Learning Fingerprints for a Database Intrusion Detection System
ESORICS '02 Proceedings of the 7th European Symposium on Research in Computer Security
Visualization-enabled multi-document summarization by Iterative Residual Rescaling
Natural Language Engineering
| Hi-index | 0.00 |
Summarization technologies today work, in essence, by performing data reduction over the original document source. Document fragments, identified as particularly representative of content, are extracted and offered to the user; typically, such fragments are sentence-sized, and the summary is nothing more than a concatenation of these sentences. We argue that for content characterisation, phrasal units with certain discourse properties are more representative than sentences. From such a position, we outline a model of document content abstraction based on a notion of topically prominent topic stamps. For such abstractions to be useful, they need to retain contextual highlights of their occurrences in the documents; to be usable, they further need to be able to function as windows into the full documents, with suitably designed interfaces for navigation into areas of particular interest. This paper proposes a way for contextualing document highlights, relates this to our model of salience-based content characterization, and demonstrates how the document abstractions derived from such principles facilitate dynamic document content presentation. We argue that dynamic document abstractions effectively mediate different levels of granularity analysis, from terse document highlights to full contextualised foci of particular interest. We close by describing a range of dynamic document viewers which embody novel presentation metaphors for delivery of document content.