DefiningImageAccess/Project/RichTags
From ImageWeb
| DefiningImageAccess/Project/RichTags | |
|---|---|
| homepage:=http://www.mspace.fm/projects/richtags/}} | |
| [[has sub-project::{{{Subproject}}}]]}} | |
| [[sub-project of::{{{Supproject}}}]]}} | |
| [[start date:={{{Start date}}}]]}} | |
| [[end date:={{{End date}}}]]}} | |
| Status:=Initial Stage}} | |
| JISCProject:=True}} | |
| Image Materials:=Optional label, e.g. 3D or 2D or videos}} | |
| Focus:=Metadata Creation, Tagging}} | |
| [[Publishes::{{{Publishes}}}]]}} | |
| [[References::{{{References}}}]]}} | |
| [[Uses::{{{Uses}}}]]}} | |
| [[Creates::{{{Creates}}}]]}} | |
| Partner::University of Southampton}} | |
| Contact::Dan Smith}} | |
| Defining Image Access | DefiningImageAccess/RelatedWork |
Rich Tags
- Supporting Better Exploration of Digital Repositories with Semantic Social Tagging
- See Also: Meetings/20070209/DefiningImageAccess-ToolsAndTechnologies and mSpace
"Category search within digital repositories is poorly supported. This means that people wishing to access the assets of digital repositories are largely limited to keyword search, which means they must know what they want in order to look for it."
"The aim of Semantic Tagging is to add structure/meaning to tags by connecting keyword tags to relational structures such as folksonomies, taxonomies and/or ontologies. Associating tags with such structures means that tags acquire structured meanings ... it becomes possible to infer other, non-explicitly stated relations."
It seems to me that the idea of Rich Tagging is loosely associated with "relevance feedback". The system builds a graph of related tags -- I speculate this may be based on multiple tags applied to the same resource -- and then offers "nearby" tags as possible matches (both when applying tags, and when performing retrieval?) for an offered tag.
Then there's this: "Tagged items can easily be found from multiple publication repositories through the central taxonomy of semantic tags being used" -- which appears to be very closely related to what we want to do with data webs.
Also, they say: "Our approach to Semantic Tagging is based on our experience that a little bit of semantics goes a long way" - which is just what we say of our data web philosophy! The approach being described here seems to have strong similarities with some aspects of Alsitair Miles' "Gardens of meaning" work with SKOS. It seems that users will be asked to add some associations when the use a new tag: "... the widget seeing no other terms matching open access requests that the person add some meanings to the tag ...".
This projects appears to differ from our approach in that they envisage being a service that is accessed by the repository -- implying new versions of the repository software. This is very good for us, as it suggests that the outcome of this project may be something that can be fruitfully exploited by our data webs. I also anticipate that some of their user interface work may be something that we can employ in applications that build upon data webs (e.g. "Tag lookup will either take a single keyword or sentence and return a list of matching tag details, or will take a whole paragraph and return a list of tag matches based on the parsing of the paragraph").
They further aim to undertake usability studies, which, while being focused on semantic tagging, may have some bearing on future directions of our work to develop user interfaces for data webs.
Further review
Since the original comments were posted, we have found more details about Rich Tags on the web (or maybe have come to understand it better). These further notes are prompted by a new review of the project overview, and discussions with various people.
Rich tags appear to be a very promising approach for augmenting metadata from repositories with user-supplied information about content. It is developing a service, initially for use with ePrints, that will allow users to easily associate tags with dynamically created "micro ontologies", providing additional information that enhances the potential for tags to help researchers find relevant information.
| [As used with current applications] tags are used mainly to show some sense of levels of interest in a topic within a community, since they are not particularly effective for exploring back into the Tag Cloud to look for interesting content associated with that tag. |
Effectively, Rich tags will associate some additional information with tags that can be used as a basis for content filtering and navigation:
| Our approach to Semantic Tagging is based on our experience that a little bit of semantics goes a long way. ... the widget seeing no other terms matching open access requests that the person add some meanings to the tag ... The next time someone wishes to tag a document with Open Source, they can choose which of those meanings is most apt for what they propose, add a new meaning, or refine an existing meaning. |
(There are echos here of a project reported at the NCeSS ontologies meeting in Manchester -- see Meetings/20070323/NCeSS-Ontogies, segment by Pete Edwards on work relating to PolicyGrid. Here, they use a very simple core ontology and attach tags to particular concepts within that ontology using an innovative form-free user interface. Again, we see: A little semantics ... a long way ... tags filling acceptance/participation gap.)
Digging in, an attractive approach of the Rich Tags architecture is its sepaation of the Rich Tags service from the repository and user interface embedding. This means that, if the service interface is well-defined and stable, the repository/user interfaces and the service can be evolved independently. It would be very interesting to see if we could combine the rich tag service with our proposed schema and coreference services, and the technical approach should make such experimentation very easy. The overview mentions using a SOAP interface, but I would prefer a simple REST style of interface for ease of expoloratory developments. (As described, I think their AJAX approach may fall foul of browser Javascript security models, but I'm sure there are ways round that.)
I do wonder if their approach of "tagging the tags" (see quotation above) may fall foul of vocabulary divergence that the system aims to address for primary tags. Can the system be applied recursively? Can the use of a seed ontology help to contain such divergence?
It is not clear to me if the tagging service will maintain provenance information. This seems likely, as a project goal is to explore the role of trust in take-up and use of social tagging for ePrints exploration.
In summary, I propose that incorporation of Rich Tags ideas and services into an image web development, preferably in collaboration with the team at Southampton, would be a very desirable line of investigation. There are two particular reasons for this: (a) the opaque nature of images, and the importance of associated information, makes the accummulation of metadata, and (b) the fact that is emerging from our survey work that domain specific metadata is not common in institutional repositories.
Another area of investigation to consider would be the extent to which a single tagging service can effectively support multiple data webs with differing domain focus. A common service would have the great advantage of enabling a higher degree of "crowdsourcing", but will it also lead to weakening of usable semantics when notions of schema matching or coreference may be somewhat domain dependent. Also, there's a concern of over-centralization; will it be feasible to federate rich tag services so that there is not a dependency on a single central service?
