TY - GEN
T1 - Labeled phrase latent dirichlet allocation
AU - Tang, Yi Kun
AU - Mao, Xian Ling
AU - Huang, Heyan
N1 - Publisher Copyright:
© Springer International Publishing AG 2016.
PY - 2016
Y1 - 2016
N2 - In recent years,topic modeling,such as Latent Dirichlet Allocation (LDA) and its variations,has been widely used to discover the abstract topics in text corpora. There are two state-of-the-art topic models: Labeled LDA (LLDA) and PhraseLDA. LLDA is a supervised generative model which considers the label information,but it does not take into consideration word order under the bag-of-words assumption. On the contrary,PhraseLDA regards each document as a mixture of phrases,which partly considers the word order. However,PhraseLDA cannot model the supervised label information. In this paper,in order to overcome the defects of two models above while combining their merits,we propose a novel topic model,called Labeled Phrase LDA,which synchronously considers the supervised information and word order. Lots of experiments were conducted among the proposed model and two state-ofthe- art models,which show the proposed model significantly outperforms baselines in terms of case study,perplexity and scalability.
AB - In recent years,topic modeling,such as Latent Dirichlet Allocation (LDA) and its variations,has been widely used to discover the abstract topics in text corpora. There are two state-of-the-art topic models: Labeled LDA (LLDA) and PhraseLDA. LLDA is a supervised generative model which considers the label information,but it does not take into consideration word order under the bag-of-words assumption. On the contrary,PhraseLDA regards each document as a mixture of phrases,which partly considers the word order. However,PhraseLDA cannot model the supervised label information. In this paper,in order to overcome the defects of two models above while combining their merits,we propose a novel topic model,called Labeled Phrase LDA,which synchronously considers the supervised information and word order. Lots of experiments were conducted among the proposed model and two state-ofthe- art models,which show the proposed model significantly outperforms baselines in terms of case study,perplexity and scalability.
KW - Labeled phrase LDA
KW - Multi-labeled corpus
KW - Topic model
UR - https://www.scopus.com/pages/publications/84996598341
U2 - 10.1007/978-3-319-48740-3_39
DO - 10.1007/978-3-319-48740-3_39
M3 - Conference contribution
AN - SCOPUS:84996598341
SN - 9783319487397
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 525
EP - 536
BT - Web Information Systems Engineering – WISE 2016 - 17th International Conference, Proceedings
A2 - Cellary, Wojciech
A2 - Wang, Jianmin
A2 - Mokbel, Mohamed F.
A2 - Wang, Hua
A2 - Zhou, Rui
A2 - Zhang, Yanchun
PB - Springer Verlag
T2 - 17th International Conference on Web Information Systems Engineering, WISE 2016
Y2 - 8 November 2016 through 10 November 2016
ER -