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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451
|
/*
* Open Chinese Convert
*
* Copyright 2015 BYVoid <byvoid@byvoid.com>
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <cmath>
#include <unordered_map>
#include "darts.h"
#include "PhraseExtract.hpp"
namespace opencc {
namespace internal {
bool ContainsPunctuation(const PhraseExtract::UTF8StringSlice8Bit& word) {
static const vector<PhraseExtract::UTF8StringSlice8Bit> punctuations = {
" ", "\n", "\r", "\t", "-", ",", ".", "?", "!", "*", " ",
",", "。", "、", ";", ":", "?", "!", "…", "“", "”", "「",
"」", "—", "-", "(", ")", "《", "》", ".", "/", "\"};
for (const auto& punctuation : punctuations) {
if (word.FindBytePosition(punctuation) !=
static_cast<PhraseExtract::UTF8StringSlice8Bit::LengthType>(-1)) {
return true;
}
}
return false;
}
} // namespace internal
class PhraseExtract::DictType {
public:
typedef PhraseExtract::Signals ValueType;
typedef std::pair<UTF8StringSlice8Bit, ValueType> ItemType;
PhraseExtract::Signals& Get(const UTF8StringSlice8Bit& key) {
Darts::DoubleArray::result_pair_type result;
daTrie.exactMatchSearch(key.CString(), result, key.ByteLength());
if (result.value != -1) {
return items[result.value].second;
} else {
throw ShouldNotBeHere();
}
}
PhraseExtract::Signals& AddKey(const UTF8StringSlice8Bit& key) {
return dict[key];
}
void Clear() {
ClearDict();
daTrie.clear();
}
const vector<ItemType>& Items() const { return items; }
void Build() {
BuildKeys();
BuildDaTrie();
}
private:
void BuildKeys() {
items.reserve(dict.size());
for (const auto& item : dict) {
items.push_back(item);
}
ClearDict();
std::sort(
items.begin(), items.end(),
[](const ItemType& a, const ItemType& b) { return a.first < b.first; });
}
void ClearDict() {
std::unordered_map<UTF8StringSlice8Bit, PhraseExtract::Signals,
UTF8StringSlice8Bit::Hasher>().swap(dict);
}
void BuildDaTrie() {
const char** keyNames = new const char* [items.size()];
size_t* keyLengths = new size_t[items.size()];
for (size_t i = 0; i < items.size(); i++) {
const auto& key = items[i].first;
keyNames[i] = key.CString();
keyLengths[i] = key.ByteLength();
}
daTrie.build(items.size(), keyNames, keyLengths);
delete[] keyNames;
delete[] keyLengths;
}
std::unordered_map<UTF8StringSlice8Bit, PhraseExtract::Signals,
UTF8StringSlice8Bit::Hasher> dict;
vector<ItemType> items;
Darts::DoubleArray daTrie;
};
using namespace internal;
bool PhraseExtract::DefaultPreCalculationFilter(
const PhraseExtract&, const PhraseExtract::UTF8StringSlice8Bit&) {
return false;
}
bool PhraseExtract::DefaultPostCalculationFilter(
const PhraseExtract& phraseExtract,
const PhraseExtract::UTF8StringSlice8Bit& word) {
const PhraseExtract::Signals& signals = phraseExtract.Signal(word);
const double logProbability = phraseExtract.LogProbability(word);
const double cohesionScore = signals.cohesion - logProbability * 0.5;
const double entropyScore =
sqrt((signals.prefixEntropy) * (signals.suffixEntropy + 1)) -
logProbability * 0.85;
bool accept = cohesionScore > 9 && entropyScore > 11 &&
signals.prefixEntropy > 0.5 && signals.suffixEntropy > 0 &&
signals.prefixEntropy + signals.suffixEntropy > 3;
return !accept;
}
PhraseExtract::PhraseExtract()
: wordMinLength(2), wordMaxLength(2), prefixSetLength(1),
suffixSetLength(1), preCalculationFilter(DefaultPreCalculationFilter),
postCalculationFilter(DefaultPostCalculationFilter), utf8FullText(""),
signals(new DictType) {
Reset();
}
PhraseExtract::~PhraseExtract() { delete signals; }
void PhraseExtract::Reset() {
prefixesExtracted = false;
suffixesExtracted = false;
frequenciesCalculated = false;
wordCandidatesExtracted = false;
cohesionsCalculated = false;
prefixEntropiesCalculated = false;
suffixEntropiesCalculated = false;
wordsSelected = false;
totalOccurrence = 0;
logTotalOccurrence = 0;
ReleasePrefixes();
ReleaseSuffixes();
wordCandidates.clear();
words.clear();
signals->Clear();
utf8FullText = UTF8StringSlice("");
preCalculationFilter = DefaultPreCalculationFilter;
postCalculationFilter = DefaultPostCalculationFilter;
}
void PhraseExtract::ExtractSuffixes() {
suffixes.reserve(utf8FullText.UTF8Length() / 2 *
(wordMaxLength + suffixSetLength));
for (UTF8StringSlice text = utf8FullText; text.UTF8Length() > 0;
text.MoveRight()) {
const LengthType suffixLength =
std::min(static_cast<LengthType>(wordMaxLength + suffixSetLength),
text.UTF8Length());
const UTF8StringSlice& slice = text.Left(suffixLength);
suffixes.push_back(UTF8StringSlice8Bit(slice.CString(),
static_cast<UTF8StringSlice8Bit::LengthType>(slice.UTF8Length()),
static_cast<UTF8StringSlice8Bit::LengthType>(slice.ByteLength())));
}
suffixes.shrink_to_fit();
// Sort suffixes
std::sort(suffixes.begin(), suffixes.end());
suffixesExtracted = true;
}
void PhraseExtract::ExtractPrefixes() {
prefixes.reserve(utf8FullText.UTF8Length() / 2 *
(wordMaxLength + prefixSetLength));
for (UTF8StringSlice text = utf8FullText; text.UTF8Length() > 0;
text.MoveLeft()) {
const LengthType prefixLength =
std::min(static_cast<LengthType>(wordMaxLength + prefixSetLength),
text.UTF8Length());
const UTF8StringSlice& slice = text.Right(prefixLength);
prefixes.push_back(UTF8StringSlice8Bit(slice.CString(),
static_cast<UTF8StringSlice8Bit::LengthType>(slice.UTF8Length()),
static_cast<UTF8StringSlice8Bit::LengthType>(slice.ByteLength())));
}
prefixes.shrink_to_fit();
// Sort suffixes reversely
std::sort(prefixes.begin(), prefixes.end(),
[](const UTF8StringSlice8Bit& a, const UTF8StringSlice8Bit& b) {
return a.ReverseCompare(b) < 0;
});
prefixesExtracted = true;
}
void PhraseExtract::CalculateFrequency() {
if (!suffixesExtracted) {
ExtractSuffixes();
}
for (const auto& suffix : suffixes) {
for (UTF8StringSlice8Bit::LengthType i = 1; i <= suffix.UTF8Length() && i <= wordMaxLength; i++) {
const UTF8StringSlice8Bit wordCandidate = suffix.Left(i);
signals->AddKey(wordCandidate).frequency++;
totalOccurrence++;
}
}
logTotalOccurrence = log(totalOccurrence);
signals->Build();
frequenciesCalculated = true;
}
void PhraseExtract::ExtractWordCandidates() {
if (!frequenciesCalculated) {
CalculateFrequency();
}
for (const auto& item : signals->Items()) {
const auto& wordCandidate = item.first;
if (wordCandidate.UTF8Length() < wordMinLength) {
continue;
}
if (ContainsPunctuation(wordCandidate)) {
continue;
}
if (preCalculationFilter(*this, wordCandidate)) {
continue;
}
wordCandidates.push_back(wordCandidate);
}
// Sort by frequency
std::sort(wordCandidates.begin(), wordCandidates.end(),
[this](const UTF8StringSlice8Bit& a, const UTF8StringSlice8Bit& b) {
const size_t freqA = Frequency(a);
const size_t freqB = Frequency(b);
if (freqA > freqB) {
return true;
} else if (freqA < freqB) {
return false;
} else {
return a < b;
}
});
wordCandidatesExtracted = true;
}
typedef std::unordered_map<PhraseExtract::UTF8StringSlice8Bit, size_t,
PhraseExtract::UTF8StringSlice8Bit::Hasher>
AdjacentSetType;
template <bool SUFFIX>
void CalculatePrefixSuffixEntropy(
const vector<PhraseExtract::UTF8StringSlice8Bit>& presuffixes,
const PhraseExtract::LengthType setLength,
const PhraseExtract::LengthType wordMinLength,
const PhraseExtract::LengthType wordMaxLength,
const std::function<void(const PhraseExtract::UTF8StringSlice8Bit& word,
AdjacentSetType& adjacentSet)>& updateEntropy) {
AdjacentSetType adjacentSet;
auto setLength8Bit = static_cast<PhraseExtract::UTF8StringSlice8Bit::LengthType>(setLength);
for (PhraseExtract::LengthType length = wordMinLength;
length <= wordMaxLength; length++) {
adjacentSet.clear();
PhraseExtract::UTF8StringSlice8Bit lastWord("");
for (const auto& presuffix : presuffixes) {
if (presuffix.UTF8Length() < length) {
continue;
}
auto length8Bit = static_cast<PhraseExtract::UTF8StringSlice8Bit::LengthType>(length);
const auto& wordCandidate =
SUFFIX ? presuffix.Left(length8Bit) : presuffix.Right(length8Bit);
if (wordCandidate != lastWord) {
updateEntropy(lastWord, adjacentSet);
lastWord = wordCandidate;
}
if (length + setLength <= presuffix.UTF8Length()) {
if (SUFFIX) {
const auto& wordSuffix = presuffix.SubString(length8Bit, setLength8Bit);
adjacentSet[wordSuffix]++;
} else {
const auto& wordPrefix = presuffix.SubString(
presuffix.UTF8Length() - length8Bit - setLength8Bit, setLength8Bit);
adjacentSet[wordPrefix]++;
}
}
}
updateEntropy(lastWord, adjacentSet);
}
}
void PhraseExtract::CalculateSuffixEntropy() {
if (!suffixesExtracted) {
ExtractSuffixes();
}
if (!frequenciesCalculated) {
CalculateFrequency();
}
CalculatePrefixSuffixEntropy<true>(
suffixes, suffixSetLength, wordMinLength, wordMaxLength,
[this](const PhraseExtract::UTF8StringSlice8Bit& word,
AdjacentSetType& adjacentSet) {
if (word.UTF8Length() > 0) {
signals->Get(word).suffixEntropy = CalculateEntropy(adjacentSet);
adjacentSet.clear();
}
});
suffixEntropiesCalculated = true;
}
void PhraseExtract::CalculatePrefixEntropy() {
if (!prefixesExtracted) {
ExtractPrefixes();
}
if (!frequenciesCalculated) {
CalculateFrequency();
}
CalculatePrefixSuffixEntropy<false>(
prefixes, prefixSetLength, wordMinLength, wordMaxLength,
[this](const PhraseExtract::UTF8StringSlice8Bit& word,
AdjacentSetType& adjacentSet) {
if (word.UTF8Length() > 0) {
signals->Get(word).prefixEntropy = CalculateEntropy(adjacentSet);
adjacentSet.clear();
}
});
prefixEntropiesCalculated = true;
}
void PhraseExtract::CalculateCohesions() {
if (!wordCandidatesExtracted) {
ExtractWordCandidates();
}
if (!frequenciesCalculated) {
CalculateFrequency();
}
for (const auto& wordCandidate : wordCandidates) {
signals->Get(wordCandidate).cohesion = CalculateCohesion(wordCandidate);
}
cohesionsCalculated = true;
}
const PhraseExtract::Signals&
PhraseExtract::Signal(const UTF8StringSlice8Bit& wordCandidate) const {
return signals->Get(wordCandidate);
}
double PhraseExtract::Cohesion(const UTF8StringSlice8Bit& word) const {
return Signal(word).cohesion;
}
double PhraseExtract::Entropy(const UTF8StringSlice8Bit& word) const {
return SuffixEntropy(word) + PrefixEntropy(word);
}
double PhraseExtract::SuffixEntropy(const UTF8StringSlice8Bit& word) const {
return Signal(word).suffixEntropy;
}
double PhraseExtract::PrefixEntropy(const UTF8StringSlice8Bit& word) const {
return Signal(word).prefixEntropy;
}
size_t PhraseExtract::Frequency(const UTF8StringSlice8Bit& word) const {
const size_t frequency = Signal(word).frequency;
return frequency;
}
double PhraseExtract::Probability(const UTF8StringSlice8Bit& word) const {
const size_t frequency = Frequency(word);
return static_cast<double>(frequency) / totalOccurrence;
}
double PhraseExtract::LogProbability(const UTF8StringSlice8Bit& word) const {
// log(frequency / totalOccurrence) = log(frequency) - log(totalOccurrence)
const size_t frequency = Frequency(word);
return log(frequency) - logTotalOccurrence;
}
double PhraseExtract::PMI(const UTF8StringSlice8Bit& wordCandidate,
const UTF8StringSlice8Bit& part1,
const UTF8StringSlice8Bit& part2) const {
// PMI(x, y) = log(P(x, y) / (P(x) * P(y)))
// = log(P(x, y)) - log(P(x)) - log(P(y))
return LogProbability(wordCandidate) - LogProbability(part1) -
LogProbability(part2);
}
double PhraseExtract::CalculateCohesion(
const UTF8StringSlice8Bit& wordCandidate) const {
// TODO Try average value
double minPMI = INFINITY;
for (UTF8StringSlice8Bit::LengthType leftLength = 1; leftLength <= wordCandidate.UTF8Length() - 1;
leftLength++) {
const auto& leftPart = wordCandidate.Left(leftLength);
const auto& rightPart =
wordCandidate.Right(wordCandidate.UTF8Length() - leftLength);
double pmi = PMI(wordCandidate, leftPart, rightPart);
minPMI = std::min(pmi, minPMI);
}
return minPMI;
}
double PhraseExtract::CalculateEntropy(const std::unordered_map<
UTF8StringSlice8Bit, size_t, UTF8StringSlice8Bit::Hasher>& choices) const {
double totalChoices = 0;
for (const auto& item : choices) {
totalChoices += item.second;
}
double entropy = 0;
for (const auto& item : choices) {
const size_t occurrence = item.second;
const double probability = occurrence / totalChoices;
entropy += probability * log(probability);
}
if (entropy != 0) {
entropy = -entropy;
}
return entropy;
}
void PhraseExtract::SelectWords() {
if (!wordCandidatesExtracted) {
ExtractWordCandidates();
}
if (!cohesionsCalculated) {
CalculateCohesions();
}
if (!prefixEntropiesCalculated) {
CalculatePrefixEntropy();
}
if (!suffixEntropiesCalculated) {
CalculateSuffixEntropy();
}
for (const auto& word : wordCandidates) {
if (!postCalculationFilter(*this, word)) {
words.push_back(word);
}
}
wordsSelected = true;
}
} // namespace opencc
|