Information Technology Reference
In-Depth Information
Fig. 1.9
Sentiment tags evaluating the perception of “Telekom” in comparison to related items
support, both to research topics and newsworthy entities as well as their perception
in the press.
1.6 Conclusion
We presented a system for aggregating and analyzing German news articles collected
from a wide range of news sources. In addition to topic detection and tracking of
news stories, and building a dynamic topic hierarchy based on the current news
situation, the central component implements methods for quotation extraction and
sentiment analysis. As a stylistic means, quotations are used to underline significant
information and can be regarded as a trustworthy piece of information, because
they reflect statements of cited people and organizations in an original and genuine
manner. Our approach to quotation extraction is rule-based and exploits text parts
surrounded by quotation marks for retrieving direct quotations and the presence
of reporting verbs and phrases for detecting indirect quotations. Each quotation is
assigned a quotation speaker by applying a set of predefined rules and including a
crude form of co-reference resolution. Evaluated on a manually created dataset with
German-language quotations, the method yields convincing results, in particular for
direct quotations. We assume quotations being the most subjective parts of news
articles and base our sentiment analysis approach on quotations previously extracted
by our quotation extraction component. We decompose the task into subjectivity
Search WWH ::
Custom Search