How to Model Probabilistic Ontologies

In this section we will show how to model a probabilistic ontologies (Matsumoto and Santos 2008). We have summarized this process in 11 steps, as follows:

  1. Identify and define the scope of the problem to be solved;
  2. Identify the entities present in the domain. If you are having trouble trying to come up with different entities, you probably do not need to use MEBN. Try using just BN, instead;
  3. Identify logical groups (there are groups of information that are or could be logically put together) for identifying sets of entities;
  4. Identify criterias that can classify in some way the identified entities. This will help you choose which entities are relevant and which are not to solve the problem (discard them from your model). This step can also help to detect uncertain about the existence of information and to identify contexts where certain informations can be considered valid;
  5. Identify the attributes the entities can have;
  6. If any attribute identified in the previous step is continuous, decide how the discretization must be done. In the case where the values are discrete, but there are too many values, try to group them some how;
  7. Identify rules related to entities present in the same group and rules related to entities in different groups;
  8. Evaluate if MEBN is necessary and/or sufficient for modeling the problem;
  9. Evaluate if the identified entities are enough for your MEBN model;
  10. Map the entities, groups, rules, and relations identified to their respective MEBN element (MFrag, Node, etc);
  11. Design the model in UnBBayes. Here, you might need to change your model a little bit, because UnBBayes has some singularities due to implementation (the way it implements recursion, possible states for a resident node, etc).