Probabilistic latent semantic indexing
Proceedings of the 22nd annual international ACM SIGIR conference on Research and development in information retrieval
The Journal of Machine Learning Research
Why we twitter: understanding microblogging usage and communities
Proceedings of the 9th WebKDD and 1st SNA-KDD 2007 workshop on Web mining and social network analysis
Labeled LDA: a supervised topic model for credit attribution in multi-labeled corpora
EMNLP '09 Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Volume 1 - Volume 1
What is Twitter, a social network or a news media?
Proceedings of the 19th international conference on World wide web
Patterns of temporal variation in online media
Proceedings of the fourth ACM international conference on Web search and data mining
The Joint Inference of Topic Diffusion and Evolution in Social Communities
ICDM '11 Proceedings of the 2011 IEEE 11th International Conference on Data Mining
Discovering geographical topics in the twitter stream
Proceedings of the 21st international conference on World Wide Web
Hi-index | 0.00 |
In this paper, we focus on the issue of characterizing microblogs' topics based on topic models. Different from dealing with traditional textual media (such as news documents), modeling microblogs has three challenges: 1) too much noise; 2) short text; and 3) content incompleteness. Previously, all these limitations have been investigated separately. Some work filters the noise through a prior classification; some enhances the text through the user's blog history; and some utilizes the social network. However, none of these work could solve all the above limitations simultaneously. To solve this problem, we make a combination of previous work in this paper, and propose a unified generative model for characterizing microblogs' topics. In the proposed unified approach, all the three limitations could be solved. A collapsed Gibbs-sampling optimization method is derived for estimating the parameters. Through both qualitative and quantitative analysis in Twitter, we demonstrate that our approach consistently outperforms previous methods at a significant scale.