A Continuous Bursty Topic Discovery from Twitter through Topic Sketch

다국어 초록 (Multilingual Abstract)

Twitter has turned out to be one of the biggest microblogging stages for clients around the globe to impart anything happening around them to companions and past. A bursty theme in Twitter is one that triggers a surge of applicable tweets inside a brief timeframe, which frequently reflects vital occasions of mass intrigue. Step by step instructions to influence Twitter for early location of bursty points has in this way turn into a vital research issue with enormous viable esteem. Regardless of the abundance of research work on subject demonstrating and investigation in Twitter, it remains a test to distinguish bursty themes progressively. As existing strategies can barely scale to handle the assignment with the tweet stream progressively, we propose in this paper TopicSketch, a draw based subject model together with an arrangement of procedures to accomplish continuous discovery. We assess our answer on a tweet stream with more than 30 million tweets. Our investigation comes about show both productivity and viability of our approach. Particularly it is additionally exhibited that TopicSketch on a solitary machine can possibly handle several millions tweets for each day, which is on the same scale of the total number of daily tweets in Twitter, and present bursty events in finer-granularity.

참고문헌 (Reference)

1 G. Cormode, "What’s hot and what’s not: tracking most frequent items dynamically" 296-306, 2003

2 C. Li, "Twevent: segment-based event detection from tweets" 155-164, 2012

3 X. Wang, "Topics over time: a non-markov continuous-time model of topical trends" 424-433, 2006

4 D. He, "Topic dynamics: an alternative model of bursts in streams of topics" 443-452, 2010

5 S. Petrovi´c, "Streaming first story detection with application to twitter" 181-189, 2010

6 S. Ross, "Stochastic processes" Wiley 1996

7 T. Hofmann, "Probabilistic latent semantic indexing" 50-57, 1999

8 G. P. C. Fung, "Parameter free bursty events detection in text streams" 181-192, 2005

9 M. D. Hoffman, "Online learning for latent dirichlet allocation" 23 : 856-864, 2010

10 K. R. Canini, "Online inference of topics with latent dirichlet allocation" 5 : 65-72, 2009

11 J. Allan, "On-line new event detection and tracking" 37-45, 1998

12 L. AlSumait, "On-line lda: adaptive topic models for mining text streams with applications to topic detection and tracking" 2008

13 J. Leskovec, "Meme-tracking and the dynamics of the news cycle" 497-506, 2009

14 D. Blei, "Latent dirichlet allocation" 3 : 993-1022, 2003

15 T. Griffiths, "Finding scientific topics" 101 (101): 5228-5235, 2004

16 Q. Diao, "Finding bursty topics from microblogs" 1 : 536-544, 2012

17 J. Weng, "Event detection in twitter" 2011

18 T. Sakaki, "Earthquake shakes twitter users: realtime event detection by social sensors" 851-860, 2010

19 C. Jin, "Dynamically maintaining frequent items over a data stream" 287-294, 2003

20 D. M. Blei, "Dynamic topic models" 113-120, 2006

21 C. Wang, "Continuous time dynamic topic models" 579-586, 2008

22 J. Kleinberg, "Bursty and hierarchical structure in streams" 7 (7): 373-397, 2003

23 G. S. Manku, "Approximate frequency counts over data streams" 346-357, 2002

24 G. Cormode, "An improved data stream summary: the count-min sketch and its applications" 55 (55): 58-75, 2005

25 A. Ihler, "Adaptive event detection with time-varying poisson processes" 207-216, 2006

26 L. Hong, "A time-dependent topic model for multiple text streams" 832-840, 2011

27 T. Brants, "A system for new event detection" 330-337, 2003