edu.stanford.nlp.ie
Interface RankedExtractor

All Known Implementing Classes:

AbstractFieldExtractor

public interface RankedExtractor

Interface for information extraction components that support merging of two Instances, using field-level confidence ratings. Such extractors should implement both FieldExtractor and RankedExtractor.

Method Summary

edu.unika.aifb.kaon.Instance

extractFields(edu.unika.aifb.kaon.Instance i, String text, Confidence c) Returns the given Instance with (potentially) additional fields extracted from the given text.

Method Detail

extractFields

public edu.unika.aifb.kaon.Instance extractFields(edu.unika.aifb.kaon.Instance i,
                                                  String text,
                                                  Confidence c)

Returns the given Instance with (potentially) additional fields extracted from the given text. This is different from the version of extractFields specified by FieldExtractor in that additionally a Confidence object is passed in. This object has parallel fields to the Instance and for each filled field in the Instance, the corresponding field in the Confidence is filled with some ranking (e.g., likelihood) of how good of a fit the extracted information is. This is generally some number between 0 (not at all confident) and 1 (completely confident).

Confidence ratings are used when two potentially compatible Instances are merged together but contain conflicting information -- if the confidence rating for field A is high in Instance X but low in Instance Y, then one might choose to use X's version of A over Y's.

For any field that is filled in the Instance, the corresponding field in the Confidence should also be filled.

Parameters:

i - the original Instance to extract into

text - the text from which to extract

c - the Confidence object corresponding to the Instance passed in, which stores numerical confidence ratings for the extracted fields

Returns:

the updated Instance