Utilizing social annotations for topical search in Twitter

1
Utilizing social annotations for topical search
in Twitter

• Saptarshi Ghosh
• BESU Shibpur
• Complex Network Research Group
• CSE, IIT Kharagpur

2
General overview

• Social networks in online world
• Twitter, folksonomies such as Delicious
• Modeling the network evolution
• Improving search services
• Socio-technological networks in offline world
• Indian Railway Network
• Traffic analysis

3
Topical attributes of Twitter users

• Twitter has emerged as an important source of
  information real-time news
• Increasing access through topical search Teevan
  WSDM 2011
• Motivation to discover topical attributes /
  expertise of users
• Potential applications
• Know credentials of a user
• Identify topical experts

4
How to discover topical attributes?

• Prior attempts rely on contents of tweets or
  user-profiles Ramage ICWSM 2010, Pochampally
  SIGIR Workshop 2011
• Many profiles do not give topical information
• Tweets often contain day-to-day conversation ?
  difficult to infer topics Java SNA-KDD 2007,
  Wagner SocialCom 2012
• Proposed methodology
• Use social annotations how a user is described
  by others
• Social annotations gathered through Twitter Lists

5
(No Transcript)
6
Mining Lists to infer topics

• Collect Lists containing a given user U
• Identify Us topics from List meta-data
• Basic IR techniques such as case-folding, remove
  domain-specific stopwords
• Extract nouns and adjectives

7
Topics inferred from Lists
politics, senator, congress, government,
republicans, Iowa, gop, conservative
politics, senate, government, congress,
democrats, Missouri, progressive, women
linux, tech, open, software, libre, gnu,
computer, developer, ubuntu, unix
8
Lists vs. other features
Profile bio
love, daily, people, time, GUI, movie, video,
life, happy, game, cool
Most common words from tweets
Most common words from Lists
celeb, actor, famous, movie, stars, comedy,
music, Hollywood, pop culture
9
Who-is-who service

• Developed a Who-is-Who service for Twitter
• Shows word-cloud for major topics for a given
  user
• http//twitter-app.mpi-sws.org/who-is-who/

N. Sharma, S. Ghosh, F. Benevenuto, N. Ganguly,
K. Gummadi, Inferring who-is-who in the Twitter
social network, WOSN 2012.
10
Search system for topic experts

• Cognos, a search system for topic experts
• http//twitter-app.mpi-sws.org/whom-to-follow/
• Given a query (topic)
• Identify users related to the topic using Lists
• Rank identified users
• Uses ranking scheme based on Lists
• Relevance of user to query
• Popularity of user

11
Cognos results for politics
12
Cognos results for stem cell
13
Evaluation of Cognos

• Evaluations through user-surveys
• Cognos gives accurate results for wide variety of
  queries
• Cognos vs. Twitter Who-To-Follow service
• Judgment by majority voting
• Out of 27 queries, Cognos judged better for 12,
  Twitter WTF better for 11 and tie for 4

S. Ghosh, N. Sharma, F. Benevenuto, N. Ganguly,
K. Gummadi, Cognos Crowdsourcing Search for
Topic Experts in Microblogs, SIGIR 2012.
14
Twitter as a source of information

• Characterizing the experts in Twitter ?
  characterizing Twitter platform as a whole
• What are the topics on which information is
  available on Twitter?

15
Topics in Twitter major topics to niche ones
16
Study on the Indian Railway Network
17
Motivation rail accidents during 2010

• Details of accidents in Wiki page on IR
  accidents
• Considered only accidents due to
• Collision between trains
• Derailment

18
IRN data collection

• Crawled schedules of express trains from
  www.indianrail.gov.in in October 2010
• 2195 express train-routes, 3041 stations
• Scheduled time of each train reaching each
  station
• Express train schedules for several years since
  1991
• From Trains At A Glance time-tables
• Obtained from National Rail Museum, New Delhi

19
Observations

• Many trunk-routes in the Indo-Gangetic Plain
  (IGP) have high daily traffic with low headway
• Bad scheduling of IR traffic
• Routes in north India have especially low headway
  during early morning hours when dense fog is
  likely
• Skewed distribution of daily traffic
• Unbalanced growth of traffic in IGP
• Traffic in some segments in IGP has increased by
  250 in 2009, compared to the traffic in 1991
• Very low construction of new tracks

20
Publication and press coverage
S. Ghosh, A. Banerjee, N. Ganguly. Some insights
on the recent spate of accidents in Indian
Railways. Physica A, Elsevier, 2012.
21
Thank You

• Questions / Suggestions?

22
Backup slides
23
Cognos vs. Twitter Who-To-Follow