src/relevance.c

   1 /* This file is part of Pazpar2.
   2    Copyright (C) 2006-2013 Index Data
   3
   4 Pazpar2 is free software; you can redistribute it and/or modify it under
   5 the terms of the GNU General Public License as published by the Free
   6 Software Foundation; either version 2, or (at your option) any later
   7 version.
   8
   9 Pazpar2 is distributed in the hope that it will be useful, but WITHOUT ANY
  10 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12 for more details.
  13
  14 You should have received a copy of the GNU General Public License
  15 along with this program; if not, write to the Free Software
  16 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17
  18 */
  19
  20 #if HAVE_CONFIG_H
  21 #include <config.h>
  22 #endif
  23
  24 #include <assert.h>
  25 #include <math.h>
  26 #include <stdlib.h>
  27
  28 #include "relevance.h"
  29 #include "session.h"
  30
  31 #ifdef WIN32
  32 #define log2(x) (log(x)/log(2))
  33 #endif
  34
  35 struct relevance
  36 {
  37     int *doc_frequency_vec;
  38     int *term_frequency_vec_tmp;
  39     int *term_pos;
  40     int vec_len;
  41     struct word_entry *entries;
  42     pp2_charset_token_t prt;
  43     int rank_cluster;
  44     double follow_factor;
  45     double lead_decay;
  46     int length_divide;
  47     NMEM nmem;
  48 };
  49
  50 struct word_entry {
  51     const char *norm_str;
  52     const char *display_str;
  53     int termno;
  54     char *ccl_field;
  55     struct word_entry *next;
  56 };
  57
  58 static struct word_entry *word_entry_match(struct relevance *r,
  59                                            const char *norm_str,
  60                                            const char *rank, int *weight)
  61 {
  62     int i = 1;
  63     struct word_entry *entries = r->entries;
  64     for (; entries; entries = entries->next, i++)
  65     {
  66         if (*norm_str && !strcmp(norm_str, entries->norm_str))
  67         {
  68             const char *cp = 0;
  69             int no_read = 0;
  70             sscanf(rank, "%d%n", weight, &no_read);
  71             rank += no_read;
  72             while (*rank == ' ')
  73                 rank++;
  74             if (no_read > 0 && (cp = strchr(rank, ' ')))
  75             {
  76                 if ((cp - rank) == strlen(entries->ccl_field) &&
  77                     memcmp(entries->ccl_field, rank, cp - rank) == 0)
  78                     *weight = atoi(cp + 1);
  79             }
  80             return entries;
  81         }
  82     }
  83     return 0;
  84 }
  85
  86 int relevance_snippet(struct relevance *r,
  87                       const char *words, const char *name,
  88                       WRBUF w_snippet)
  89 {
  90     int no = 0;
  91     const char *norm_str;
  92 #if 1
  93     yaz_log(YLOG_LOG, "relevance_snippet for field=%s content=%s",
  94             name, words);
  95 #endif
  96     pp2_charset_token_first(r->prt, words, 0);
  97
  98     while ((norm_str = pp2_charset_token_next(r->prt)))
  99     {
 100         size_t org_start, org_len;
 101         struct word_entry *entries = r->entries;
 102         int highlight = 0;
 103         int i;
 104
 105         pp2_get_org(r->prt, &org_start, &org_len);
 106         for (; entries; entries = entries->next, i++)
 107         {
 108             yaz_log(YLOG_LOG, "Compare: %s %s", norm_str, entries->norm_str);
 109             if (*norm_str && !strcmp(norm_str, entries->norm_str))
 110                 highlight = 1;
 111         }
 112         if (highlight)
 113             wrbuf_puts(w_snippet, "<match>");
 114
 115         wrbuf_xmlputs_n(w_snippet, words + org_start, org_len);
 116         if (highlight)
 117             wrbuf_puts(w_snippet, "</match>");
 118         no += highlight;
 119     }
 120     if (no)
 121     {
 122         yaz_log(YLOG_LOG, "SNIPPET match: %s", wrbuf_cstr(w_snippet));
 123     }
 124     return no;
 125 }
 126
 127 void relevance_countwords(struct relevance *r, struct record_cluster *cluster,
 128                           const char *words, const char *rank,
 129                           const char *name)
 130 {
 131     int *w = r->term_frequency_vec_tmp;
 132     const char *norm_str;
 133     int i, length = 0;
 134     double lead_decay = r->lead_decay;
 135     struct word_entry *e;
 136     WRBUF wr = cluster->relevance_explain1;
 137     int printed_about_field = 0;
 138
 139     pp2_charset_token_first(r->prt, words, 0);
 140     for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
 141     {
 142         w[i] = 0;
 143         r->term_pos[i] = 0;
 144     }
 145
 146     assert(rank);
 147     while ((norm_str = pp2_charset_token_next(r->prt)))
 148     {
 149         int local_weight = 0;
 150         e = word_entry_match(r, norm_str, rank, &local_weight);
 151         if (e)
 152         {
 153             int res = e->termno;
 154             int j;
 155
 156             if (!printed_about_field)
 157             {
 158                 printed_about_field = 1;
 159                 wrbuf_printf(wr, "field=%s content=", name);
 160                 if (strlen(words) > 50)
 161                 {
 162                     wrbuf_xmlputs_n(wr, words, 49);
 163                     wrbuf_puts(wr, " ...");
 164                 }
 165                 else
 166                     wrbuf_xmlputs(wr, words);
 167                 wrbuf_puts(wr, ";\n");
 168             }
 169             assert(res < r->vec_len);
 170             w[res] += local_weight / (1 + log2(1 + lead_decay * length));
 171             wrbuf_printf(wr, "%s: w[%d] += w(%d) / "
 172                          "(1+log2(1+lead_decay(%f) * length(%d)));\n",
 173                          e->display_str, res, local_weight, lead_decay, length);
 174             j = res - 1;
 175             if (j > 0 && r->term_pos[j])
 176             {
 177                 int d = length + 1 - r->term_pos[j];
 178                 wrbuf_printf(wr, "%s: w[%d] += w[%d](%d) * follow(%f) / "
 179                              "(1+log2(d(%d));\n",
 180                              e->display_str, res, res, w[res],
 181                              r->follow_factor, d);
 182                 w[res] += w[res] * r->follow_factor / (1 + log2(d));
 183             }
 184             for (j = 0; j < r->vec_len; j++)
 185                 r->term_pos[j] = j < res ? 0 : length + 1;
 186         }
 187         length++;
 188     }
 189
 190     for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
 191     {
 192         if (length == 0 || w[i] == 0)
 193             continue;
 194         wrbuf_printf(wr, "%s: tf[%d] += w[%d](%d)", e->display_str, i, i, w[i]);
 195         switch (r->length_divide)
 196         {
 197         case 0:
 198             cluster->term_frequency_vecf[i] += (double) w[i];
 199             break;
 200         case 1:
 201             wrbuf_printf(wr, " / log2(1+length(%d))", length);
 202             cluster->term_frequency_vecf[i] +=
 203                 (double) w[i] / log2(1 + length);
 204             break;
 205         case 2:
 206             wrbuf_printf(wr, " / length(%d)", length);
 207             cluster->term_frequency_vecf[i] += (double) w[i] / length;
 208         }
 209         cluster->term_frequency_vec[i] += w[i];
 210         wrbuf_printf(wr, " (%f);\n", cluster->term_frequency_vecf[i]);
 211     }
 212
 213     cluster->term_frequency_vec[0] += length;
 214 }
 215
 216 static void pull_terms(struct relevance *res, struct ccl_rpn_node *n)
 217 {
 218     char **words;
 219     int numwords;
 220     char *ccl_field;
 221     int i;
 222
 223     switch (n->kind)
 224     {
 225     case CCL_RPN_AND:
 226     case CCL_RPN_OR:
 227     case CCL_RPN_NOT:
 228     case CCL_RPN_PROX:
 229         pull_terms(res, n->u.p[0]);
 230         pull_terms(res, n->u.p[1]);
 231         break;
 232     case CCL_RPN_TERM:
 233         nmem_strsplit(res->nmem, " ", n->u.t.term, &words, &numwords);
 234         for (i = 0; i < numwords; i++)
 235         {
 236             const char *norm_str;
 237
 238             ccl_field = nmem_strdup_null(res->nmem, n->u.t.qual);
 239
 240             pp2_charset_token_first(res->prt, words[i], 0);
 241             while ((norm_str = pp2_charset_token_next(res->prt)))
 242             {
 243                 struct word_entry **e = &res->entries;
 244                 while (*e)
 245                     e = &(*e)->next;
 246                 *e = nmem_malloc(res->nmem, sizeof(**e));
 247                 (*e)->norm_str = nmem_strdup(res->nmem, norm_str);
 248                 (*e)->ccl_field = ccl_field;
 249                 (*e)->termno = res->vec_len++;
 250                 (*e)->display_str = nmem_strdup(res->nmem, words[i]);
 251                 (*e)->next = 0;
 252             }
 253         }
 254         break;
 255     default:
 256         break;
 257     }
 258 }
 259 void relevance_clear(struct relevance *r)
 260 {
 261     if (r)
 262     {
 263         int i;
 264         for (i = 0; i < r->vec_len; i++)
 265             r->doc_frequency_vec[i] = 0;
 266     }
 267 }
 268
 269 struct relevance *relevance_create_ccl(pp2_charset_fact_t pft,
 270                                        struct ccl_rpn_node *query,
 271                                        int rank_cluster,
 272                                        double follow_factor, double lead_decay,
 273                                        int length_divide)
 274 {
 275     NMEM nmem = nmem_create();
 276     struct relevance *res = nmem_malloc(nmem, sizeof(*res));
 277
 278     res->nmem = nmem;
 279     res->entries = 0;
 280     res->vec_len = 1;
 281     res->rank_cluster = rank_cluster;
 282     res->follow_factor = follow_factor;
 283     res->lead_decay = lead_decay;
 284     res->length_divide = length_divide;
 285     res->prt = pp2_charset_token_create(pft, "relevance");
 286
 287     pull_terms(res, query);
 288
 289     res->doc_frequency_vec = nmem_malloc(nmem, res->vec_len * sizeof(int));
 290
 291     // worker array
 292     res->term_frequency_vec_tmp =
 293         nmem_malloc(res->nmem,
 294                     res->vec_len * sizeof(*res->term_frequency_vec_tmp));
 295
 296     res->term_pos =
 297         nmem_malloc(res->nmem, res->vec_len * sizeof(*res->term_pos));
 298
 299     relevance_clear(res);
 300     return res;
 301 }
 302
 303 void relevance_destroy(struct relevance **rp)
 304 {
 305     if (*rp)
 306     {
 307         pp2_charset_token_destroy((*rp)->prt);
 308         nmem_destroy((*rp)->nmem);
 309         *rp = 0;
 310     }
 311 }
 312
 313 void relevance_newrec(struct relevance *r, struct record_cluster *rec)
 314 {
 315     if (!rec->term_frequency_vec)
 316     {
 317         int i;
 318
 319         // term frequency [1,..] . [0] is total length of all fields
 320         rec->term_frequency_vec =
 321             nmem_malloc(r->nmem,
 322                         r->vec_len * sizeof(*rec->term_frequency_vec));
 323         for (i = 0; i < r->vec_len; i++)
 324             rec->term_frequency_vec[i] = 0;
 325
 326         // term frequency divided by length of field [1,...]
 327         rec->term_frequency_vecf =
 328             nmem_malloc(r->nmem,
 329                         r->vec_len * sizeof(*rec->term_frequency_vecf));
 330         for (i = 0; i < r->vec_len; i++)
 331             rec->term_frequency_vecf[i] = 0.0;
 332     }
 333 }
 334
 335 void relevance_donerecord(struct relevance *r, struct record_cluster *cluster)
 336 {
 337     int i;
 338
 339     for (i = 1; i < r->vec_len; i++)
 340         if (cluster->term_frequency_vec[i] > 0)
 341             r->doc_frequency_vec[i]++;
 342
 343     r->doc_frequency_vec[0]++;
 344 }
 345
 346 // Prepare for a relevance-sorted read
 347 void relevance_prepare_read(struct relevance *rel, struct reclist *reclist)
 348 {
 349     int i;
 350     float *idfvec = xmalloc(rel->vec_len * sizeof(float));
 351
 352     reclist_enter(reclist);
 353     // Calculate document frequency vector for each term.
 354     for (i = 1; i < rel->vec_len; i++)
 355     {
 356         if (!rel->doc_frequency_vec[i])
 357             idfvec[i] = 0;
 358         else
 359         {
 360             /* add one to nominator idf(t,D) to ensure a value > 0 */
 361             idfvec[i] = log((float) (1 + rel->doc_frequency_vec[0]) /
 362                             rel->doc_frequency_vec[i]);
 363         }
 364     }
 365     // Calculate relevance for each document
 366     while (1)
 367     {
 368         int relevance = 0;
 369         WRBUF w;
 370         struct word_entry *e = rel->entries;
 371         struct record_cluster *rec = reclist_read_record(reclist);
 372         if (!rec)
 373             break;
 374         w = rec->relevance_explain2;
 375         wrbuf_rewind(w);
 376         wrbuf_puts(w, "relevance = 0;\n");
 377         for (i = 1; i < rel->vec_len; i++)
 378         {
 379             float termfreq = (float) rec->term_frequency_vecf[i];
 380             int add = 100000 * termfreq * idfvec[i];
 381
 382             wrbuf_printf(w, "idf[%d] = log(((1 + total(%d))/termoccur(%d));\n",
 383                          i, rel->doc_frequency_vec[0],
 384                          rel->doc_frequency_vec[i]);
 385             wrbuf_printf(w, "%s: relevance += 100000 * tf[%d](%f) * "
 386                          "idf[%d](%f) (%d);\n",
 387                          e->display_str, i, termfreq, i, idfvec[i], add);
 388             relevance += add;
 389             e = e->next;
 390         }
 391         if (!rel->rank_cluster)
 392         {
 393             struct record *record;
 394             int cluster_size = 0;
 395
 396             for (record = rec->records; record; record = record->next)
 397                 cluster_size++;
 398
 399             wrbuf_printf(w, "score = relevance(%d)/cluster_size(%d);\n",
 400                          relevance, cluster_size);
 401             relevance /= cluster_size;
 402         }
 403         else
 404         {
 405             wrbuf_printf(w, "score = relevance(%d);\n", relevance);
 406         }
 407         rec->relevance_score = relevance;
 408     }
 409     reclist_leave(reclist);
 410     xfree(idfvec);
 411 }
 412
 413 /*
 414  * Local variables:
 415  * c-basic-offset: 4
 416  * c-file-style: "Stroustrup"
 417  * indent-tabs-mode: nil
 418  * End:
 419  * vim: shiftwidth=4 tabstop=8 expandtab
 420  */
 421