lib-php/tokenizer/icu_tokenizer.php

   1 <?php
   2
   3 namespace Nominatim;
   4
   5 require_once(CONST_LibDir.'/SimpleWordList.php');
   6
   7 class Tokenizer
   8 {
   9     private $oDB;
  10
  11     private $oNormalizer;
  12     private $oTransliterator;
  13
  14     public function __construct(&$oDB)
  15     {
  16         $this->oDB =& $oDB;
  17         $this->oNormalizer = \Transliterator::createFromRules(CONST_Term_Normalization_Rules);
  18         $this->oTransliterator = \Transliterator::createFromRules(CONST_Transliteration);
  19     }
  20
  21     public function checkStatus()
  22     {
  23         $sSQL = 'SELECT word_id FROM word WHERE word_id is not null limit 1';
  24         $iWordID = $this->oDB->getOne($sSQL);
  25         if ($iWordID === false) {
  26             throw new \Exception('Query failed', 703);
  27         }
  28         if (!$iWordID) {
  29             throw new \Exception('No value', 704);
  30         }
  31     }
  32
  33
  34     public function normalizeString($sTerm)
  35     {
  36         if ($this->oNormalizer === null) {
  37             return $sTerm;
  38         }
  39
  40         return $this->oNormalizer->transliterate($sTerm);
  41     }
  42
  43
  44     public function mostFrequentWords($iNum)
  45     {
  46         $sSQL = "SELECT word FROM word WHERE type = 'W'";
  47         $sSQL .= "ORDER BY info->'count' DESC LIMIT ".$iNum;
  48         return $this->oDB->getCol($sSQL);
  49     }
  50
  51
  52     private function makeStandardWord($sTerm)
  53     {
  54         return trim($this->oTransliterator->transliterate(' '.$sTerm.' '));
  55     }
  56
  57
  58     public function tokensForSpecialTerm($sTerm)
  59     {
  60         $aResults = array();
  61
  62         $sSQL = "SELECT word_id, info->>'class' as class, info->>'type' as type ";
  63         $sSQL .= '   FROM word WHERE word_token = :term and type = \'S\'';
  64
  65         Debug::printVar('Term', $sTerm);
  66         Debug::printSQL($sSQL);
  67         $aSearchWords = $this->oDB->getAll($sSQL, array(':term' => $this->makeStandardWord($sTerm)));
  68
  69         Debug::printVar('Results', $aSearchWords);
  70
  71         foreach ($aSearchWords as $aSearchTerm) {
  72             $aResults[] = new \Nominatim\Token\SpecialTerm(
  73                 $aSearchTerm['word_id'],
  74                 $aSearchTerm['class'],
  75                 $aSearchTerm['type'],
  76                 \Nominatim\Operator::TYPE
  77             );
  78         }
  79
  80         Debug::printVar('Special term tokens', $aResults);
  81
  82         return $aResults;
  83     }
  84
  85
  86     public function extractTokensFromPhrases(&$aPhrases)
  87     {
  88         $sNormQuery = '';
  89         $aWordLists = array();
  90         $aTokens = array();
  91         foreach ($aPhrases as $iPhrase => $oPhrase) {
  92             $sNormPhrase = $this->normalizeString($oPhrase->getPhrase());
  93             Debug::printVar('Phrase', $sNormPhrase);
  94
  95             $oWordList = new SimpleWordList($sNormPhrase);
  96
  97             foreach ($oWordList->getTokens() as $sToken) {
  98                 $sTransToken = $this->makeStandardWord($sToken);
  99                 if (!isset($aTokens[$sTransToken])) {
 100                     $aTokens[$sTransToken] = array();
 101                 }
 102                 $aTokens[$sTransToken][$sToken] = $sToken;
 103             }
 104
 105             $aWordLists[] = $oWordList;
 106         }
 107
 108         Debug::printVar('Tokens', $aTokens);
 109         Debug::printVar('WordLists', $aWordLists);
 110
 111         $oValidTokens = $this->computeValidTokens($aTokens);
 112
 113         foreach ($aPhrases as $iPhrase => $oPhrase) {
 114             $oPhrase->setWordSets($aWordLists[$iPhrase]->getWordSets($oValidTokens));
 115         }
 116
 117         return $oValidTokens;
 118     }
 119
 120
 121     private function computeValidTokens($aTokens)
 122     {
 123         $oValidTokens = new TokenList();
 124
 125         if (!empty($aTokens)) {
 126             $this->addTokensFromDB($oValidTokens, $aTokens);
 127
 128             // Try more interpretations for Tokens that could not be matched.
 129             foreach ($aTokens as $sToken => $aNormalized) {
 130                 if (!$oValidTokens->contains($sToken)) {
 131                     if (preg_match('/^([0-9]{5}) [0-9]{4}$/', $sToken, $aData)) {
 132                         // US ZIP+4 codes - merge in the 5-digit ZIP code
 133                         $oValidTokens->addToken(
 134                             $sToken,
 135                             new Token\Postcode(null, $aData[1], 'us')
 136                         );
 137                     } elseif (preg_match('/^[0-9]+$/', $sToken)) {
 138                         // Unknown single word token with a number.
 139                         // Assume it is a house number.
 140                         $oValidTokens->addToken(
 141                             $sToken,
 142                             new Token\HouseNumber(null, trim($sToken))
 143                         );
 144                     }
 145                 }
 146             }
 147         }
 148
 149         return $oValidTokens;
 150     }
 151
 152
 153     private function addTokensFromDB(&$oValidTokens, $aTokens)
 154     {
 155         // Check which tokens we have, get the ID numbers
 156         $sSQL = 'SELECT word_id, word_token, type, word,';
 157         $sSQL .= "      info->>'op' as operator,";
 158         $sSQL .= "      info->>'class' as class, info->>'type' as ctype,";
 159         $sSQL .= "      info->>'count' as count";
 160         $sSQL .= ' FROM word WHERE word_token in (';
 161         $sSQL .= join(',', $this->oDB->getDBQuotedList(array_keys($aTokens))).')';
 162
 163         Debug::printSQL($sSQL);
 164
 165         $aDBWords = $this->oDB->getAll($sSQL, null, 'Could not get word tokens.');
 166
 167         foreach ($aDBWords as $aWord) {
 168             $iId = (int) $aWord['word_id'];
 169             $sTok = $aWord['word_token'];
 170             $aNorms = $aTokens[$sTok];
 171
 172             switch ($aWord['type']) {
 173                 case 'C':  // country name tokens
 174                     if ($aWord['word'] !== null) {
 175                         foreach ($aNorms as $sNorm) {
 176                             $oValidTokens->addToken(
 177                                 $sNorm,
 178                                 new Token\Country($iId, $aWord['word'])
 179                             );
 180                         }
 181                     }
 182                     break;
 183                 case 'H':  // house number tokens
 184                     foreach ($aNorms as $sNorm) {
 185                         $oValidTokens->addToken($sNorm, new Token\HouseNumber($iId, $aWord['word_token']));
 186                     }
 187                     break;
 188                 case 'P':  // postcode tokens
 189                     // Postcodes are not normalized, so they may have content
 190                     // that makes SQL injection possible. Reject postcodes
 191                     // that would need special escaping.
 192                     if ($aWord['word'] !== null
 193                         && pg_escape_string($aWord['word']) == $aWord['word']
 194                     ) {
 195                         $sNormPostcode = $this->normalizeString($aWord['word']);
 196                         foreach ($aNorms as $sNorm) {
 197                             if ($sNormPostcode == $sNorm) {
 198                                 $oValidTokens->addToken(
 199                                     $sNorm,
 200                                     new Token\Postcode($iId, $aWord['word'], null)
 201                                 );
 202                             }
 203                         }
 204                     }
 205                     break;
 206                 case 'S':  // tokens for classification terms (special phrases)
 207                     if ($aWord['class'] !== null && $aWord['ctype'] !== null) {
 208                         foreach ($aNorms as $sNorm) {
 209                             if ($aWord['word'] == $sNorm) {
 210                                 $oValidTokens->addToken($sTok, new Token\SpecialTerm(
 211                                     $iId,
 212                                     $aWord['class'],
 213                                     $aWord['ctype'],
 214                                     (isset($aWord['operator'])) ? Operator::NEAR : Operator::NONE
 215                                 ));
 216                             }
 217                         }
 218                     }
 219                     break;
 220                 case 'W': // full-word tokens
 221                     foreach ($aNorms as $sNorm) {
 222                         $oValidTokens->addToken($sNorm, new Token\Word(
 223                             $iId,
 224                             (int) $aWord['count'],
 225                             substr_count($aWord['word_token'], ' ') + 1,
 226                             levenshtein($aWord['word'], $sNorm) + 1
 227                         ));
 228                     }
 229                     break;
 230                 case 'w':  // partial word terms
 231                     foreach ($aNorms as $sNorm) {
 232                         $oValidTokens->addToken($sNorm, new Token\Partial(
 233                             $iId,
 234                             $aWord['word_token'],
 235                             (int) $aWord['count']
 236                         ));
 237                     }
 238                     break;
 239                 default:
 240                     break;
 241             }
 242         }
 243     }
 244 }