Given a gold-standard corpus, one of the main advantages of this algorithm is the opportunity for learning diverse values for the three parameters, subject to a particular goal. For example, the above-mentioned assumption (i.e., diseases affect a very limited set of major organs) can be transformed into a learning task based on disease categories. We experimented with the 41 manually curated diseases, split into 13 categories dictated by the top-level terms (e.g., cardiovascular diseases, integumentary system diseases, etc.) in the Disease Ontology (DO), and aimed to maximize the category-based true-positive rate. This can be realized by learning sets of parameters corresponding to each disease category. The experimental results showed an overall resulting precision of 66.8%, including highlights such as over 70% precision for diseases by infectious agents (73.0%), diseases of the nervous system (77.8%), or immune system diseases (82.8%). Similarly, we experimented with targeting a maximized overall F-score (i.e., the harmonic mean of precision and recall—a balance between coverage and true-positive rate) and achieved a value of 45.1%. This value is equivalent to an average precision of around 60% associated with a recall of around 40%.