src/relevance.c

   1 /* This file is part of Pazpar2.
   2    Copyright (C) 2006-2012 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 struct relevance
  32 {
  33     int *doc_frequency_vec;
  34     int *term_frequency_vec_tmp;
  35     int *term_pos;
  36     int vec_len;
  37     struct word_entry *entries;
  38     pp2_charset_token_t prt;
  39     int rank_cluster;
  40     double follow_factor;
  41     double lead_decay;
  42     int length_divide;
  43     NMEM nmem;
  44 };
  45
  46 struct word_entry {
  47     const char *norm_str;
  48     const char *display_str;
  49     int termno;
  50     char *ccl_field;
  51     struct word_entry *next;
  52 };
  53
  54 static struct word_entry *word_entry_match(struct relevance *r,
  55                                            const char *norm_str,
  56                                            const char *rank, int *weight)
  57 {
  58     int i = 1;
  59     struct word_entry *entries = r->entries;
  60     for (; entries; entries = entries->next, i++)
  61     {
  62         if (*norm_str && !strcmp(norm_str, entries->norm_str))
  63         {
  64             const char *cp = 0;
  65             int no_read = 0;
  66             sscanf(rank, "%d%n", weight, &no_read);
  67             rank += no_read;
  68             while (*rank == ' ')
  69                 rank++;
  70             if (no_read > 0 && (cp = strchr(rank, ' ')))
  71             {
  72                 if ((cp - rank) == strlen(entries->ccl_field) &&
  73                     memcmp(entries->ccl_field, rank, cp - rank) == 0)
  74                     *weight = atoi(cp + 1);
  75             }
  76             return entries;
  77         }
  78     }
  79     return 0;
  80 }
  81
  82 void relevance_countwords(struct relevance *r, struct record_cluster *cluster,
  83                           const char *words, const char *rank,
  84                           const char *name)
  85 {
  86     int *w = r->term_frequency_vec_tmp;
  87     const char *norm_str;
  88     int i, length = 0;
  89     double lead_decay = r->lead_decay;
  90     struct word_entry *e;
  91     WRBUF wr = cluster->relevance_explain1;
  92     int printed_about_field = 0;
  93
  94     pp2_charset_token_first(r->prt, words, 0);
  95     for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
  96     {
  97         w[i] = 0;
  98         r->term_pos[i] = 0;
  99     }
 100
 101     assert(rank);
 102     while ((norm_str = pp2_charset_token_next(r->prt)))
 103     {
 104         int local_weight = 0;
 105         e = word_entry_match(r, norm_str, rank, &local_weight);
 106         if (e)
 107         {
 108             int res = e->termno;
 109             int j;
 110
 111             if (!printed_about_field)
 112             {
 113                 printed_about_field = 1;
 114                 wrbuf_printf(wr, "field=%s content=", name);
 115                 if (strlen(words) > 50)
 116                 {
 117                     wrbuf_xmlputs_n(wr, words, 49);
 118                     wrbuf_puts(wr, " ...");
 119                 }
 120                 else
 121                     wrbuf_xmlputs(wr, words);
 122                 wrbuf_puts(wr, ";\n");
 123             }
 124             assert(res < r->vec_len);
 125             w[res] += local_weight / (1 + log2(1 + lead_decay * length));
 126             wrbuf_printf(wr, "%s: w[%d] += w(%d) / "
 127                          "(1+log2(1+lead_decay(%f) * length(%d)));\n",
 128                          e->display_str, res, local_weight, lead_decay, length);
 129             j = res - 1;
 130             if (j > 0 && r->term_pos[j])
 131             {
 132                 int d = length + 1 - r->term_pos[j];
 133                 wrbuf_printf(wr, "%s: w[%d] += w[%d](%d) * follow(%f) / "
 134                              "(1+log2(d(%d));\n",
 135                              e->display_str, res, res, w[res],
 136                              r->follow_factor, d);
 137                 w[res] += w[res] * r->follow_factor / (1 + log2(d));
 138             }
 139             for (j = 0; j < r->vec_len; j++)
 140                 r->term_pos[j] = j < res ? 0 : length + 1;
 141         }
 142         length++;
 143     }
 144
 145     for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
 146     {
 147         if (length == 0 || w[i] == 0)
 148             continue;
 149         wrbuf_printf(wr, "%s: tf[%d] += w[%d](%d)", e->display_str, i, i, w[i]);
 150         switch (r->length_divide)
 151         {
 152         case 0:
 153             cluster->term_frequency_vecf[i] += (double) w[i];
 154             break;
 155         case 1:
 156             wrbuf_printf(wr, " / log2(1+length(%d))", length);
 157             cluster->term_frequency_vecf[i] +=
 158                 (double) w[i] / log2(1 + length);
 159             break;
 160         case 2:
 161             wrbuf_printf(wr, " / length(%d)", length);
 162             cluster->term_frequency_vecf[i] += (double) w[i] / length;
 163         }
 164         cluster->term_frequency_vec[i] += w[i];
 165         wrbuf_printf(wr, " (%f);\n", cluster->term_frequency_vecf[i]);
 166     }
 167
 168     cluster->term_frequency_vec[0] += length;
 169 }
 170
 171 static void pull_terms(struct relevance *res, struct ccl_rpn_node *n)
 172 {
 173     char **words;
 174     int numwords;
 175     char *ccl_field;
 176     int i;
 177
 178     switch (n->kind)
 179     {
 180     case CCL_RPN_AND:
 181     case CCL_RPN_OR:
 182     case CCL_RPN_NOT:
 183     case CCL_RPN_PROX:
 184         pull_terms(res, n->u.p[0]);
 185         pull_terms(res, n->u.p[1]);
 186         break;
 187     case CCL_RPN_TERM:
 188         nmem_strsplit(res->nmem, " ", n->u.t.term, &words, &numwords);
 189         for (i = 0; i < numwords; i++)
 190         {
 191             const char *norm_str;
 192
 193             ccl_field = nmem_strdup_null(res->nmem, n->u.t.qual);
 194
 195             pp2_charset_token_first(res->prt, words[i], 0);
 196             while ((norm_str = pp2_charset_token_next(res->prt)))
 197             {
 198                 struct word_entry **e = &res->entries;
 199                 while (*e)
 200                     e = &(*e)->next;
 201                 *e = nmem_malloc(res->nmem, sizeof(**e));
 202                 (*e)->norm_str = nmem_strdup(res->nmem, norm_str);
 203                 (*e)->ccl_field = ccl_field;
 204                 (*e)->termno = res->vec_len++;
 205                 (*e)->display_str = nmem_strdup(res->nmem, words[i]);
 206                 (*e)->next = 0;
 207             }
 208         }
 209         break;
 210     default:
 211         break;
 212     }
 213 }
 214
 215 struct relevance *relevance_create_ccl(pp2_charset_fact_t pft,
 216                                        struct ccl_rpn_node *query,
 217                                        int rank_cluster,
 218                                        double follow_factor, double lead_decay,
 219                                        int length_divide)
 220 {
 221     NMEM nmem = nmem_create();
 222     struct relevance *res = nmem_malloc(nmem, sizeof(*res));
 223     int i;
 224
 225     res->nmem = nmem;
 226     res->entries = 0;
 227     res->vec_len = 1;
 228     res->rank_cluster = rank_cluster;
 229     res->follow_factor = follow_factor;
 230     res->lead_decay = lead_decay;
 231     res->length_divide = length_divide;
 232     res->prt = pp2_charset_token_create(pft, "relevance");
 233
 234     pull_terms(res, query);
 235
 236     res->doc_frequency_vec = nmem_malloc(nmem, res->vec_len * sizeof(int));
 237     for (i = 0; i < res->vec_len; i++)
 238         res->doc_frequency_vec[i] = 0;
 239
 240     // worker array
 241     res->term_frequency_vec_tmp =
 242         nmem_malloc(res->nmem,
 243                     res->vec_len * sizeof(*res->term_frequency_vec_tmp));
 244
 245     res->term_pos =
 246         nmem_malloc(res->nmem, res->vec_len * sizeof(*res->term_pos));
 247
 248     return res;
 249 }
 250
 251 void relevance_destroy(struct relevance **rp)
 252 {
 253     if (*rp)
 254     {
 255         pp2_charset_token_destroy((*rp)->prt);
 256         nmem_destroy((*rp)->nmem);
 257         *rp = 0;
 258     }
 259 }
 260
 261 void relevance_newrec(struct relevance *r, struct record_cluster *rec)
 262 {
 263     if (!rec->term_frequency_vec)
 264     {
 265         int i;
 266
 267         // term frequency [1,..] . [0] is total length of all fields
 268         rec->term_frequency_vec =
 269             nmem_malloc(r->nmem,
 270                         r->vec_len * sizeof(*rec->term_frequency_vec));
 271         for (i = 0; i < r->vec_len; i++)
 272             rec->term_frequency_vec[i] = 0;
 273
 274         // term frequency divided by length of field [1,...]
 275         rec->term_frequency_vecf =
 276             nmem_malloc(r->nmem,
 277                         r->vec_len * sizeof(*rec->term_frequency_vecf));
 278         for (i = 0; i < r->vec_len; i++)
 279             rec->term_frequency_vecf[i] = 0.0;
 280     }
 281 }
 282
 283 void relevance_donerecord(struct relevance *r, struct record_cluster *cluster)
 284 {
 285     int i;
 286
 287     for (i = 1; i < r->vec_len; i++)
 288         if (cluster->term_frequency_vec[i] > 0)
 289             r->doc_frequency_vec[i]++;
 290
 291     r->doc_frequency_vec[0]++;
 292 }
 293
 294 // Prepare for a relevance-sorted read
 295 void relevance_prepare_read(struct relevance *rel, struct reclist *reclist)
 296 {
 297     int i;
 298     float *idfvec = xmalloc(rel->vec_len * sizeof(float));
 299
 300     reclist_enter(reclist);
 301     // Calculate document frequency vector for each term.
 302     for (i = 1; i < rel->vec_len; i++)
 303     {
 304         if (!rel->doc_frequency_vec[i])
 305             idfvec[i] = 0;
 306         else
 307         {
 308             /* add one to nominator idf(t,D) to ensure a value > 0 */
 309             idfvec[i] = log((float) (1 + rel->doc_frequency_vec[0]) /
 310                             rel->doc_frequency_vec[i]);
 311         }
 312     }
 313     // Calculate relevance for each document
 314     while (1)
 315     {
 316         int relevance = 0;
 317         WRBUF w;
 318         struct word_entry *e = rel->entries;
 319         struct record_cluster *rec = reclist_read_record(reclist);
 320         if (!rec)
 321             break;
 322         w = rec->relevance_explain2;
 323         wrbuf_rewind(w);
 324         wrbuf_puts(w, "relevance = 0;\n");
 325         for (i = 1; i < rel->vec_len; i++)
 326         {
 327             float termfreq = (float) rec->term_frequency_vecf[i];
 328             int add = 100000 * termfreq * idfvec[i];
 329
 330             wrbuf_printf(w, "idf[%d] = log(((1 + total(%d))/termoccur(%d));\n",
 331                          i, rel->doc_frequency_vec[0],
 332                          rel->doc_frequency_vec[i]);
 333             wrbuf_printf(w, "%s: relevance += 100000 * tf[%d](%f) * "
 334                          "idf[%d](%f) (%d);\n",
 335                          e->display_str, i, termfreq, i, idfvec[i], add);
 336             relevance += add;
 337             e = e->next;
 338         }
 339         if (!rel->rank_cluster)
 340         {
 341             struct record *record;
 342             int cluster_size = 0;
 343
 344             for (record = rec->records; record; record = record->next)
 345                 cluster_size++;
 346
 347             wrbuf_printf(w, "score = relevance(%d)/cluster_size(%d);\n",
 348                          relevance, cluster_size);
 349             relevance /= cluster_size;
 350         }
 351         else
 352         {
 353             wrbuf_printf(w, "score = relevance(%d);\n", relevance);
 354         }
 355         rec->relevance_score = relevance;
 356     }
 357     reclist_leave(reclist);
 358     xfree(idfvec);
 359 }
 360
 361 /*
 362  * Local variables:
 363  * c-basic-offset: 4
 364  * c-file-style: "Stroustrup"
 365  * indent-tabs-mode: nil
 366  * End:
 367  * vim: shiftwidth=4 tabstop=8 expandtab
 368  */
 369