Automatic Generation of (LOM) Metadata
I’ve been advocating for some time that “electronic forms must die”, meaning that metadata can be generated without asking end users to fill in bothersome electronic forms. On the ARIADNE Wiki, we now provide some information on how we can make this happen through our “Automatic Indexing Framework”. Basically, we have developed a software framework for automatic metadata generation. As you would expect, the framework can generate metadata from the content of the learning object: as an example, a simple analysis of a limited amount of textual information suffices to determine the language of the learning object with very reasonable accuracy. More interestingly, we can also generate metadata from the context that the learning object is deployed in. At the moment, we have two interesting such context based indexers:
- a building block for Blackboard that enables us to capture metadata from the Blackboard “courses” that a learning object is deployed in, and
- a component for the MIT Open CourseWare initiative, that mines the (quite consistent) web site.
A generic negotiation framework reconciles clashes between indexers, based on a “confidence value” that models how reliable certain inferences are. Most importantly, this is an open framework, so we welcome additional components, and we will maintain this work as an open source effort: sourceforge currently hosts the code.