001/** 002 * Copyright (c) 2011, The University of Southampton and the individual contributors. 003 * All rights reserved. 004 * 005 * Redistribution and use in source and binary forms, with or without modification, 006 * are permitted provided that the following conditions are met: 007 * 008 * * Redistributions of source code must retain the above copyright notice, 009 * this list of conditions and the following disclaimer. 010 * 011 * * Redistributions in binary form must reproduce the above copyright notice, 012 * this list of conditions and the following disclaimer in the documentation 013 * and/or other materials provided with the distribution. 014 * 015 * * Neither the name of the University of Southampton nor the names of its 016 * contributors may be used to endorse or promote products derived from this 017 * software without specific prior written permission. 018 * 019 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND 020 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 021 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 022 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR 023 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 024 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 025 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON 026 * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 027 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 028 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 029 */ 030package org.openimaj.text.nlp.sentiment.lexicon; 031 032import java.util.ArrayList; 033import java.util.HashSet; 034import java.util.Iterator; 035import java.util.LinkedList; 036import java.util.List; 037import java.util.Set; 038 039import org.arabidopsis.ahocorasick.AhoCorasick; 040import org.arabidopsis.ahocorasick.SearchResult; 041import org.openimaj.text.nlp.textpipe.annotations.AnnotationUtils; 042import org.openimaj.text.nlp.textpipe.annotations.RawTextAnnotation; 043import org.openimaj.text.nlp.textpipe.annotations.TokenAnnotation; 044import org.openimaj.text.nlp.textpipe.annotators.MissingRequiredAnnotationException; 045import org.openimaj.text.nlp.textpipe.annotators.OpenNLPTokenAnnotator; 046 047/** 048 * An implementation of Hatzivassiloglou and McKeown's approach to a 049 * semisupervised method of building a bipolar sentiment lexicon. This is a one 050 * pass version, in that the corpus to build from is fixed. 051 * 052 */ 053public class HMLexiconBuilder { 054 055 Set<String> positiveLexicon; 056 Set<String> negativeLexicon; 057 List<String> newPos; 058 List<String> newNeg; 059 List<String> corpus; 060 OpenNLPTokenAnnotator tokA; 061 062 public HMLexiconBuilder(List<String> posBootStrap, List<String> negBootStrap) { 063 this.positiveLexicon = new HashSet<String>(); 064 this.negativeLexicon = new HashSet<String>(); 065 this.newPos = new LinkedList<String>(); 066 this.newNeg = new LinkedList<String>(); 067 this.tokA = new OpenNLPTokenAnnotator(); 068 for (String s : posBootStrap) { 069 addToLexicon(positiveLexicon, newPos, s); 070 } 071 for (String s : negBootStrap) { 072 addToLexicon(negativeLexicon, newNeg, s); 073 } 074 } 075 076 private void addToLexicon(Set<String> compSet, List<String> q, String token) { 077 if (compSet.add(token)) 078 q.add(token); 079 } 080 081 public void buildFromCorpus(List<String> corpus) { 082 this.corpus = corpus; 083 process(); 084 } 085 086 private void process() { 087 while (!newPos.isEmpty()) 088 processNewLexTokens(positiveLexicon, newPos, negativeLexicon, 089 newNeg); 090 while (!newNeg.isEmpty()) 091 processNewLexTokens(negativeLexicon, newNeg, positiveLexicon, 092 newPos); 093 // Make sure that they have not added to each other after processing. 094 if (!(newPos.isEmpty() || !newNeg.isEmpty())) 095 process(); 096 } 097 098 private void processNewLexTokens(Set<String> lexicon, List<String> q, 099 Set<String> anti_lexicon, List<String> anti_q) { 100 AhoCorasick<String> tri; 101 tri = new AhoCorasick<String>(); 102 for (String string : q) { 103 String syno = string + " and"; 104 String anti = string + " but"; 105 tri.add(syno.getBytes(), syno); 106 tri.add(anti.getBytes(), anti); 107 } 108 tri.prepare(); 109 q.clear(); 110 for (String doc : corpus) { 111 String lcdoc = doc.toLowerCase(); 112 Iterator<SearchResult<String>> result = tri.search(lcdoc.getBytes()); 113 List<String> hits = new ArrayList<String>(); 114 while (result.hasNext()) { 115 SearchResult<String> sr = result.next(); 116 for (String s : sr.getOutputs()) { 117 hits.add(s); 118 } 119 } 120 for (String hit : hits) { 121 int tokeniseFrom = lcdoc.indexOf(hit) + hit.length(); 122 List<String> tokens = tokenise(lcdoc.substring(tokeniseFrom)); 123 Iterator<String> it = tokens.iterator(); 124 String newLex=null; 125 boolean anti =false; 126 if(it.hasNext()){ 127 String first = it.next(); 128 if(first.equals("not")){ 129 anti=true; 130 if(it.hasNext()){ 131 newLex=it.next(); 132 } 133 } 134 else{ 135 newLex = first; 136 } 137 } 138 139 if(hit.endsWith("but"))anti=!anti; 140 if (newLex!=null) { 141 if (!anti)addToLexicon(lexicon, q, newLex); 142 else addToLexicon(anti_lexicon, anti_q, newLex); 143 } 144 } 145 } 146 } 147 148 private List<String> tokenise(String text) { 149 RawTextAnnotation rta = new RawTextAnnotation(text); 150 try { 151 tokA.annotate(rta); 152 return AnnotationUtils.getStringTokensFromTokenAnnotationList(rta.getAnnotationsFor(TokenAnnotation.class)); 153 } catch (MissingRequiredAnnotationException e) { 154 e.printStackTrace(); 155 } 156 return null; 157 } 158 159}