These are used to specify the linguistic processing for sense analysis.
The recipes specify the following:
- The sense resolution strategies that are employed;
- The number of times each strategy is employed.
As one might expect, the run time of recipes vary greatly. Some recipes run for considerably longer than others and achieve slight precision increases (less ambiguity, more accuracy). This trade-off must be considered for each application.
In the table below, the relative values are provided against the “highestPrecision” recipe which runs the slowest and provides the most accurate results.
|Recipe||Description||Relative Processing Rate||Charged Units per Token||Relative Precision|
|highestPrecision||Best possible results. Suitable for sense tagging applications.||1.0||3.0||1.0||1.0|
|defaultPrecision||Compromise between the precision of the results and the computing cost suitable for most applications.||250%||1.0||-2.0%||-0.6%|
|quickResults||Sacrifices some accuracy for speed. Suitable for applications where only the best results are taken.||350%||0.75||-3.6%||-2.4%|
|ner||This recipe only annotates positions that are most probably a named entity.||350%||0.50||not applicable||not available|
The accumulated service usage shown at My Projects takes into account the relative processing rates of the recipes. The last column of the table indicates how many units are charged per token processed. See Application Projects for more information.