1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
|
package com.wcohen.ss;
import java.util.ArrayList;
import java.util.Collections;
import java.util.Iterator;
import java.util.List;
import com.wcohen.ss.api.StringDistance;
import com.wcohen.ss.api.StringWrapper;
import com.wcohen.ss.api.Token;
import com.wcohen.ss.api.Tokenizer;
/**
* TFIDF-based distance metric, extended to use "soft" token-matching. Specifically, tokens are considered a partial
* match if they get a good score using an inner string comparator.
*
* <p>
* On the WHIRL datasets, thresholding JaroWinkler at 0.9 or 0.95 seems to be about right.
*/
public class SoftTFIDF extends TFIDF {
// distance to use to compare tokens
private StringDistance tokenDistance;
// threshold beyond which tokens are considered a match
private double tokenMatchThreshold;
// default token distance
private static final StringDistance DEFAULT_TOKEN_DISTANCE = new JaroWinkler();
public SoftTFIDF(Tokenizer tokenizer, StringDistance tokenDistance, double tokenMatchThreshold) {
super(tokenizer);
this.tokenDistance = tokenDistance;
this.tokenMatchThreshold = tokenMatchThreshold;
}
public SoftTFIDF(StringDistance tokenDistance, double tokenMatchThreshold) {
super();
this.tokenDistance = tokenDistance;
this.tokenMatchThreshold = tokenMatchThreshold;
}
public SoftTFIDF(StringDistance tokenDistance) {
this(tokenDistance, 0.9);
}
public SoftTFIDF() {
this(new JaroWinkler(), 0.9);
}
public void setTokenMatchThreshold(double d) {
tokenMatchThreshold = d;
}
public void setTokenMatchThreshold(Double d) {
tokenMatchThreshold = d.doubleValue();
}
public double getTokenMatchThreshold() {
return tokenMatchThreshold;
}
private class Similarity implements Comparable<Similarity> {
int r1;
int r2;
double sim;
public Similarity(int r1, int r2, double sim) {
this.r1 = r1;
this.r2 = r2;
this.sim = sim;
}
public int compareTo(Similarity o) {
if (sim > o.sim)
return 1;
else if (sim < o.sim)
return -1;
return 0;
}
}
public double score(StringWrapper s, StringWrapper t) {
checkTrainingHasHappened(s, t);
UnitVector sBag = asUnitVector(s);
UnitVector tBag = asUnitVector(t);
List<Similarity> similarities = new ArrayList<Similarity>(sBag.size());
double sim = 0.0;
int i = 0;
for (Iterator<Token> ti = sBag.tokenIterator(); ti.hasNext(); i++) {
Token tok = ti.next();
int j = 0;
for (Iterator<Token> tj = tBag.tokenIterator(); tj.hasNext(); j++) {
Token tokJ = tj.next();
double distItoJ = tokenDistance.score(tok.getValue(), tokJ.getValue());
if (distItoJ >= tokenMatchThreshold) {
similarities.add(new Similarity(i, j, distItoJ * sBag.getWeight(tok) * tBag.getWeight(tokJ)));
}
}
}
/*
* This could be O(sBag.size() * tBag.size() * log (sBag.size() * tBag.size())) in the worst case but usually
* the threshold will make it much better and likely less than O(sBag.size() * tBag.size()) which is the current
* complexity.
*/
Collections.sort(similarities, Collections.reverseOrder());
boolean[] sUsed = new boolean[sBag.size()];
boolean[] tUsed = new boolean[tBag.size()];
// enforce that each word is only used for one similarity, to make sure normalization works
for (int k = 0; k < similarities.size(); k++) {
Similarity similarity = similarities.get(k);
if (sUsed[similarity.r1] || tUsed[similarity.r2])
continue;
sim += similarity.sim;
sUsed[similarity.r1] = true;
tUsed[similarity.r2] = true;
}
return sim;
}
/**
* Explain how the distance was computed. In the output, the tokens in S and T are listed, and the common tokens are
* marked with an asterisk.
*/
public String explainScore(StringWrapper s, StringWrapper t) {
BagOfTokens sBag = (BagOfTokens) s;
BagOfTokens tBag = (BagOfTokens) t;
StringBuilder buf = new StringBuilder("");
PrintfFormat fmt = new PrintfFormat("%.3f");
buf.append("Common tokens: ");
for (Iterator<Token> i = sBag.tokenIterator(); i.hasNext();) {
Token tok = i.next();
if (tBag.contains(tok)) {
buf.append(" " + tok.getValue() + ": ");
buf.append(fmt.sprintf(sBag.getWeight(tok)));
buf.append("*");
buf.append(fmt.sprintf(tBag.getWeight(tok)));
} else {
// find best matching token
double matchScore = tokenMatchThreshold;
Token matchTok = null;
for (Iterator<Token> j = tBag.tokenIterator(); j.hasNext();) {
Token tokJ = j.next();
double distItoJ = tokenDistance.score(tok.getValue(), tokJ.getValue());
if (distItoJ >= matchScore) {
matchTok = tokJ;
matchScore = distItoJ;
}
}
if (matchTok != null) {
buf.append(" '" + tok.getValue() + "'~='" + matchTok.getValue() + "': ");
buf.append(fmt.sprintf(sBag.getWeight(tok)));
buf.append("*");
buf.append(fmt.sprintf(tBag.getWeight(matchTok)));
buf.append("*");
buf.append(fmt.sprintf(matchScore));
}
}
}
buf.append("\nscore = " + score(s, t));
return buf.toString();
}
public String toString() {
return "[SoftTFIDF thresh=" + tokenMatchThreshold + ";" + tokenDistance + "]";
}
}
|
