Social media sites (by some referred to as the web 2.0) allow their users to interact with each other, for example in collecting and sharing so-called user-generated content - these can be just bookmarks, but also blogs, images, and videos. Social media support co-creation: processes where customers (or users, if you prefer) do not just consume but play an active role in defining and shaping the end product. Famous examples include Six Degrees, LiveJournal, Digg, Epinions, Myspace, Flickr, YouTube, Linked-in, and Pinterest. Of course, today's internet giants Facebook and Twitter are key new developments. Finally, Wikipedia should not be overlooked - a major resource in many language technologies including information retrieval! The second part of the lecture looks into the opportunities for information retrieval research. Social media platforms tend to provide access to user profiles, connections between users, the content these users publish or share, and how they react to each other's content through commenting and rating. Also, the large majority of social media platforms allow their users to categorize content by means of tags (or, in direct communication, through hash-tags), resulting in collaborative ways of information organization known as folksonomies. However, these social media also form a challenge for information retrieval research: the many platforms vary in functionalities, and we have only very little understanding of clearly desirable features like combining tag usage and ratings in content recommendation! A unifying approach based on random walks will be discussed to illustrate how we can answer some of these questions [1], but clearly the area has ample opportunity to leave your own marks. In the final part of the lecture I will briefly touch upon an even wider range of opportunities, where data derived from social media form a key component to enable new research and insights. I will review a few important results from research centered on Wikipedia, facebook and twitter data, as well as a diverse range of new information sources including the geo- and temporal information derived from images and tweets, product reviews and comments on youtube videos, and how url shorteners may give a view on what is popular on the web. [1] Maarten Clements, Arjen P. De Vries, and Marcel J. T. Reinders. 2010. The task-dependent effect of tags and ratings on social media access. ACM Trans. Inf. Syst. 28, 4, Article 21 (November 2010), 42 pages. http://doi.acm.org/10.1145/1852102.1852107

1. 9th European Summer School in Information Retrieval September 4th, 2013
http://bit.ly/ESSIR13IRSocMedia
IR and Social Media
Arjen P. de Vries
arjen@acm.org
Centrum Wiskunde & Informatica
Delft University of Technology
Spinque B.V.

2. On slideshare,
IR = Investor Relations

3. Social Media
Noun
social media (plural only)
Interactive forms of media that allow users
to interact with and publish to each other,
generally by means of the Internet.
The early 21st century saw a huge increase in social
media thanks to the widespread availability of the
Internet.

4. http://www.webanalyticsworld.net/2010/11/history-of-social-media-infographic.html

5. Social Media
 “Social bookmarking” sites
 “User generated content”
 Images (flickr) and videos (youtube, vimeo), but also
blogs
 Social network services
 Twitter, facebook

6. Not just one beast!

9. IR and Social Media?

10. Red Hot Chili Peppers

11. “Rock group” in
author’s metadata...
Organisation in
groups may help
disambiguate
query!
More implicit
metadata...

12. Information Science
“Search for the fundamental knowledge
which will allow us to postulate and utilize
the most efficient combination of [human
and machine] resources”
 M.E. Senko. Information systems: records, relations, sets, entities,
and things. Information systems, 1(1):3–13, 1975.

13. Core Questions
 How to represent information?
 The information need and search requests
 The objects to be shown in response to an
information request
 How to match information
representations?

14. IR and Social Media
 Richer information representations!

15. Richer representations
 User profiles
 User name, full name, description, image,
homepage url, etc.
 Connections between users
 Networks of friends, followers, etc
 Comments/reactions
 Endorsing and sharing

16. Q: Web ancient social media?

17. (C) 2008, The New York Times Company
Anchor tekst:
“continue reading”

18. Not a lot of info
to represent
the page…
Een fan’s hyves page:
Kyteman's HipHop Orchestra: www.kyteman.com
Kaartverkoop luxor theater:
22 mei - Kyteman's hiphop Orkest - www.kyteman.com
Kluun.nl:
De site van Kyteman
Blog Rockin’ Beats:
De 21-jarige Kyteman
(trompettist, componist en
Producer Colin Benders),
heeft drie jaar gewerkt aan
zijn debuut:
the Hermit sessions.
Jazzenzo:
...een optreden van het populaire
Kyteman’s Hiphop Orkest

20. ‘Co-creation’
 Social Media:
 Consumer becomes a co-creator
 ‘Data consumption’ traces
 In essence: many new sources to play the
role of anchor text
 Tags and/or ratings
 Tweets
 Comments, reviews

21. Potential Benefits for IR
 Expand content representation
 Reduce the vocabulary gap(s) between
creators of content, indexers, and users
 More diverse views on the same content

22. Potential Benefits for IR
 Relevance depends on user context
 User task
 User knowledge

23. Potential Benefits for IR
 Relevance depends on user context
 User task
 User knowledge
 Social media provide an opportunity to
make much better assumptions about
user context
 A specific user’s context
 The variety of user contexts that may exist

24. Maarten Clements, Arjen P. de Vries and Marcel J.T. Reinders.
The task dependent effect of tags and ratings on social media access.
TOIS 28, 4, article 21 (November 2010), 42 pages.

25. LibraryThing

26. LibraryThing
 Items
 People
 Tags
 Ratings
See also: http://www.macle.nl/tud/LT/

27. Synonyms

28. Synonyms

30. Examples
 Humour
 Classic

31. LibraryThing
 Items
 People
 Tags
 Ratings
See also: http://www.macle.nl/tud/LT/

33. Search with Random Walk
 Present nodes according to estimated
probability that a random walk that starts
from (task dependent) starting nodes,
would end at this node
 E.g., tag suggestion starts in a tag node;
personalized search in tag and user nodes

34. Tagging Relationships

36. An item recommendation walk

37. Ratings
 Ratings may enhance the graph, or just
be used for evaluation

38. Personalized Search
 Assume a user who types a single tag as
query

39. Personalized Search

40.  A soft clustering effect smoothly relates
similar concepts before converging to the
background probability

41.  Homographs like “Java” are
disambiguated because the walk starts in
both the query tag and the target user
 So, content that matches the user’s
preference is more likely to be found first

42. Common System Designs

43. Analysis results
 Allowing all users to tag all available
content improves retrieval tasks
 Combining tags and ratings may improve
both search and recommendation tasks

44. Ternary relation lost!
 The UIT matrix represents a ternary
relation, that is lost when creating the
three UI, IT and UT matrices

45. Ternary relation lost!
 The UIT matrix represents a ternary
relation, that is lost when creating the
three UI, IT and UT matrices
 Potentially a problem if tags express opinion
about an item; e.g.,
 “poetry” can independent from item still describe
the user
 “awful” requires to know what item the term
belongs to

47. Tags vs. rating
 Most tags do not deviate far from the
mean rating
 Only few tags strongly correlated with
opinion
 Note: poetry higher quality than chicklit

48. Metadata
 Scientific articles have many types of
metadata associated:
 Abstract
 Author
 Booktitle
 Description
 Journal
 Tags
 Are all these types of metadata useful for
item recommendation?

49. Metadata
 According to Toine Bogers’ PhD thesis:
 Concatenate all fields associated to a single
user’s profile’s items into one huge text field,
and use an off-the-shelf IR model to match
the profile against metadata of the items.
“Profile-centric Matching”
 Or, construct item profiles from meta-data of
all users for that item, and apply an item-
based collaborative filtering approach
“Item-based Hybrid Filtering”
 Author, description, tags, title, url, journal
and booktitle all contribute

50. Finally: a recent case study

51. Artist Popularity?
 Let’s ask widely used social media music
platforms!
 I.e., query their APIs

53. Artist Popularity (1-3)
 Top-5 popular artists in dataset
 Jan 21 – Mar 21
 3 hourly timestamped popularity indices

54. http://bit.ly/ESSIR13IRSocMedia

55. Artist Popularity

56. Artist Popularity (?!)
 Top-5 popular artists in dataset
 Jan 21 – Mar 21
 3 hourly timestamped popularity indices

57. The Black Keys

58. The Black Keys
 Three grammy awards received!

59. The Black Keys
 Web responds, while service based
popularity index is static

60. Implications
 An “artist popularity” index depends on
the platform and its user population
 Web based popularity – estimated via URL
shortener’s API – “reacts” to real-world
events
 Suitable as an academics’ search log
replacement?

61. Implications
 An “artist popularity” index depends on
the platform and its user population
 Web based popularity – estimated via URL
shortener’s API – “reacts” to real-world
events
 Suitable as an academics’ search log
replacement?
 Q: What is the most useful popularity –
one that changes dynamically or one that
lasts?

63. Many topics I skipped…

65. Tweets about blip.tv
 “Twanchor text”
 E.g.: http://blip.tv/file/2168377
 Amazing
 Watching “World’s most realistic 3D city
models?”
 Google Earth/Maps killer
 Ludvig Emgard shows how maps/satellite pics
on web is done (learn Google and MS!)
 and ~120 more Tweets

66. Wikipedia
 Wikipedia contains semantically very rich
annotations:
 Wikipedia Categories
 Wikipedia Lists
 Times (1930, 1931, 1932, etc. etc.)
 Names
 Disambiguation pages
Etc.
 Note: DBPedia is just Wikipedia 

67. Wikipedia
 People have used Wikipedia edit history to
look for events

68. Geotags / POIs
 Many social media items carry explicit geo
information
 Geotags are low-level “coordinates”
 POIs are high-level “point-of-interest” labels
 Applications
 Recommend geo-locations to people
 Predict POI tags from (tweet) text
 Predict where a user will go next

69. Map text to locations
 Build a language model from all tags
assigned to flickr images that belong to a
predefined grid cell
 Neighbouring cells used for smoothing
(like hierarchic language models used
previously for video / scene / shot)
 User frequency of a term in a location
(instead of term frequency)
Neil O’Hare and Vanessa Murdock
Modeling Locations with Social Media
Information Retrieval, February 2013, Volume 16, Issue 1, pp 30-62

70. Placing Images: Easy
http://www.flickr.com/photos/63666148@N00/3615989115/
Athens, Ohio or Athens, Greece?

71. Placing Images: Hard
Ballooning company
in Ottawa

72. Searching the Social Graph
 Search entities, and the relationships
between them, in the (facebook) social
graph
 Clearly IR problems, but who has the data
to work with?
Micheal Curtiss et al.
Unicorn: A System for Searching the Social Graph
PVLDB, Vol. 6, No. 11

73. Crawling
 How to get “the” data?
 Rate limited APIs
 ToS
HEADACHES!

74. Fred Morstatter, Jürgen Pfeffer, Huan Liu and Kathleen M. Carley
Is the Sample Good Enough? Comparing Data from Twitter’s Streaming
API with Twitter’s Firehose
ICWSM 2013

75. Not IR yet, but…
Interesting stuff nevertheless!
de Volkskrant, March 13, 2013
Michal Kosinski, David Stillwell, and Thore Graepel
Private traits and attributes are predictable from digital records of
human behavior
PNAS 2013 ; published ahead of print March 11, 2013,
doi:10.1073/pnas.1218772110

76. Take home message(s)

77. Take home message(s)
 Social media give us IR researchers
access to a rich resource of context
 Including time & location!

78. Take home message(s)
 Social media give us IR researchers
access to a rich resource of context
 Including time & location!
 Gather the right data for your problem
domain, and it may be a good alternative
for not having the click data we all want
so badly

79. Take home message(s)
 Social media give us IR researchers
access to a rich resource of context
 Including time & location!
 Gather the right data for your problem
domain, and it may be a good alternative
for not having the click data we all want
so badly
 Various recommendation and retrieval
tasks exist in social media – can one
theory address all of these?

80. C U @ #ECIR2014 ? !