Query Translation - The Knowledge Base

Pattern Matching: Looks at all data models and attempts to find a best approximation of a universal data model.

I like to think of this as being similar to social tagging. There is not one specified tag for a certain article, book, etc. You have many tags associated with the item but there are ones that stand out more than the others. This is indicated by the size and the boldness of the text.

Similar to social tagging. Some mappings are stronger than others, but the KB should be able to keep track of this for us.

The approximation of the universal data model should be open to input by all medical professionals. As with social tagging, rating and ranking medical professionals could log into a system (AMA keeps track of physicians and there may be other organizations that identify medical professionals) and give their input. The input is then applied to the KB and the more agreement there is on a particular mapping, the stronger the mapping becomes.

There should be a notion of a threshold with the data model. If you want a high specificity for a query (clear and precise definitions), then the KB will take note of this and look for areas of high specificity in the underlying databases. However, with high specificity there will be a large amount of data lost in the retrieval process.

If you want a low specificity for a query (more generalized definitions), then the KB will take not of this and look for areas that are generally related. However, the low specificity will produce ambiguous results with lots of errors.

Therefore, there needs to be a "sweet spot" in the range of specificity (or approximations of the global data model) that will produce as much relevant information as possible with minimal errors and ambiguity.