Anzeige
Erschienen in:
2016 | OriginalPaper | Buchkapitel
Labeled Phrase Latent Dirichlet Allocation
verfasst von : Yi-Kun Tang, Xian-Ling Mao, Heyan Huang
Erschienen in: Web Information Systems Engineering – WISE 2016
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
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-of-the-art models, which show the proposed model significantly outperforms baselines in terms of case study, perplexity and scalability.