 20a referenced in selecting the target households 20c may be actual attributes based on known information about the members of the sample household 20a, or attributes estimated by the first model from the device-specific viewing log information (household viewing log information) of the sample household 20a.

[0059] The attribute-specific viewing information generation unit 15 is a processing unit that generates attribute-specific viewing information indicating whether or not members of each attribute of the target household 20c viewed the television broadcasts that are the subject of the audience rating survey (hereinafter referred to as survey target broadcasts) among the television broadcasts corresponding to each combination of the television broadcast channel (or broadcast station affiliate) and date and time data (data such as month, day of the week, time slot, etc.) related to the date and time of viewing of the television broadcast, for each target household 20c selected by the target household selection unit 14. Note that the survey target broadcasts are not limited to program broadcasts such as predetermined news programs, drama programs, etc., but may also include any television broadcasts (e.g., advertising broadcasts) in a predetermined time slot on the channel or broadcast station affiliate that is the subject of the survey.

[0060] As shown in Fig. 5, the attribute-specific viewing information generation unit 15 first estimates whether each target household 20c (in Fig. 5, target households 20c with identification information IDs x1, x4, and x5) selected by the target household selection unit 14 has viewed a survey target broadcast using the second model (more specifically, the second model verified to be capable of properly estimating whether a television broadcast corresponding to each combination of channel (or broadcast station affiliate) and date and time data has been viewed for each attribute of the constituent members of each device-specific household 20b in the survey target area) created by the second model learning processing unit 12 for each attribute of the constituent members of the target household 20c from device-specific viewing log information for a predetermined period acquired via the viewing data acquisition unit 13 (hereinafter, this estimation process is referred to as attribute-specific viewing estimation process). Note that Fig. 5 illustrates, for example, a television broadcast aired on broadcast station affiliate x on Wednesday, July 7th from 18:00 to 19:00. FIG. 5 also illustrates a case where each of the target households 20c having the identification information IDs x1, x4, and x5 has members with the same attribute At(n).

[0061] In addition, the attribute-specific viewing information generation unit 15 determines, from the device-specific viewing log information for a specified period of each target household 20c, whether or not viewing (viewing by target household 20c) of a television broadcast on the channel or broadcasting station affiliate (broadcasting station affiliate x in Figure 5) of the broadcast to be surveyed actually took place during the time period of the broadcast to be surveyed (the time period from 18:00 to 19:00 on Wednesday, July 7th in Figure 5) (hereinafter, this determination process will be referred to as device-specific viewing determination process).

[0062] Then, for each target household 20c, the attribute-specific viewing information generation unit 15 determines whether or not members of each attribute belonging to each target household 20c have viewed the survey target broadcast, based on the results of the attribute-specific viewing estimation process and the device-specific viewing determination process, and generates the determined results as attribute-specific viewing information. In this case, if the attribute-specific viewing estimation process estimates that a member of a certain attribute of the target household 20c has viewed the survey target broadcast, and the device-specific viewing determination process determines that the member has viewed the survey target broadcast during the time slot of the survey target broadcast on the same channel (or broadcast station affiliate) as the survey target broadcast, it is determined that the member of that attribute has viewed the survey target broadcast.

[0063] For example, in Figure 5, for target household 20c with identification information ID x1, the attribute-specific viewing estimation process for the constituent members of attribute At(n) estimates that the survey target broadcast was viewed, and the device-specific viewing determination process determines that the constituent members of attribute At(n) viewed the survey target broadcast during the time period of the survey target broadcast on the same broadcasting station network as the survey target broadcast, so it is determined that the constituent members of attribute At(n) viewed the survey target broadcast.

[0064] Furthermore, if the attribute-specific viewing estimation process estimates that a member of the target household 20c with a certain attribute has not viewed the surveyed broadcast, it is determined that the member with that attribute has not viewed the surveyed broadcast, regardless of the results of the device-specific viewing determination process.

[0065] For example, in Figure 5, for target household 20c with identification information ID x4, the attribute-specific viewing estimation process for members with attribute At(n) estimates that the member did not watch the survey target broadcast, and therefore, regardless of the results of the device-specific viewing determination process, it is determined that the member with attribute At(n) did not watch the survey target broadcast.

[0066] Furthermore, if the device-specific viewing determination process determines that, for target household 20c, there was no viewing of the survey target broadcast on the same broadcasting station network as the survey target broadcast during the time period of the survey target broadcast (i.e., no television broadcast was viewed during the time period of the survey target broadcast, or a television broadcast on a different channel (or a different broadcasting station network) than the survey target broadcast was viewed), then it is determined that no member of any attribute of target household 20c did not watch the survey target broadcast, regardless of the results of the attribute-specific viewing estimation process.

[0067] For example, in Figure 5, for target household 20c with identification information ID x5, the device-specific viewing determination process determines that there was no viewing of the survey target broadcast on the same broadcasting station as the survey target broadcast during the time period of the survey target broadcast, so regardless of the result of the attribute-specific viewing estimation process, it is determined that no members of target household 20c with any attribute (including members with attribute At(n)) watched the survey target broadcast.

[0068] The audience rating estimation unit 16 is a processing unit that estimates audience status data (data equivalent to audience ratings) of the survey target broadcast for each attribute of all constituent members of the target household 20c, based on the attribute-specific audience information obtained by processing of the attribute-specific audience information generation unit 15. In this case, if the total number of constituent members for each attribute At(i) (i = 1, 2, ..., N) in all of the target households 20c is denoted as M(i) and the total number of constituent members who have been confirmed to have watched the survey target broadcast for each attribute At(i) is denoted as m(i), the audience rating estimation unit 16 calculates the audience status data of the survey target broadcast for each attribute At(i) using the following formula (1). Viewing status data for attribute At(i) = m(i) / M(i) ……(1)

[0069] For example, in Figure 5, if, of all the target households 20c, there are three target households 20c with members having the attribute At(n): ID=x1, ID=x4, and ID=x5, then the only target household 20c whose members with the attribute At(n) have been confirmed to have watched the survey target broadcast is the target household 20c with ID=x1, and therefore the viewing status data for the survey target broadcast calculated using equation (1) for the attribute At(n) is 1 / 3 (≒0.33). In this embodiment, each functional unit of the audience rating survey device 1 is configured to execute the processes as described above.

[0070] Next, the overall processing of the audience rating survey device 1 when conducting an audience rating survey for an arbitrary survey target broadcast will be described with reference to Fig. 6. In this case, it is assumed that the first and second models appropriate for the survey target area have already been created.

[0071] The audience rating survey device 1 executes the processing of steps 1 to 3 by the target household selection unit 14. In this case, in step 1, the target household selection unit 14 acquires device-specific viewing log information for a predetermined period of each device-specific household 20b in the survey target area via the viewing data acquisition unit 13. Note that in this case, the acquired device-specific viewing log information may include household viewing log information of a sample household 20a in the survey target area. In this case, the sample household 20a is considered to be a device-specific household 20b.

[0072] Next, in STEP 2, the target household selection unit 14 estimates the attributes of the members of each device-specific household 20b in the survey area using the first model from the device-specific viewing log information of the device-specific household 20b. Note that for the sample household 20a deemed to be the device-specific household 20b, the attributes of its members may be estimated using the first model, but instead of this estimation, the attributes of each member may be identified from known information about each member of the sample household 20a.

[0073] Next, in STEP 3, the target household selection unit 14 selects target households 20c from the device-specific households 20b in the survey area based on a standard distribution of attributes predetermined for the survey area. In this case, as described above, the target household selection unit 14 selects target households 20c so that the distribution of attributes of the entire constituent members of the target households 20c matches (or nearly matches) the standard distribution. Note that the selected target households 20c may include sample households 20a that are considered to be device-specific households 20b.

[0074] Next, the audience rating survey device 1 executes the processing of STEPs 4 and 5 using the attribute-specific viewing information generation unit 15. In this case, the attribute-specific viewing information generation unit 15 acquires device-specific viewing log information for each target household 20c in STEP 4. The device-specific viewing log information is acquired via the viewing data acquisition unit 13. Alternatively, device-specific viewing log information corresponding to each target household 20c is acquired from the entire device-specific viewing log information of device-specific households 20b acquired by the target household selection unit 14 in STEP 1.

[0075] Next, in STEP 5, the attribute-specific viewing information generation unit 15 uses the device-specific viewing log information and the second model to generate attribute-specific viewing information for each of the target households 20c, indicating whether the survey target broadcast was viewed for each attribute of the constituent members of the target household 20c. In this case, as described above with reference to Fig. 5, the attribute-specific viewing information generation unit 15 executes a process (attribute-specific viewing estimation process) to estimate whether the survey target broadcast was viewed for each attribute of the constituent members of the target household 20c using the second model from the device-specific viewing log information of each target household 20c, and a process (device-specific viewing determination process) to determine whether the survey target broadcast was actually viewed in the target household 20c during the time slot of the survey target broadcast, and based on the results of these processes, determines whether the survey target broadcast was viewed for each attribute of the constituent members of the target household 20c.

[0076] Next, in STEP 6, the audience rating survey device 1 estimates the viewing situation data of the survey target broadcast for the entire target household 20c for each attribute of the constituent members by the audience rating estimation unit 16. In this case, the audience rating estimation unit 16 estimates the viewing situation data for each attribute based on the attribute-specific viewing information generated by the attribute-specific viewing information generation unit 15, as described above, based on the above formula (1).

[0077] Supplementally, in this embodiment, the above STEPs 1 to 3 correspond to the first step in the present invention, and STEPs 1, 2, and 3 correspond to step 1a, step 1b, and step 1c, respectively, in the present invention. Also, STEPs 4, 5, and 6 correspond to step 2, step 3, and step 4, respectively, in the present invention. Also, the first model and the second model correspond to model B and model A, respectively, in the present invention, and the processing of first model learning processing unit 11 corresponds to step 5 in the present invention.

[0078] According to the embodiment described above, by creating appropriate first and second models corresponding to the survey area and appropriately setting a reference distribution corresponding to the survey area, it is possible to accurately estimate viewing situation data for each attribute of the survey area with high reliability without requiring personal viewing log information of sample households 20a in the survey area. Furthermore, since the audience rating survey can be conducted without using the viewing data of sample households 20a, the audience rating survey can be conducted at low cost. [Explanation of symbols]

[0079] 20a...sample household, 21...receiver, 22b...viewing data output device, 23b...television device, television device 23b of target household 20c...target device.

Claims

1. A viewer rating survey method executed in a viewer rating survey device that executes processing related to a viewer rating survey of television broadcasts, comprising: a first step of selecting a plurality of target devices from a plurality of television devices each including a television broadcast receiver and a viewing data output device capable of outputting viewing log information indicating which channel or broadcast station affiliated television broadcast was viewed and at what time via the receiver, and belonging to a predetermined survey target area; a second step of acquiring viewing log information output from the viewing data output device of each of the plurality of target devices selected in the first step; a third step of generating attribute-specific viewing information indicating which channel or broadcasting station affiliated television broadcast was viewed and at what time from the viewing log information for each target device acquired in the second step, using a first model A created in advance by machine learning processing for each attribute of the constituent members of the viewers of each target device; and a fourth step of conducting an audience rating survey for each of the plurality of target devices based on the attribute-specific audience information obtained in the third step, a step 1b of estimating the attributes of the constituent members of the viewers of each television device from the viewing log information collected in step 1a using a model B created in advance by machine learning processing; and a step 1c of selecting the target devices from the plurality of television devices using the attributes of the constituent members of the viewers of each television device estimated in step 1b so that the distribution of the attributes of the constituent members of the viewers of all the target devices matches a predetermined standard distribution.

2. 2. The audience rating survey method according to claim 1, The audience rating survey method, wherein the reference distribution in step 1c is an actual distribution according to the attributes in the target area of ​​the audience rating survey.

3. 3. The audience rating survey method according to claim 1 or 2, A method for audience rating surveys, further comprising a fifth step of creating the A model and the B model by machine learning processing using viewing data obtained regarding television broadcast viewing in a plurality of sample households belonging to the same area as the survey target area as learning data.

4. 3. The audience rating survey method according to claim 1 or 2, A method for audience rating surveys, further comprising a fifth step of creating the A model and the B model by machine learning processing using viewing data obtained regarding television broadcast viewing in a plurality of sample households belonging to an area different from the target area of ​​the audience rating survey as learning data.

Images