Skip to content

Latest commit

 

History

History
103 lines (72 loc) · 2.95 KB

README.md

File metadata and controls

103 lines (72 loc) · 2.95 KB
Raw

NER Tagger

A Maximum Entropy Sequence Model for Named Entity Recognition (NER) in Scala

Currently implements a trigram Hidden Markov Model

Inference is done using the Viterbi algorithm

Build

A fully functional jar — including all dependencies — can be built using sbt-assembly.

In the project's top level directory, run:

$ sbt assembly

Usage

For a list of all available options, run the jar with --help option:

$ java -jar NERtagger.jar --help

NERtagger 1.0
Usage: NERtagger [score|save|tag] [options] <args>...

  --help                   ouptut usage information
  -c, --corpus <str>       training corpus; must be one of {wiki|gene}
  -r, --rare <int>         rare count threshold
  -t, --train              flag indicating whether or not to retrain the model
  -p, --pp <str>           Rare word preprocessor; must be one of {pattern|replace}
Command: score [options]
Scores tagger on dev/key set and outputs tagged development set to file if provided.
  -o, --out <file>         output file name (optional)
Command: save [options]
Saves model rule counts to file for quicker loading
  -o, --out <file>         output file name
Command: tag [options] [<text>]
Tag sentence using trained model; output to file if provided
  <text>                   text to tag
  -i, --input <file>       input file name
  -o, --out <file>         output file name (optional)

Example Usage

Scoring

$ java -jar NERtagger.jar score -c wiki

  Using Wikipedia Documents
  Loading tagger...
  Scoring tagger...

  Total Scores
  ------------------
  Accuracy  | 0.9627
  Precision | 0.9377
  Recall    | 0.7916
  F1-Score  | 0.8585

Tagging

$ java -jar NERtagger.jar tag 'Matthew Chamberlin lives in San Francisco , but his code lives on GitHub .'

  Using Wikipedia Documents
  Loading tagger...

  Output:
  Matthew|I-PER Chamberlin|I-PER lives|O in|O San|I-LOC Francisco|I-LOC ,|O but|O his|O code|O lives|O on|O GitHub|I-MISC .|O

For a tagger trained on any of the provided corpora to work effectively with new input text, preprocessing and tokenization should be done before hand. The tagger expects whitespace-separated tokens.

Provided Corpora

Wikipedia NER Tags

The Wikipedia Tag Corpus from the Univeristy of Syndey's Schwa Lab. Tag symbols are based on CoNLL named entity types.

Tag set:

  • I-MISC
  • I-PER
  • I-ORG
  • I-LOC
  • O

BioCreative GENE Tags

A slightly modified version of the MedTag/GENETAG corpus used in the BioCreative I challenge.

Available in it's original form from the National Center for Biotechnology Information:

ftp://ftp.ncbi.nlm.nih.gov/pub/lsmith/MedTag/

Tag set:

  • I-GENE
  • O