conllner2003sentence2tokensequence.java

常用机器学习算法,java编写源代码,内含常用分类算法,包括说明文档

/* Copyright (C) 2002 Univ. of Massachusetts Amherst, Computer Science Dept.
   This file is part of "MALLET" (MAchine Learning for LanguagE Toolkit).
   http://www.cs.umass.edu/~mccallum/mallet
   This software is provided under the terms of the Common Public License,
   version 1.0, as published by http://www.opensource.org.  For further
   information, see the file `LICENSE' included with this distribution. */


/** 
   @author Andrew McCallum <a href="mailto:[e-mail]">[e-mail]</a>
 */

/*
	An error?  CoNLLTrue MalletTrue MalletPred
	O O O
	I-MISC B-MISC B-MISC
	B-MISC B-MISC I-MISC
	I-MISC B-MISC I-MISC
	O O O
	O O O
	O O O
*/

package edu.umass.cs.mallet.share.mccallum.ner;

import edu.umass.cs.mallet.base.pipe.*;
import edu.umass.cs.mallet.base.types.*;
import java.util.regex.*;

public class ConllNer2003Sentence2TokenSequence extends Pipe
{
	static final String[] endings = new String[]
	{"ing", "ed", "ogy", "s", "ly", "ion", "tion", "ity", "ies"};
	static Pattern[] endingPatterns = new Pattern[endings.length];
	// Indexed by {forward,backward} {0,1,2 offset} {ending char ngram index}
	static final String[][][] endingNames = new String[2][3][endings.length];

	{
		for (int i = 0; i < endings.length; i++) {
			endingPatterns[i] = Pattern.compile (".*"+endings[i]+"$");
			for (int j = 0; j < 3; j++) {
				for (int k = 0; k < 2; k++)
					endingNames[k][j][i] = "W"+(k==1?"-":"")+j+"=<END"+endings[i]+">";
			}
		}
	}

	boolean saveSource = false;
	boolean doConjunctions = false;
	boolean doTags = true;
	boolean doPhrases = true;
	boolean doSpelling = false;
	boolean doDigitCollapses = true;
	boolean doDowncasing = false;
	
	public ConllNer2003Sentence2TokenSequence ()
	{
		super (null, LabelAlphabet.class);
	}

	public ConllNer2003Sentence2TokenSequence (boolean extraFeatures)
	{
		super (null, LabelAlphabet.class);
		if (!extraFeatures) {
			doDigitCollapses = doConjunctions = doSpelling = doPhrases = doTags = false;
			doDowncasing = true;
		}
	}
	
	/* Lines look like this:
		 -DOCSTART- -X- -X- O

		 EU NNP I-NP I-ORG
		 rejects VBZ I-VP O
		 German JJ I-NP I-MISC
		 call NN I-NP O
		 to TO I-VP O
		 boycott VB I-VP O
		 British JJ I-NP I-MISC
		 lamb NN I-NP O
		 . . O O

		 Peter NNP I-NP I-PER
		 Blackburn NNP I-NP I-PER

		 BRUSSELS NNP I-NP I-LOC
		 1996-08-22 CD I-NP O

		 The DT I-NP O
		 European NNP I-NP I-ORG
		 Commission NNP I-NP I-ORG
		 said VBD I-VP O
		 on IN I-PP O
		 ...
	*/

	public Instance pipe (Instance carrier)
	{
		String sentenceLines = (String) carrier.getData();
		String[] tokens = sentenceLines.split ("\n");
		TokenSequence data = new TokenSequence (tokens.length);
		LabelSequence target = new LabelSequence ((LabelAlphabet)getTargetAlphabet(), tokens.length);
		boolean [][] ending = new boolean[3][endings.length];
		boolean [][] endingp1 = new boolean[3][endings.length];
		boolean [][] endingp2 = new boolean[3][endings.length];
		StringBuffer source = saveSource ? new StringBuffer() : null;

		String prevLabel = "NOLABEL";
		Pattern ipattern = Pattern.compile ("I-.*");
		String word, tag, phrase, label;
		for (int i = 0; i < tokens.length; i++) {
			if (tokens[i].length() != 0) {
				String[] features = tokens[i].split (" ");
				if (features.length != 4)
					throw new IllegalStateException ("Line \""+tokens[i]+"\" doesn't have four elements");
				word = features[0]; // .toLowerCase();
				tag = features[1];
				phrase = features[2];
				label = features[3];
			} else {
				word = "-<S>-";
				tag = "-<S>-";
				phrase = "-<S>-";
				label = "O";
			}

			// Transformations
			if (doDigitCollapses) {
				if (word.matches ("19\\d\\d"))
					word = "<YEAR>";
				else if (word.matches ("19\\d\\ds"))
					word = "<YEARDECADE>";
				else if (word.matches ("19\\d\\d-\\d+"))
					word = "<YEARSPAN>";
				else if (word.matches ("\\d+\\\\/\\d"))
					word = "<FRACTION>";
				else if (word.matches ("\\d[\\d,\\.]*"))
					word = "<DIGITS>";
				else if (word.matches ("19\\d\\d-\\d\\d-\\d--d"))
					word = "<DATELINEDATE>";
				else if (word.matches ("19\\d\\d-\\d\\d-\\d\\d"))
					word = "<DATELINEDATE>";
				else if (word.matches (".*-led"))
					word = "<LED>";
				else if (word.matches (".*-sponsored"))
					word = "<LED>";
			}

			if (doDowncasing)
				word = word.toLowerCase();
			Token token = new Token (word);
			
			// Word and tag unigram at current time
			if (doSpelling) {
				for (int j = 0; j < endings.length; j++) {
					ending[2][j] = ending[1][j];
					ending[1][j] = ending[0][j];
					ending[0][j] = endingPatterns[j].matcher(word).matches();
					if (ending[0][j]) token.setFeatureValue (endingNames[0][0][j], 1);
				}
			}

			if (doTags) {
				token.setFeatureValue ("T="+tag, 1);
			}

			if (doPhrases) {
				token.setFeatureValue ("P="+phrase, 1);
			}

			if (true) {
				// Change so each segment always begins with a "B-",
				// even if previous token did not have this label.
				String oldLabel = label;
				if (ipattern.matcher(label).matches ()
						&& (prevLabel.length() < 3		// prevLabel is "O"
								|| !prevLabel.substring(2).equals (label.substring(2)))) {
					label = "B" + oldLabel.substring(1);
				}
				prevLabel = oldLabel;
			}

			// Append
			data.add (token);
			//target.add (bigramLabel);
			target.add (label);
			//System.out.print (label + ' ');
			if (saveSource) {
				source.append (word); source.append (" ");
				//source.append (bigramLabel); source.append ("\n");
				source.append (label); source.append ("\n");
			}

		}
		//System.out.println ("");
		carrier.setData(data);
		carrier.setTarget(target);
		if (saveSource)
			carrier.setSource(source);
		return carrier;
	}
}