The following are representative samples obtained from Sense Analysis.
The documents are random documents from Wikipedia (i.e., obtained using “Random article”). The Internet Search queries were selected randomly from a collection of search queries. Both sets were processed using the “highestPrecision” recipe.
Several confidence values are attached to each result as described in Understanding the Confidence Thresholds. In the following results, each sense is classified as “usable”, “doubtful”, or “unreliable” using the following logic:
- ccfmp = 0, or (ccfmp < 0.3 and fine sense positive confidence — pc < 0.8) are unreliable;
- (0.3 < ccfmp < 0.5) and fine sense pc < 0.60 are doubtful;
- others are reliable;
These are only provided as an example filtering scenario. Applications may use stricter or looser criteria depending on their objectives.