DMCA
Topic Models for Semantically Annotated Document Collections
Cached
Download Links
| Citations: | 1 - 0 self |
BibTeX
@MISC{Bundschus_topicmodels,
author = {Markus Bundschus and Volker Tresp and Hans-peter Kriegel},
title = {Topic Models for Semantically Annotated Document Collections},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Increasingly, web document collections such as PubMed and DBPedia, but also social bookmarking systems, are annotated with semantic meta data. Given that the number of semantically annotated document collections is expected to increase in the near future, it is of interest to analyze if topic models might be able to play a larger role. Since most of the time, annotations are noisy and even human experts annotate inconsistently, a probabilistic view, as provided by topic models, is appropriate. Besides a number of interesting knowledge discovery tasks, representing topics by meta data has an additional advantage: if the concepts refer to real-world objects, the readability of the topics is greatly improved. In this paper, we present several suitable strategies to model this type of data and show experiments on two large semantically annotated document collections. 1
