Text field candidate evaluation works like classification evaluation.
For example, you have a given text field, a given document, and Brainware extraction as an evaluation method that returns weights for each candidate.
Candidate evaluation uses two values defined at the field level to decide whether there is a valid candidate.
- A minimum weight called threshold (by default 50 percent)
- A minimum difference in weight between the best candidate and the second-best competitor called distance (by default 10 percent.)
Both requirements must be true to extract a valid candidate.
Example
Weight of Candidate A | Weight of Candidate B | Weight of Candidate C | |
---|---|---|---|
Field A | 0.45 | 0.36 | 0.28 |
Field B | 0.30 | 1.00 | 0.00 |
Field C | 0.55 | 0.49 | n/a |
Result
- For field A, there is no valid candidate. The weight of candidate A is below the threshold, and the distance to candidate B is below the minimum distance.
- For field B, candidate B is a valid candidate. Its weight is above the threshold. In addition, the distance to the second-best competitor, Candidate A is above the minimum distance.
- For field C, there is no valid candidate. The weight of candidate A is above the threshold, but the distance to candidate B is below the minimum distance.