Title:	AUTOMATED BLOG CLASSIFICATION: A CROSS DOMAIN APPROACH
Author(s):	Elisabeth Lex , Christin Seifert , Michael Granitzer , Andreas Juffinger
ISBN:	978-972-8924-93-5
Editors:	Pedro Isaías, Bebo White and Miguel Baptista Nunes
Year:	2009
Edition:	1
Keywords:	Blogs, Classification, Cross-Domain
Type:	Full Paper
First Page:	598
Last Page:	605
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	The automated classification of blogs is highly important for the relatively new field of blog analysis. To classify blogs into topics or other categories, usually supervised text classification algorithms are applied. However, supervised text classifiers need a sufficient large amount of labeled data to learn a good model. Especially for blogs, data labeled with terms that capture current and actual topics are not available and data labeled in the past is usually not applicable due to topic drifts. Besides, tagged blogs collected from the web exhibit a vocabulary that is rather heterogeneous, diverse and not commonly agreed upon. In our work, we focus on news-related blogs dealing with current events. Our goal is to classify blog posts into given, common newspaper categories. As a baseline, we have high quality labeled data from a German news corpus. Our approach is to exploit the labeled data from the news corpus and use this knowledge to perform cross-domain classification on the unlabeled blogs. We need a solution with high performance, because both our corpora are dynamic and our classifier model needs to be up-to-date. In this work, we evaluated a number of text classification algorithms with different parameter settings by means of accuracy and complexity. Qualitative and quantitative analysis revealed that a recently proposed centroid-based algorithm, the Class-Feature-Centroid classifier (CFC), serves best for our setting because it achieves a comparable accuracy with state-of-the-art text classifiers and outperforms all other algorithms regarding complexity and memory consumption.
	Go Back