Output details
11 - Computer Science and Informatics
University of Aberdeen
Weakly supervised joint sentiment-topic detection from text
<24> TKDE is one of the most prestigious journals in the field of data mining. This paper presents a novel weakly-supervised probabilistic model for sentiment analysis, which is the first topic model able to detect document-level sentiment polarity and extract sentiment-bearing topics simultaneously from text. This work builds upon a full conference paper (14.5% acceptance rate for 800+ submissions) by Lin and He, which was presented at CIKM 2009 (147 Google Scholar citations). The contribution of this work led to a research project ViolenceDET funded by EPSRC and DSTL and the proposed model has also been commercialized by Temis (www.temis.com).