DMCA
Genre Classification of Web Pages: User Study and Feasibility Analysis (2004)
| Venue: | IN: BIUNDO S., FRUHWIRTH T., PALM G. (EDS.): ADVANCES IN ARTIFICIAL INTELLIGENCE |
| Citations: | 17 - 1 self |
BibTeX
@INPROCEEDINGS{Eissen04genreclassification,
author = {Sven Meyer zu Eissen and Benno Stein},
title = {Genre Classification of Web Pages: User Study and Feasibility Analysis},
booktitle = {IN: BIUNDO S., FRUHWIRTH T., PALM G. (EDS.): ADVANCES IN ARTIFICIAL INTELLIGENCE},
year = {2004},
pages = {256--269},
publisher = {Springer}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Genre classification means to discriminate between documents by means of their form, their style, or their targeted audience. Put another way, genre classification is orthogonal to a classification based on the documents’ contents. While most of the existing investigations of an automated genre classification are based on news articles corpora, the idea here is applied to arbitrary Web pages. We see genre classification as a powerful instrument to bring Web-based search services closer to a user’s information need. This objective raises two questions: (1) What are useful genres when searching the WWW? (2) Can these genres be reliably identified? The paper in hand presents results from a user study on Web genre usefulness as well as results from the construction of a genre classifier using discriminant analysis, neural network learning, and support vector machines. Particular attention is turned to a classifier’s underlying feature set: Aside from the standard feature types we introduce new features that are based on word frequency classes and that can be computed with minimum computational effort. They allow us to construct compact feature sets with few elements, with which a satisfactory genre diversification is achieved. About 70 % of the Web-documents are assigned to their true genre; note in this connection that no genre classification benchmark for Web pages has been published so far.
Keyphrases
genre classification web page user study feasibility analysis user information need document content new feature automated genre classification genre classifier discriminant analysis arbitrary web page minimum computational effort hand present result standard feature satisfactory genre diversification news article underlying feature targeted audience powerful instrument true genre particular attention neural network learning useful genre web genre usefulness support vector machine compact feature set genre classification benchmark word frequency class web-based search service
