Chapter 8: Timeshifting Life 327 Various kinds of personalization and adaptation are needed when using metadata from multiple sources. For instance, we should have a method for “normalizing” the values that others provide to my own personal scale: your one star song rating is actually my 1.8 stars. • Metadata semantics vary between people Content-based retrieval (CBR) is a promising technique for search- ing content (section 5.2). In CBR, the content itself is analyzed in order to identify objects or other features. For instance, the visual character- istics of a photo can be analyzed and checked for people, objects, and weather. The identifi ed information is then stored as metadata, in textual or other form, to allow rapid machine operations. As usual, any successful content recognition will not function universally, but will be domain- or context-specifi c. • Content-based retrieval is powerful in limited domains The roadmap towards more advanced content-based retrieval tech- niques is incremental. That is, some unsophisticated analyses can be performed fi rst, say, “this photo was shot at night,” and once the methods become more advanced, the created metadata can be aug- mented with additional information. Context-awareness research will bring in several enablers. Online communities are crucial for metadata creation. Therefore, technologies that support easy creation, verifi cation, and sharing of not only content, but also metadata, are of specifi c importance. As an example, we envision a “Whole Earth” distributed catalogue, where a P2P network recognizes and annotates content. Furthermore, intelli- gent metadata sharing could be enabled by global author identifi cation and personal global metadata namespaces (RDF). It is also likely that we will have good and solid techniques for making the globally shared metadata anonymous, or verinymous, depending on the situation. • P2P indexes world’s content However, communities may produce more than just mere meta- data. Communities may be harnessed to provide entire ontologies on specifi c domains, especially once the payoff is high enough. Even today, user-contributed ontologies provide semantics for restricted 1 domains (for instance, the dmoz open directory project (ODP) had 1 http://www.dmoz.org/