The Value of Social TV

Knowledge at Wharton

Ask about Knowledge at Wharton

Episode Description

📂 Technology 🏷️ Research

Read the full transcriptHide the full transcript
  1. our guest today is professor Chandra Hill whose research focuses on social television which is a combination of
  2. social media and television Chandra thank you so much for joining us today thank you for having me
  3. so many people tweet as they were watching TV shows why is this so important to the media industry both is
  4. part of figuring out how many people are actually watching the show and also why
  5. does this matter to advertisers so there are a number of reasons why it's interesting and also important the first
  6. one is that for TV shows they can observe in real time immediate response and engagement to the content in the
  7. show whether that be sort of organic response to just the script or TV shows actually incorporating social media
  8. content and the shows and asking people to do for example voting while watching TV but then in addition to looking at
  9. real-time engagement one could also look at the viewers and estimate things like
  10. demographics interest and get a sense for who the people are that are actually watching the TV show so the Nielsen has
  11. just announced that it will start providing demographic data why is this notable and or is it really notable well
  12. we've been estimating demographics from tweets for for quite a long time as the tweets relate to television viewers and
  13. television content so while it's notable in that it's an added service for for Nielsen customers it's something that
  14. can actually be done with publicly available data for free and what's nice about it is that if you have a
  15. methodology to infer demographics of groups or individuals you can do so for a really large number of people and
  16. Twitter handles so Twitter handles represent shows and brands and so it's nice about the ability to actually infer
  17. demographics is that you can do so for a wide range of Twitter handles more so than the popular televisions that
  18. television shows that Nielsen typically follows can you tell us a little bit more about your approach and how it
  19. differs from what Nielsen has announced that it's going to do well so our approach works in the following way
  20. so we start with Twitter handles so usually we focus on television shows and brands but the Twitter handles can
  21. represent anyone for example I could use the Twitter handle of my account and we
  22. start with those twitter handles and for the Twitter handles we grab the followers of the Twitter handle for
  23. those followers we grab their tweets and so each Twitter handle then in our method is represented by all of the
  24. tweets of all of the followers so you can think of this as for a particular show we have all of the follower tweets
  25. not just tweets about the show but tweets about their daily lives once we have this document if you will of all of
  26. the tweets of the followers of a particular handle then we basically create what's called a bag of words so
  27. think of it as just creating you know one big vector of words and the associated counts with them we then
  28. correlate the count so we've normalized the counts in a special way but we correlate the counts on these words with
  29. aggregate level demographics of the shows so we get this data in an interesting way from Facebook through
  30. the Facebook advertising API which allows us to get estimates of the aggregate level proportions of people
  31. that follow a particular thing it could be a television show like I mentioned or
  32. a brand or a person and what we do is correlate the proportions of people that follow a brand or television or person
  33. with these words and find the words that are correlated with different proportions of demographics so examples
  34. of those demographics that we've looked at our gender age education level but then in addition to that Facebook also
  35. allows you to target different interests so people with different interests so we
  36. could even look at estimates of the population for people that like gardening for example or cooking and so
  37. it's fascinating even at this aggregate level where we take the words associated with these
  38. shows the people that you know the words associated with the people that follow the shows and these aggregate level
  39. demographics we can do a really good job when we build models to correlate the words in the demographics at predicting
  40. the demographics of held out groups of people and so why this is so powerful is that you don't have to restrict yourself
  41. to just popular things that you know perhaps a company like Nielsen would typically have estimates for a reliable
  42. estimate you can make estimates for just about anything that has a group of people that talk about their daily lives
  43. that you can sort of isolate this group so speaking of that you have you have something that you call the taco graphic
  44. profile explain exactly what that is so we've just coined this term to basically
  45. mean that people are what they say so groups and individuals you know sort of used specific language on Twitter and on
  46. social media and basically they are what they say so words that people use are highly indicative of both their
  47. demographics and interests the needs of service offers demographic data to industry but your approach takes that
  48. one step further and actually makes recommendations based on the data how do you do that and why is it different than
  49. traditional recommendation systems right so we have built a recommendation engine
  50. on top of what we call these profiles for shows and so we wanted to show that these profiles actually had value and so
  51. what we do is simply calculate the similarity between shows or really Twitter handles based on the words that
  52. people who follow the shows say so in doing so we can calculate the similarity between anything any Twitter handle and
  53. when we have a new set of users we can then sort of give one of the shows or Twitter handles that that user follows
  54. into our big correlation matrix of items or shows and ask based on the similarity
  55. between the item that we give our system in all of the calculations that we've done what
  56. are the things that that user would most likely follow and we find that calculating the similarity between shows
  57. in this way just using the words is is highly predictive when we sort of build this recommendation engine predicting
  58. what people will follow and so the nice thing about this is while there are a lot of strategies for building
  59. recommendation engine so for example using the product network or the network of Twitter handles that form by looking
  60. at Twitter handles that are commonly followed by a lot of people and using the text it means that you don't need
  61. those networks so while Twitter has a large network of users there are a lot of websites that don't write and so here
  62. it's saying that perhaps the the tweets or text can be used as a substitute for those network data when it's not there
  63. and when it is there it can be used to complement it in addition to kind of showing the value there it also helps
  64. with what's called a cold start problem for recommendation engines and basically
  65. yeah so this problem is when you have an item or a product or in the case of TV as we're talking about now that doesn't
  66. really have that many followers then you you're not going to recommend it right so it's quicker to get these tweets you
  67. only need a few followers to start you know calculating the similarity between the twitter handle's
  68. so it'll be more likely to be recommended sooner and it also tends towards making more diverse
  69. recommendations so as opposed to making sort of more popular recommendations in terms of our methodology that we've got
  70. so can you give me an example of a specific instance where you've been able to make some recommendations using a
  71. recommendation engine and others also so we test this on Twitter users and so we're assuming that the things that
  72. Twitter users follow a representative of the things that they're interested in so we basically build our models on one
  73. subset of users and then make predictions on a holdout set of users so we've done this in the
  74. context of television shows reliably and then we've also done this in the context
  75. of brands so so we've focused mostly on television shows and brands but because we're using all of the words that people
  76. say without going in with any ontology meaning you know some list of words which would be a lexicon or
  77. like relationships between known words that are meaningful for television or brands our approach is generalizable so
  78. we take all of the words that people say without restricting them in any way and
  79. what that means is we can calculate the similarity between any two things not just television shows not just brands
  80. and so and it makes their approach very flexible that's a good company or a brand use your approach in-house and if
  81. so how would they go about it absolutely I mean so what's nice about the approach
  82. that that we've developed is it relies strictly on publicly available data so which means the data are free so we've
  83. tested our approach on Twitter users that have revealed their preferences by following there would still be this
  84. extra step needed for a firm to test it in their particular context but if they are trying to just infer the
  85. demographics of Twitter users then it would work that way but if they wanted to use the approach for making
  86. recommendations in their context they would have to test it against their own users now TechCrunch story about
  87. Nielsen's new offering says the company has found it while there are a significant number of people tweeting
  88. about TV and even larger number are consuming that content and sometimes those consumers shed new light on what
  89. type of demographic groups watch a particular show have you seen that in your research and why is it so important
  90. to tease out passive tweet readers along with people who are actively creating content right so we haven't we haven't
  91. focused on that distinction mostly because we decide that somebody's interested in a show based on the fact
  92. that they fall that show not based on the fact that they're tweeting about it and so all of
  93. the users that followers show would be in our data set most those that actively tweet and those two that don't we could
  94. easily compare them my guess is that they're not extremely different but perhaps they are but we could compare
  95. them pretty easily so the nice thing about that would be like if there are in fact differences then it would provide
  96. insights to a company but we focused mostly on you know of the people that follow your brand can we infer or
  97. tv-show can we infer the overall demographics but it would be easy because our approach infers information
  98. for groups of users to infer the demographics of the subset of people that tweet and the subset of people that
  99. don't for a particular show finally one last question people both in and outside
  100. the media industry have complained for years that the traditional system of Nielsen ratings doesn't accurately count
  101. how many people are actually watching the show what are the stakes here and why is doing that so important how can
  102. social media be a game changer in this so the main thing is that the way that you can easily collect the data and sort
  103. of watch viewers enables us to watch a larger number of viewers for free and therefore make inferences about a larger
  104. number of users and TV shows for free so most of the criticism of Nielsen data there there are many but meant one of
  105. the main ones is that there's not a lot of coverage for niche shows or shows that aren't that popular because the way
  106. that historically they've collected data is based on a relatively small panel of
  107. users that have a device in their home to track their viewing patterns and so what social media does is it opens up
  108. the space of users to pretty much anybody that's that's tweeting and so you're not restricted then to infer only
  109. the demographics of popular shows because there will be coverage for all shows and all things on Twitter now say
  110. that with the main caveat that not everybody's on Twitter right so there's going to be a bias of course
  111. if you use sort of only the social media data so while you can make inferences for a larger number of shows you're
  112. going to be biased and you'll have to correct for that bias of the fact that not everybody is on Twitter it's skews
  113. younger and that has to be accounted for but the promise is that you know you can
  114. make inferences cheaper faster for more people and for more shows would you like
  115. to say anything about a question that I should have asked but I haven't so I think maybe just talking about sort of
  116. the future I'm producting maybe what all this means for the TV industry so it's nice to see Nielsen and other companies
  117. beginning to combine different types of data to make more comprehensive pictures
  118. of their their viewing audience but what would be nice to see from companies like
  119. Nielsen our partnerships with data having partnerships with people that have different types of data that the
  120. masses couldn't otherwise get so with Twitter you know sort of inferring demographics is something that we do
  121. it's something that a lot of researchers now are starting to do for various reasons and so that's kind of easy what
  122. would be nice to see would be now you know perhaps customers wouldn't want this but to see them do things like
  123. partner with credit card companies and partner with companies for which it's really difficult to get that data and
  124. see sort of what insights can be drawn about TV viewers from combining data from disparate sources that you know
  125. aren't easy to cut genres thanks so much for speaking with us today you

Brief Summary

Professor Chandra Hill discusses social television's impact on audience measurement and how social media data can enhance viewer insights.

Key points

  • The Taco Graphic Profile. Professor Chandra explains how language on social media reflects demographics and interests.
  • Social Media vs. Nielsen Ratings. The discussion highlights the limitations of traditional Nielsen ratings and the potential of social media data.

Tension

40/ 100Low tension

Related episodes