IMPROVING NEWS RANKING BY COMMUNITY TWEETS Xin Shuai, Xiaozhong Liu, Johan Bollen Sunday, April 15, 2012
Different information needs by the same short query Will Obama sing at the next American Idol? Alice What's Will Obama Obama's lower policy towards taxes? China? Carl Bob Sunday, April 15, 2012
Persons from the same community share similar interests Demographic features Age Profession Gender Location ...... Sunday, April 15, 2012
Users interests in news can be inferred from online social media community User interests Sunday, April 15, 2012
Community Tweets Voting Model (CTVM) C1 and C2 are two di ff erent communities! C1 interest updated tweets from V otes ranking original C1 for C1 ranked news documents C2 interest from updated search V otes ranking tweets from engine for C2 C2 Sunday, April 15, 2012
CTVM re-ranks google news results of ‘obama’ for California users Re-Ranking Based on CA tweets 2011-01-30, CA 12:00am, 2011-01-30 0 0 0.041 1 0.047 0 0 0.037 0.011 2 0 0 0.03 0.011 0 0.109 0 3 0.025 0 0 Sunday, April 15, 2012
Four types of data is collected for evaluation Queries: 50 popular queries were manually identified through Google Insight Tweets: Daily tweets for two weeks were collected from three states: California, New Y ork and Taxes by streaming API News documents: Top 10 ranked retrieval documents from Google & Y ahoo News were collected every day Turk evaluations: Sunday, April 15, 2012
More users in CA participate in turking and tweeting than NY and TX # of tweets/day total # of Google news documents evaluated on MTurk total # of Y ahoo news document evaluated on MTurk 5320 news documents from 33 queries are evaluated by 105 Turkers Sunday, April 15, 2012
CTVM performs well for Yahoo News @ CA & NY users CA NY TX Sunday, April 15, 2012
Future works remains... Why our method does not work well for TX Better representation of users interests: named entity, historical e ff ect, ... Query subject di ff erence Other demographic features to form community Sunday, April 15, 2012
Recommend
More recommend
