TGDK.2.3.1.pdf
- Filesize: 2.89 MB
- 34 pages
In multimedia annotation, referencing specific segments of a document is often desired due to its richness and multimodality, but no universal representation for such references exists. This significantly hampers the usage of multimedia content in knowledge graphs, as it is modeled as one large atomic information container. Unstructured data - such as text, audio, images, and video - can commonly be decomposed into its constituent parts, as such documents rarely contain only one semantic concept. Hence, it is reasonable to assume that these advances will make it possible to decompose these previous atomic components into logical segments. To be processable by the knowledge graph stack, however, one needs to break the atomic nature of multimedia content, providing a mechanism to address media segments. This paper proposes a Unified Segmentation Model capable of depicting arbitrary segmentations on any media document type. The work begins with a formal analysis of multimedia and segmentation, exploring segmentation operations and how to describe them. Building on this analysis, it then develops a practical scheme for expressing segmentation in Uniform Resource Identifiers (URIs). Given that this approach makes segments of multimedia content referencable, it breaks their atomic nature and makes them first-class citizens within knowledge graphs. The proposed model is implemented as a proof of concept in the MediaGraph Store, a multimedia knowledge graph storage and querying engine.
Feedback for Dagstuhl Publishing