The Cognitive Workbench Graph Database stores knowledge in the form of entities and relations between the entities. With this it is possible to incorporate any type of knowledge representation. For example it is possible to model the user his private life into four basic entity types: An email account has a type Place and can intuitively been seen to contain a repository of emails called “Inbox”, also of type Place which contains a specific email of type Thing. An email is naturally associated with a send-time of type Time and a sender or recipient of type Person. Given this entity modeling it is then possible to introduce untyped and types relations between these entities, such as beneficiary or agent.
The workbench acquires relations through several means:
- Obvious relations (if an email arrived at time t, there is a relation “arrived at” between t and the email entity
- Shallow co-occurrence-based relations for the exploration of potentially new types of relations
- deep linguistic analysis based relations where our NLP tools really analyzed the full text word by word and came to the conclusion that some text formulates a specific relation between two entities
Here’s is an example of how a few emails might form a very small knowledge modeling in the Cognitive WorkbenchGraph Database:
This thematic structure does not follow linguistic usage closely, and exactly how these relations are established is determined independently for each kind of user data that the Cognitive Workbench supports.