On Tue, Nov 6, 2012 at 9:26 AM, Yasen Kiprov <[email protected]> wrote: > I'm writing a named entity recognition system for text excerpts from the > social/public domain: blogs, news, etc. I'm testing different approaches with > rules and ML and I need to evaluate annotations accuracy (in terms of f-score > against a gold corpus). My plan is to use the MASC corpus or build a custom > one but the first task is to find the right tools for evaluation.
This sounds a lot like an example we have in ClearTK (it also uses the MASC named entity data). The full cross-validation evaluation code is here: https://code.google.com/p/cleartk/source/browse/cleartk-examples/src/main/java/org/cleartk/examples/chunking/EvaluateNamedEntityChunker.java And the class that actually calculates F-score, etc. over annotations is here: https://code.google.com/p/cleartk/source/browse/cleartk-eval/src/main/java/org/cleartk/eval/AnnotationStatistics.java Steve
