The Economy of Distributed Metadata Authoring by Stefano Mazzocchi The presentation will sketch the differences between data creation and metadata creation, outlining the impact of these differences on the economy of distributed content creation and consumption. It will be shown how these economical effects might impact both semantically-enhanced distributed technologies and communities of users of these technologies. Finally, it will be suggested how the economical and social projections can be used as a metric for the feasibility of a proposed technologies that involve highly distributed environments. Experts' Workshop - Perspectives on networked knowledge spaces 25/26 October 2002, Sankt Augustin, Germany Organised by: MARS Exploratory Media Lab at the Fraunhofer Institut für Medienkommunikation
What is Metadata Metadata is information about information
Classic Examples of Metadata Keywords Author Date of creation/modification Address/Identifier
More provocative examples Punctuation in text Layout on a page Font size/weight/style in text Commentary audio tracks on DVD
General Metadata Properties Metadata is data about data, but it’s still data Metadata should be semantically orthogonal: data should be understandable even without metadata
Markup and Metadata Markup languages can be seen as metadata-driven languages. Markup syntax is designed to keep data and metadata orthogonal
The Importance of Metadata Key to semantic analysis Key to multidimensional augmentation of information Key to information relationability In short: key to more powerful datamining
Types of Metadata Human authored Automatically Inferred
Human Authoring (1) In-process: data and metadata are created at the same time Out-of-process: metadata is added after data has been created
Human Authoring (2) By the data author: data and metadata are written by the same person By another author: data and metadata are created by different people
Automatic Inference Recogniction of patterns and trends in data Semantic assumption of data-metadata correlations
Types of Automatic Inference Heuristic: some algorithm performs analysis on the data set (artificial reproduction of intelligent behavior) Transparent: some mechanically extracted information is transparently associated with some metadata performed by human semantic analysis
Transparent Inference Examples Google’s PageRank Amazon’s related items NEC’s CiteSeer
Google PageRank is the system that ranks the pages found after a query against their database It works on hyperlink topology analysis Metadata is inferred from the hyperlinks contained into the page
Amazon Relation between items is inferred from the analysis of the articles bought by the other users The act of a user buying two products is assumed to be a sign of relation between the items Simply by buying, the users are collectively filling up product metadata on relations
CiteSeer Digital Library of IT papers Ranks searches on ‘citations’ topology analysis Bibliographies become the source of relevance metadata
The Issues with Metadata Quality of metadata heavily influences the quality of all search/retrieval systems
First Law of Metadata Quality Artificial intelligence is just that: artificial! So: for a system that feels smart to humans, you need human-created metadata
First Law of Metadata Quantity The more high-quality metadata, the better. But: the more human-created metadata, the more expensive the authoring process gets.
Metrics In order to estimate the value of proposed technological solutions, a metric is required Economical feasibility is one possible metric
Consequences All current markup-based semantic web solutions (RDF, topic maps, ontologies) are economically infeasible. The best semantic solutions are those based on transparent inference
Suggestions (1) Plan the impact of metadata authoring costs on technology decisions. Don’t underestimate the importance of user feeling. Think about what can be inferred transparently without requiring heuristics
Suggestions (2) Do all efforts to make instant return on the investment of metadata authoring Don’t ask too much Be smart but not smarter
Thanks!