lib/tokeniserlib.php

   1 <?php // $Id$
   2
   3 ///////////////////////////////////////////////////////////////////////////
   4 //                                                                       //
   5 // NOTICE OF COPYRIGHT                                                   //
   6 //                                                                       //
   7 // Original code:                                                        //
   8 //                                                                       //
   9 // Drupal - The copyright of both the Drupal software and the            //
  10 //          "Druplicon" logo belongs to all the original authors,        //
  11 //          though both are licensed under the GPL.                      //
  12 //          http://drupal.org                                            //
  13 //                                                                       //
  14 // Modifications:                                                        //
  15 //                                                                       //
  16 // Moodle - Modular Object-Oriented Dynamic Learning Environment         //
  17 //          http://moodle.com                                            //
  18 //                                                                       //
  19 // Copyright (C) 2001-3001 Martin Dougiamas        http://dougiamas.com  //
  20 //           (C) 2001-3001 Eloy Lafuente (stronk7) http://contiento.com  //
  21 //           (C) 2001-3001 Antonio Vicent          http://ludens.es      //
  22 //                                                                       //
  23 // This program is free software; you can redistribute it and/or modify  //
  24 // it under the terms of the GNU General Public License as published by  //
  25 // the Free Software Foundation; either version 2 of the License, or     //
  26 // (at your option) any later version.                                   //
  27 //                                                                       //
  28 // This program is distributed in the hope that it will be useful,       //
  29 // but WITHOUT ANY WARRANTY; without even the implied warranty of        //
  30 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         //
  31 // GNU General Public License for more details:                          //
  32 //                                                                       //
  33 //          http://www.gnu.org/copyleft/gpl.html                         //
  34 //                                                                       //
  35 ///////////////////////////////////////////////////////////////////////////
  36
  37 /// Based on Drupal's search.module version 1.224
  38 /// http://cvs.drupal.org/viewcvs/drupal/drupal/modules/search/search.module?view=markup
  39
  40 /// Usage: $tokens = tokenise_text($text)
  41 /// Returns an array of tokens (key) with their score (value)
  42 /// (see function definition for more info)
  43
  44 /// Some constants
  45
  46 define ('MINIMUM_WORD_SIZE',  3); /// Minimum word size to index and search
  47 define ('MAXIMUM_WORD_SIZE', 50); /// Maximum word size to index and search
  48
  49 define ('START_DELIM',  "\xce\xa9\xc3\x91\xc3\x91"); // Use these (omega) and (ntilde) utf-8 combinations
  50 define ('CENTER_DELIM', "\xc3\x91\xce\xa9\xc3\x91"); // as safe delimiters (not puncuation nor usual) while
  51 define ('END_DELIM',    "\xc3\x91\xc3\x91\xce\xa9"); // processing numbers ($join_numbers = false)
  52
  53 /**
  54  * Matches Unicode character classes to exclude from the search index.
  55  *
  56  * See: http://www.unicode.org/Public/UNIDATA/UCD.html#General_Category_Values
  57  *
  58  * The index only contains the following character classes:
  59  * Lu         Letter, Uppercase
  60  * Ll         Letter, Lowercase
  61  * Lt         Letter, Titlecase
  62  * Lo         Letter, Other
  63  * Nd         Number, Decimal Digit
  64  * No         Number, Other
  65  */
  66 define('PREG_CLASS_SEARCH_EXCLUDE',
  67 '\x{0}-\x{2f}\x{3a}-\x{40}\x{5b}-\x{60}\x{7b}-\x{bf}\x{d7}\x{f7}\x{2b0}-'.
  68 '\x{385}\x{387}\x{3f6}\x{482}-\x{489}\x{559}-\x{55f}\x{589}-\x{5c7}\x{5f3}-'.
  69 '\x{61f}\x{640}\x{64b}-\x{65e}\x{66a}-\x{66d}\x{670}\x{6d4}\x{6d6}-\x{6ed}'.
  70 '\x{6fd}\x{6fe}\x{700}-\x{70f}\x{711}\x{730}-\x{74a}\x{7a6}-\x{7b0}\x{901}-'.
  71 '\x{903}\x{93c}\x{93e}-\x{94d}\x{951}-\x{954}\x{962}-\x{965}\x{970}\x{981}-'.
  72 '\x{983}\x{9bc}\x{9be}-\x{9cd}\x{9d7}\x{9e2}\x{9e3}\x{9f2}-\x{a03}\x{a3c}-'.
  73 '\x{a4d}\x{a70}\x{a71}\x{a81}-\x{a83}\x{abc}\x{abe}-\x{acd}\x{ae2}\x{ae3}'.
  74 '\x{af1}-\x{b03}\x{b3c}\x{b3e}-\x{b57}\x{b70}\x{b82}\x{bbe}-\x{bd7}\x{bf0}-'.
  75 '\x{c03}\x{c3e}-\x{c56}\x{c82}\x{c83}\x{cbc}\x{cbe}-\x{cd6}\x{d02}\x{d03}'.
  76 '\x{d3e}-\x{d57}\x{d82}\x{d83}\x{dca}-\x{df4}\x{e31}\x{e34}-\x{e3f}\x{e46}-'.
  77 '\x{e4f}\x{e5a}\x{e5b}\x{eb1}\x{eb4}-\x{ebc}\x{ec6}-\x{ecd}\x{f01}-\x{f1f}'.
  78 '\x{f2a}-\x{f3f}\x{f71}-\x{f87}\x{f90}-\x{fd1}\x{102c}-\x{1039}\x{104a}-'.
  79 '\x{104f}\x{1056}-\x{1059}\x{10fb}\x{10fc}\x{135f}-\x{137c}\x{1390}-\x{1399}'.
  80 '\x{166d}\x{166e}\x{1680}\x{169b}\x{169c}\x{16eb}-\x{16f0}\x{1712}-\x{1714}'.
  81 '\x{1732}-\x{1736}\x{1752}\x{1753}\x{1772}\x{1773}\x{17b4}-\x{17db}\x{17dd}'.
  82 '\x{17f0}-\x{180e}\x{1843}\x{18a9}\x{1920}-\x{1945}\x{19b0}-\x{19c0}\x{19c8}'.
  83 '\x{19c9}\x{19de}-\x{19ff}\x{1a17}-\x{1a1f}\x{1d2c}-\x{1d61}\x{1d78}\x{1d9b}-'.
  84 '\x{1dc3}\x{1fbd}\x{1fbf}-\x{1fc1}\x{1fcd}-\x{1fcf}\x{1fdd}-\x{1fdf}\x{1fed}-'.
  85 '\x{1fef}\x{1ffd}-\x{2070}\x{2074}-\x{207e}\x{2080}-\x{2101}\x{2103}-\x{2106}'.
  86 '\x{2108}\x{2109}\x{2114}\x{2116}-\x{2118}\x{211e}-\x{2123}\x{2125}\x{2127}'.
  87 '\x{2129}\x{212e}\x{2132}\x{213a}\x{213b}\x{2140}-\x{2144}\x{214a}-\x{2b13}'.
  88 '\x{2ce5}-\x{2cff}\x{2d6f}\x{2e00}-\x{3005}\x{3007}-\x{303b}\x{303d}-\x{303f}'.
  89 '\x{3099}-\x{309e}\x{30a0}\x{30fb}-\x{30fe}\x{3190}-\x{319f}\x{31c0}-\x{31cf}'.
  90 '\x{3200}-\x{33ff}\x{4dc0}-\x{4dff}\x{a015}\x{a490}-\x{a716}\x{a802}\x{a806}'.
  91 '\x{a80b}\x{a823}-\x{a82b}\x{d800}-\x{f8ff}\x{fb1e}\x{fb29}\x{fd3e}\x{fd3f}'.
  92 '\x{fdfc}-\x{fe6b}\x{feff}-\x{ff0f}\x{ff1a}-\x{ff20}\x{ff3b}-\x{ff40}\x{ff5b}-'.
  93 '\x{ff65}\x{ff70}\x{ff9e}\x{ff9f}\x{ffe0}-\x{fffd}');
  94
  95 /**
  96  * Matches all 'N' Unicode character classes (numbers)
  97  */
  98 define('PREG_CLASS_NUMBERS',
  99 '\x{30}-\x{39}\x{b2}\x{b3}\x{b9}\x{bc}-\x{be}\x{660}-\x{669}\x{6f0}-\x{6f9}'.
 100 '\x{966}-\x{96f}\x{9e6}-\x{9ef}\x{9f4}-\x{9f9}\x{a66}-\x{a6f}\x{ae6}-\x{aef}'.
 101 '\x{b66}-\x{b6f}\x{be7}-\x{bf2}\x{c66}-\x{c6f}\x{ce6}-\x{cef}\x{d66}-\x{d6f}'.
 102 '\x{e50}-\x{e59}\x{ed0}-\x{ed9}\x{f20}-\x{f33}\x{1040}-\x{1049}\x{1369}-'.
 103 '\x{137c}\x{16ee}-\x{16f0}\x{17e0}-\x{17e9}\x{17f0}-\x{17f9}\x{1810}-\x{1819}'.
 104 '\x{1946}-\x{194f}\x{2070}\x{2074}-\x{2079}\x{2080}-\x{2089}\x{2153}-\x{2183}'.
 105 '\x{2460}-\x{249b}\x{24ea}-\x{24ff}\x{2776}-\x{2793}\x{3007}\x{3021}-\x{3029}'.
 106 '\x{3038}-\x{303a}\x{3192}-\x{3195}\x{3220}-\x{3229}\x{3251}-\x{325f}\x{3280}-'.
 107 '\x{3289}\x{32b1}-\x{32bf}\x{ff10}-\x{ff19}');
 108
 109 /**
 110  * Matches all 'P' Unicode character classes (punctuation)
 111  */
 112 define('PREG_CLASS_PUNCTUATION',
 113 '\x{21}-\x{23}\x{25}-\x{2a}\x{2c}-\x{2f}\x{3a}\x{3b}\x{3f}\x{40}\x{5b}-\x{5d}'.
 114 '\x{5f}\x{7b}\x{7d}\x{a1}\x{ab}\x{b7}\x{bb}\x{bf}\x{37e}\x{387}\x{55a}-\x{55f}'.
 115 '\x{589}\x{58a}\x{5be}\x{5c0}\x{5c3}\x{5f3}\x{5f4}\x{60c}\x{60d}\x{61b}\x{61f}'.
 116 '\x{66a}-\x{66d}\x{6d4}\x{700}-\x{70d}\x{964}\x{965}\x{970}\x{df4}\x{e4f}'.
 117 '\x{e5a}\x{e5b}\x{f04}-\x{f12}\x{f3a}-\x{f3d}\x{f85}\x{104a}-\x{104f}\x{10fb}'.
 118 '\x{1361}-\x{1368}\x{166d}\x{166e}\x{169b}\x{169c}\x{16eb}-\x{16ed}\x{1735}'.
 119 '\x{1736}\x{17d4}-\x{17d6}\x{17d8}-\x{17da}\x{1800}-\x{180a}\x{1944}\x{1945}'.
 120 '\x{2010}-\x{2027}\x{2030}-\x{2043}\x{2045}-\x{2051}\x{2053}\x{2054}\x{2057}'.
 121 '\x{207d}\x{207e}\x{208d}\x{208e}\x{2329}\x{232a}\x{23b4}-\x{23b6}\x{2768}-'.
 122 '\x{2775}\x{27e6}-\x{27eb}\x{2983}-\x{2998}\x{29d8}-\x{29db}\x{29fc}\x{29fd}'.
 123 '\x{3001}-\x{3003}\x{3008}-\x{3011}\x{3014}-\x{301f}\x{3030}\x{303d}\x{30a0}'.
 124 '\x{30fb}\x{fd3e}\x{fd3f}\x{fe30}-\x{fe52}\x{fe54}-\x{fe61}\x{fe63}\x{fe68}'.
 125 '\x{fe6a}\x{fe6b}\x{ff01}-\x{ff03}\x{ff05}-\x{ff0a}\x{ff0c}-\x{ff0f}\x{ff1a}'.
 126 '\x{ff1b}\x{ff1f}\x{ff20}\x{ff3b}-\x{ff3d}\x{ff3f}\x{ff5b}\x{ff5d}\x{ff5f}-'.
 127 '\x{ff65}');
 128
 129 /**
 130  * Matches all CJK characters that are candidates for auto-splitting
 131  * (Chinese, Japanese, Korean).
 132  * Contains kana and BMP ideographs.
 133  */
 134 define('PREG_CLASS_CJK', '\x{3041}-\x{30ff}\x{31f0}-\x{31ff}\x{3400}-\x{4db5}'.
 135 '\x{4e00}-\x{9fbb}\x{f900}-\x{fad9}');
 136
 137
 138 /**
 139  * This function process the text passed at input, extracting all the tokens
 140  * and scoring each one based in their number of ocurrences and relation with
 141  * some well-known html tags
 142  *
 143  * @param string  $text the text to be tokenised.
 144  * @param array   $stop_words array of utf-8 words than can be ignored in
 145  *                the text being processed. There are some cool lists of
 146  *                stop words at http://snowball.tartarus.org/
 147  * @param boolean $overlap_cjk option to split CJK text into some overlapping
 148  *                tokens is order to allow them to be searched. Useful to build
 149  *                indexes and search systems.
 150  * @param boolean $join_numbers option to join in one unique token sequences of numbers
 151  *                separated by puntuaction chars. Useful to build indexes and
 152  *                search systems.
 153  * @return array one sorted array of tokens, with tokens being the keys and scores in the values.
 154  */
 155 function tokenise_text($text, $stop_words = array(), $overlap_cjk = false, $join_numbers = false) {
 156
 157     $textlib = textlib_get_instance();
 158
 159     // Multipliers for scores of words inside certain HTML tags.
 160     // Note: 'a' must be included for link ranking to work.
 161     $tags = array('h1' => 25,
 162                   'h2' => 18,
 163                   'h3' => 15,
 164                   'h4' => 12,
 165                   'h5' => 9,
 166                   'h6' => 6,
 167                   'u' => 3,
 168                   'b' => 3,
 169                   'i' => 3,
 170                   'strong' => 3,
 171                   'em' => 3,
 172                   'a' => 10);
 173
 174     // Strip off all ignored tags to speed up processing, but insert space before/after
 175     // them to keep word boundaries.
 176     $text = str_replace(array('<', '>'), array(' <', '> '), $text);
 177     $text = strip_tags($text, '<'. implode('><', array_keys($tags)) .'>');
 178
 179     // Split HTML tags from plain text.
 180     $split = preg_split('/\s*<([^>]+?)>\s*/', $text, -1, PREG_SPLIT_DELIM_CAPTURE);
 181     // Note: PHP ensures the array consists of alternating delimiters and literals
 182     // and begins and ends with a literal (inserting $null as required).
 183
 184     $tag = FALSE; // Odd/even counter. Tag or no tag.
 185     $score = 1; // Starting score per word
 186     $accum = ' '; // Accumulator for cleaned up data
 187     $tagstack = array(); // Stack with open tags
 188     $tagwords = 0; // Counter for consecutive words
 189     $focus = 1; // Focus state
 190
 191     $results = array(0 => array()); // Accumulator for words for index
 192
 193     foreach ($split as $value) {
 194         if ($tag) {
 195             // Increase or decrease score per word based on tag
 196             list($tagname) = explode(' ', $value, 2);
 197             $tagname = $textlib->strtolower($tagname);
 198             // Closing or opening tag?
 199             if ($tagname[0] == '/') {
 200                 $tagname = substr($tagname, 1);
 201                 // If we encounter unexpected tags, reset score to avoid incorrect boosting.
 202                 if (!count($tagstack) || $tagstack[0] != $tagname) {
 203                     $tagstack = array();
 204                     $score = 1;
 205                 }
 206                 else {
 207                     // Remove from tag stack and decrement score
 208                     $score = max(1, $score - $tags[array_shift($tagstack)]);
 209                 }
 210             }
 211             else {
 212                 if (isset($tagstack[0]) && $tagstack[0] == $tagname) {
 213                     // None of the tags we look for make sense when nested identically.
 214                     // If they are, it's probably broken HTML.
 215                     $tagstack = array();
 216                     $score = 1;
 217                 }
 218                 else {
 219                     // Add to open tag stack and increment score
 220                     array_unshift($tagstack, $tagname);
 221                     $score += $tags[$tagname];
 222                 }
 223             }
 224             // A tag change occurred, reset counter.
 225             $tagwords = 0;
 226         }
 227         else {
 228             // Note: use of PREG_SPLIT_DELIM_CAPTURE above will introduce empty values
 229             if ($value != '') {
 230                 $words = tokenise_split($value, $stop_words, $overlap_cjk, $join_numbers);
 231                 foreach ($words as $word) {
 232                     // Add word to accumulator
 233                     $accum .= $word .' ';
 234                     $num = is_numeric($word);
 235                     // Check word length
 236                     if ($num || $textlib->strlen($word) >= MINIMUM_WORD_SIZE) {
 237                         // Normalize numbers
 238                         if ($num && $join_numbers) {
 239                             $word = (int)ltrim($word, '-0');
 240                         }
 241
 242                         if (!isset($results[0][$word])) {
 243                             $results[0][$word] = 0;
 244                         }
 245
 246                         $results[0][$word] += $score * $focus;
 247                         // Focus is a decaying value in terms of the amount of unique words up to this point.
 248                         // From 100 words and more, it decays, to e.g. 0.5 at 500 words and 0.3 at 1000 words.
 249                         $focus = min(1, .01 + 3.5 / (2 + count($results[0]) * .015));
 250                     }
 251                     $tagwords++;
 252                     // Too many words inside a single tag probably mean a tag was accidentally left open.
 253                     if (count($tagstack) && $tagwords >= 15) {
 254                         $tagstack = array();
 255                         $score = 1;
 256                     }
 257                 }
 258             }
 259         }
 260         $tag = !$tag;
 261     }
 262
 263     $res = array();
 264
 265     if (isset($results[0]) && count($results[0]) > 0) {
 266         $res = $results[0];
 267         arsort($res, SORT_NUMERIC);
 268     }
 269
 270     return $res;
 271 }
 272
 273 ///
 274 /// Some helper functions (should be considered private)
 275 ///
 276
 277 /**
 278  * Splits a string into tokens
 279  */
 280 function tokenise_split($text, $stop_words, $overlap_cjk, $join_numbers) {
 281     static $last = NULL;
 282     static $lastsplit = NULL;
 283
 284     if ($last == $text) {
 285         return $lastsplit;
 286     }
 287     // Process words
 288     $text = tokenise_simplify($text, $overlap_cjk, $join_numbers);
 289     $words = explode(' ', $text);
 290     // Limit word length to 50
 291     array_walk($words, 'tokenise_truncate_word');
 292
 293     // We have stop words, apply them
 294     if (is_array($stop_words) && !empty($stop_words)) {
 295         // Normalise them
 296         $simp_stop_words = explode(' ', tokenise_simplify(implode(' ', $stop_words), $overlap_cjk, $join_numbers));
 297         // Extract from the list
 298         $words = array_diff($words, $simp_stop_words);
 299     }
 300     // Save last keyword result
 301     $last = $text;
 302     $lastsplit = $words;
 303
 304     return $words;
 305 }
 306
 307 /**
 308  * Simplifies a string according to indexing rules.
 309  */
 310 function tokenise_simplify($text, $overlap_cjk, $join_numbers) {
 311
 312     $textlib = textlib_get_instance();
 313
 314     // Decode entities to UTF-8
 315     $text = $textlib->entities_to_utf8($text, true);
 316
 317     // Lowercase
 318     $text = $textlib->strtolower($text);
 319
 320     // Simple CJK handling
 321     if ($overlap_cjk) {
 322         $text = preg_replace_callback('/['. PREG_CLASS_CJK .']+/u', 'tokenise_expand_cjk', $text);
 323     }
 324
 325     // To improve searching for numerical data such as dates, IP addresses
 326     // or version numbers, we consider a group of numerical characters
 327     // separated only by punctuation characters to be one piece.
 328     // This also means that searching for e.g. '20/03/1984' also returns
 329     // results with '20-03-1984' in them.
 330     // Readable regexp: ([number]+)[punctuation]+(?=[number])
 331     if ($join_numbers) {
 332         $text = preg_replace('/(['. PREG_CLASS_NUMBERS .']+)['. PREG_CLASS_PUNCTUATION .']+(?=['. PREG_CLASS_NUMBERS .'])/u', '\1', $text);
 333     } else {
 334     // Keep all the detected numbers+punctuation in a safe place in order to restore them later
 335         preg_match_all('/['. PREG_CLASS_NUMBERS .']+['. PREG_CLASS_PUNCTUATION . PREG_CLASS_NUMBERS .']+/u', $text, $foundseqs);
 336         $storedseqs = array();
 337         foreach (array_unique($foundseqs[0]) as $ntkey => $value) {
 338             $prefix = (string)(count($storedseqs) + 1);
 339             $storedseqs[START_DELIM.$prefix.CENTER_DELIM.$ntkey.END_DELIM] = $value;
 340         }
 341         if (!empty($storedseqs)) {
 342             $text = str_replace($storedseqs, array_keys($storedseqs), $text);
 343         }
 344     }
 345
 346     // The dot, underscore and dash are simply removed. This allows meaningful
 347     // search behaviour with acronyms and URLs.
 348     $text = preg_replace('/[._-]+/', '', $text);
 349
 350     // With the exception of the rules above, we consider all punctuation,
 351     // marks, spacers, etc, to be a word boundary.
 352     $text = preg_replace('/['. PREG_CLASS_SEARCH_EXCLUDE .']+/u', ' ', $text);
 353
 354     // Restore, if not joining numbers, recover the original strings
 355     if (!$join_numbers) {
 356         if (!empty($storedseqs)) {
 357             $text = str_replace(array_keys($storedseqs), $storedseqs, $text);
 358         }
 359     }
 360
 361     return $text;
 362 }
 363
 364 /**
 365  * Basic CJK tokeniser. Simply splits a string into consecutive, overlapping
 366  * sequences of characters (MINIMUM_WORD_SIZE long).
 367  */
 368 function tokenise_expand_cjk($matches) {
 369
 370     $textlib = textlib_get_instance();
 371
 372     $str = $matches[0];
 373     $l = $textlib->strlen($str);
 374     // Passthrough short words
 375     if ($l <= MINIMUM_WORD_SIZE) {
 376         return ' '. $str .' ';
 377     }
 378     $tokens = ' ';
 379     // FIFO queue of characters
 380     $chars = array();
 381     // Begin loop
 382     for ($i = 0; $i < $l; ++$i) {
 383         // Grab next character
 384         $current = $textlib->substr($str, 0, 1);
 385         $str = substr($str, strlen($current));
 386         $chars[] = $current;
 387         if ($i >= MINIMUM_WORD_SIZE - 1) {
 388             $tokens .= implode('', $chars) .' ';
 389             array_shift($chars);
 390         }
 391     }
 392     return $tokens;
 393 }
 394
 395 /**
 396  * Helper function for array_walk in search_index_split.
 397  * Truncates one string (token) to MAXIMUM_WORD_SIZE
 398  */
 399 function tokenise_truncate_word(&$text) {
 400
 401     $textlib = textlib_get_instance();
 402     $text = $textlib->substr($text, 0, MAXIMUM_WORD_SIZE);
 403 }
 404
 405 ?>