2. Pew Internet report:
“75% of online news consumers say they get
news forwarded through email or posts on
social networking sites
and 52% say they share links to news with
others via those means.”
3. Twitter Lists
• Filtering main friends timeline is a bad idea
• Twitter Lists: manually created set of users who
often post on a certain topic
• For example:
– @huffingtonpost/apple-news
– @IndieFlix/film-people-to-follow
– @alisohani/bigdata-analytics
• A Twitter user can be included into different lists.
• Me for example:
http://twitter.com/mariagrineva/lists/memberships
4. What kind of noise?
• People tweet on other topics too, including
personal stuff
• Global news widely spread, often really
annoying: IPad launch, ash clouds, Christmas,
Michael Jackson
5. Our Approach
• Identifying niche topic of Twitter list
automatically, at real-time
• Improve the niche topic with respect to the
Global Twitter Stream
– If there is a burst related to Apple, IPad => check
maybe all Twitter is talking about that
6. Filtering = Classification
• Traditional approaches to filter news use only
textual features
• We use both textual and social features for
classification
– Twitter lists is a community of interconnected
users => see who is the center and who is an
outsider
7. What is done
• Method for identification list’s topic
signature with respect to Global Twitter
Stream
• Social features identification
• Evaluation framework