Estimating demographic compositions of television audiences from audience similarities

• US 8,453,173 B1
• Filed: 03/31/2011
• Issued: 05/28/2013
• Est. Priority Date: 12/13/2010
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Lambert, Diane, Zigmond, Daniel J., Anderson, Per R.
• Primary Examiner(s): Goodarzi, Nasser
• Assistant Examiner(s): ALAM, MUSHFIKH I
• Application Number:
  US13/077,493
• Time in Patent Office: 789 Days
• Field of Search: 725 9- 21, 725 32- 36, 725114-116
• US Class Current: 725/35
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: H04H 60/45 for identifying users
• CPC Class Codes: H04N 21/252 Processing of multiple end-users' preferences to derive collaborative data
• CPC Class Codes: H04N 21/254 Management at additional data server, e.g. shopping server, rights management server arrangements for maintenance or administration in data networks H04L41/00; network services using third party service providers H04L67/53
• CPC Class Codes: H04N 21/442 Monitoring of processes or resources, e.g. detecting the failure of a recording device, monitoring the downstream bandwidth, the number of times a movie has been viewed, the storage space available from the internal hard disk arrangements for monitoring broadcast services or broadcast-related services H04H60/29; arrangements for identifying or recognising characteristics with a direct linkage to broadcast information H04H60/35; monitoring of user activities for profile generation for accessing a video database G06F16/739; monitoring in wireless networks H04W24/00
• CPC Class Codes: H04N 7/025 Systems for the transmission of digital non-picture data, e.g. of text during the active part of a television frame transmission of digital non-picture data during the vertical blanking interval only H04N7/088

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for estimating demographic compositions of television audiences from audience similarities. In an aspect, pairwise cosine similarities of segment of viewers of rated and unrated television programs are determined, demographic information of rated programs is attributed to the unrated programs from the cosine similarity measures and demographic information for the rated programs.

Claims

1. A system, comprising:

a data processing apparatus; and

software stored on a computer storage apparatus and comprising instructions executable by the data processing apparatus and upon such execution cause the data processing apparatus to perform operations comprising;

accessing ratings data describing, for each of a plurality of first channel time blocks, demographic characteristics of a viewing audience of the channel time block;

generating for the first channel time blocks, a demographic data structure from the ratings data, the demographic data structure specifying, for each demographic characteristic and first channel time block, a value proportional to a number of viewers having the demographic characteristic that are determined to have viewed the first channel time block;

accessing segment data describing, for each of the plurality of first channel time blocks and a plurality of second channel time blocks for which demographic characteristics are not described by the ratings data, segment characteristics of a viewing audience of the channel time blocks, the segment data being data determined independently from the ratings data;

generating, for the first and second channel time blocks, a segment data structure from the segment data, the segment data structure specifying, for each segment characteristic and channel time block, a value proportional to a number of viewers having the segment characteristic that are determined to have viewed the channel time block;

for each of the second channel time blocks;

determining, from the segment data, respective similarity measures that measure the similarity of the second channel time block to each of the first channel time blocks;

determining, for each of the demographic characteristics, a respective central tendency value of the values that correspond to the demographic characteristic that are stored in the demographic data structure;

for each of the demographic characteristics, identifying first and second subsets of first channel time blocks, the first subset of first channel time blocks having values that meet a threshold with respect to the respective central tendency value of the demographic characteristic, and the second subset of the first channel time blocks having values that do not meet the threshold with respect to the respective central tendency value of the demographic characteristic;

for each of the demographic characteristics, selecting respective first channel time blocks from the second subset of first channel time blocks; and

determining estimated ratings data describing demographic characteristics of a viewing audience of the second channel time block from the respective similarity measures and the demographic characteristics of the selected first channel time blocks.

2. The system of claim 1, wherein the segment data structure comprises:

a segment matrix including rows indexed by the first and second channel time blocks and columns indexed by the segment characteristics, and wherein each cell defined by an intersection of a column and a row stores a value proportional to a number of viewers having the segment characteristic of the column that are determined to have viewed the channel time block indexing the row; and

wherein determining, from the segment data structure, respective similarity measures that measure the similarity of the second channel time block to each of the first channel time blocks comprises determining respective cosine similarity measures from the row indexed by the second channel time block and each respective row indexed by the first channel time blocks.

3. The system of claim 2, wherein the demographic data structure comprises:

a demographic matrix including rows indexed by the first channel time blocks and columns indexed by the demographic characteristics, and wherein each cell defined by an intersection of a column and a row stores a value proportional to a number of viewers having the demographic characteristic of the column that are determined to have viewed the first channel time block indexing the row; and

wherein the respective central tendency value for each demographic characteristic is determined from of the values stored in the cells of the column of the demographic matrix that corresponds to the demographic characteristic.

4. The system of claim 1, wherein the central tendency value is a median value.

5. The system of claim 3, wherein determining the estimated ratings data describing demographic characteristics of a viewing audience of the second channel time block comprises:

for each of the respective first channel time blocks for each column;

selecting the cosine similarity measures that measure the similarity of the second channel time block to first channel time block;

multiplying, by the cosine similarity measure, the value stored in the cell defined by the intersection of the column and the row indexed by the respective first channel time block in the demographic matrix to obtain a respective scaled value; and

generating a vector for the second channel time block, the vector including elements corresponding to the columns of the demographic matrix, and each element storing an estimated value based on the respective scaled values for the column to which the element corresponds, each estimated value proportional to an estimated number of viewers having the demographic characteristic of the column for the scaled values.

6. The system of claim 5, wherein selecting respective first channel time blocks from the second subset of channel time blocks comprises selecting K respective first channel time blocks, wherein K is a value that is predefined for each second channel time block.

7. The system of claim 6, wherein the value of K is selected for each second channel time block, and differs for at least two of the second channel time blocks.

8. The system of claim 7, wherein the instructions further cause the data processing apparatus to perform operations comprising:

for each second channel time block;

selecting, as a training channel time block, a first channel time block that is determined to be most similar to the second channel time block based on the respective similarity measures that measure the similarity of the second channel time block to each of the first channel time blocks;

for each of a plurality of values of K, estimated ratings data describing demographic characteristics of a viewing audience of the training channel time block from the respective similarity measures and the demographic characteristics of K selected first channel time blocks;

for each of the plurality of values of K, determining an error from the ratings data and the estimated ratings data for the training channel time block;

selecting, for the second channel time block, the value of K for which a lowest error is determined.

9. The system of claim 3, wherein determining the estimated ratings data describing demographic characteristics of a viewing audience of the second channel time block comprises:

for each of the respective first channel time blocks for each column;

selecting the cosine similarity measures that measure the similarity of the second channel time block to first channel time block;

normalizing the cosine similarity measures to a range from 0 to 1;

multiplying, by the normalized cosine similarity measure, the value stored in the cell defined by the intersection of the column and the row indexed by the respective first channel time block in the demographic matrix to obtain a respective scaled value; and

generating a vector for the second channel time block, the vector including elements corresponding to the columns of the demographic matrix, and each element storing an estimated value based on the respective scaled values for the column to which the element corresponds, each estimated value proportional to an estimated number of viewers having the demographic characteristic of the column from the scaled values.

10. The system of claim 1, wherein a channel time block defines a network, a television provider, and a time period.

11. The system of claim 1, wherein a channel time block defines a television program and a time period the television program aired.

12. The system of claim 1, wherein the demographic characteristics comprise age and gender.

13. The system of claim 1, wherein segment characteristics are predefined segment clusters.

14. A computer implemented method performed by a data processing apparatus, comprising:

accessing, by the data processing apparatus ratings, data describing, for each of a plurality of first channel time blocks, demographic characteristics of a viewing audience of the channel time block;

generating, by the data processing apparatus and for the first channel time blocks, a demographic data structure from the ratings data, the demographic data structure specifying, for each demographic characteristic and first channel time block, a value proportional to a number of viewers having the demographic characteristic that are determined to have viewed the first channel time block;

accessing, by the data processing apparatus ratings, segment data describing, for each of the plurality of first channel time blocks and a plurality of second channel time blocks for which demographic characteristics are not described by the ratings data, segment characteristics of a viewing audience of the channel time blocks;

generating, for the first and second channel time blocks, a segment data structure from the segment data, the segment data structure specifying, for each segment characteristic and channel time block, a value proportional to a number of viewers having the segment characteristic that are determined to have viewed the channel time block;

for each of the second channel time blocks;

determining, by the data processing apparatus and from the segment data structure, respective similarity measures that measure the similarity of the second channel time block to each of the first channel time blocks;

determining, by the data processing apparatus and for each of the demographic characteristics, a respective central tendency value of the values that correspond to the demographic characteristic that are stored in the demographic data structure;

for each of the demographic characteristics, identifying first and second subsets of first channel time blocks, the first subset of first channel time blocks having values that meet a threshold with respect to the respective central tendency value of the demographic characteristic, and the second subset of the first channel time blocks having values that do not meet the threshold with respect to the respective central tendency value of the demographic characteristic;

for each of the demographic characteristics, selecting, by the data processing apparatus, respective first channel time blocks from the second subset of first channel time blocks; and

determining, by the data processing apparatus ratings, estimated ratings data describing demographic characteristics of a viewing audience of the second channel time block from the respective similarity measures and the demographic characteristics of the selected first channel time blocks.

15. The method of claim 14, wherein the segment data structure comprises:

a segment matrix including rows indexed by the first and second channel time blocks and columns indexed by the segment characteristics, and wherein each cell defined by an intersection of a column and a row stores a value proportional to a number of viewers having the segment characteristic of the column that are determined to have viewed the channel time block indexing the row; and

wherein determining, from the segment data structure, respective similarity measures that measure the similarity of the second channel time block to each of the first channel time blocks comprises determining respective cosine similarity measures from the row indexed by the second channel time block and each respective row indexed by the first channel time blocks.

16. The method of claim 15, wherein the demographic data structure comprising:

a demographic matrix including rows indexed by the first channel time blocks and columns indexed by the demographic characteristics, and wherein each cell defined by an intersection of a column and a row stores a value proportional to a number of viewers having the demographic characteristic of the column that are determined to have viewed the first channel time block indexing the row; and

wherein

the respective median central tendency value for each demographic characteristic is determined from the values stored in the cells of the column of the demographic matrix that corresponds to the demographic characteristic.

17. The method of claim 16, wherein determining the estimated ratings data describing demographic characteristics of a viewing audience of the second channel time block comprises:

for each of the respective first channel time blocks for each column;

selecting the cosine similarity measures that measure the similarity of the second channel time block to first channel time block;

multiplying, by the cosine similarity measure, the value stored in the cell defined by the intersection of the column and the row indexed by the respective first channel time block in the demographic matrix to obtain a respective scaled value; and

generating a vector for the second channel time block, the vector including elements corresponding to the columns of the demographic matrix, and each element storing an estimated value based on the respective scaled values for the column to which the element corresponds, each estimated value proportional to an estimated number of viewers having the demographic characteristic of the column for the scaled values.

18. The method of claim 17, wherein selecting respective first channel time blocks from the second subset of channel time blocks comprises selecting K respective first channel time blocks, wherein K is a value that is predefined for each second channel time block.

19. The method of claim 18, wherein the value of K is selected for each second channel time block, and differs for at least two of the second channel time blocks.

20. The method of claim 19, further comprising:

for each second channel time block;

selecting, as a training channel time block, a first channel time block that is determined to be most similar to the second channel time block based on the respective similarity measures that measure the similarity of the second channel time block to each of the first channel time blocks;

for each of a plurality of values of K, estimated ratings data describing demographic characteristics of a viewing audience of the training channel time block from the respective similarity measures and the demographic characteristics of K selected first channel time blocks;

for each of the plurality of values of K, determining an error from the ratings data and the estimated ratings data for the training channel time block;

selecting, for the second channel time block, the value of K for which a lowest error is determined.

21. A non transitory computer-readable medium comprising instructions that, when executed by a data processing apparatus, cause the data processing apparatus to perform operations, comprising:

accessing ratings data describing, for each of a plurality of first channel time blocks, demographic characteristics of a viewing audience of the channel time block;

generating for the first channel time blocks, a demographic data structure from the ratings data, the demographic data structure specifying, for each demographic characteristic and first channel time block, a value proportional to a number of viewers having the demographic characteristic that are determined to have viewed the first channel time block;

accessing segment data describing, for each of the plurality of first channel time blocks and a plurality of second channel time blocks for which demographic characteristics are not described by the ratings data, segment characteristics of a viewing audience of the channel time blocks, the segment data being data determined independently from the ratings data;

generating, for the first and second channel time blocks, a segment data structure from the segment data, the segment data structure specifying, for each segment characteristic and channel time block, a value proportional to a number of viewers having the segment characteristic that are determined to have viewed the channel time block;

for each of the second channel time blocks;

determining, from the segment data, respective similarity measures that measure the similarity of the second channel time block to each of the first channel time blocks;

determining, for each of the demographic characteristics, a respective central tendency value of the values that correspond to the demographic characteristic that are stored in the demographic data structure;

for each of the demographic characteristics, identifying first and second subsets of first channel time blocks, the first subset of first channel time blocks having values that meet a threshold with respect to the respective central tendency value of the demographic characteristic, and the second subset of the first channel time blocks having values that do not meet the threshold with respect to the respective central tendency value of the demographic characteristic;

for each of the demographic characteristics, selecting respective first channel time blocks from the second subset of first channel time blocks; and

determining estimated ratings data describing demographic characteristics of a viewing audience of the second channel time block from the respective similarity measures and the demographic characteristics of the selected first channel time blocks.