1 /* This file is part of Pazpar2.
2 Copyright (C) 2006-2012 Index Data
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
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
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
28 #include "relevance.h"
33 int *doc_frequency_vec;
34 int *term_frequency_vec_tmp;
36 struct word_entry *entries;
37 pp2_charset_token_t prt;
47 const char *display_str;
51 struct word_entry *next;
54 static int word_entry_match(struct relevance *r, const char *norm_str,
55 const char *rank, int *mult)
58 struct word_entry *entries = r->entries;
59 for (; entries; entries = entries->next, i++)
61 if (*norm_str && !strcmp(norm_str, entries->norm_str))
63 int extra = r->follow_boost;
64 struct word_entry *e_follow = entries;
67 sscanf(rank, "%d%n", mult, &no_read);
71 if (no_read > 0 && (cp = strchr(rank, ' ')))
73 if ((cp - rank) == strlen(entries->ccl_field) &&
74 memcmp(entries->ccl_field, rank, cp - rank) == 0)
77 (*mult) += entries->follow_boost;
78 while ((e_follow = e_follow->next) != 0 && extra > 0)
80 e_follow->follow_boost = extra--;
82 return entries->termno;
84 entries->follow_boost = 0;
89 void relevance_countwords(struct relevance *r, struct record_cluster *cluster,
90 const char *words, const char *rank,
93 int *mult = r->term_frequency_vec_tmp;
96 int lead_mult = r->lead_boost;
98 WRBUF w = cluster->relevance_explain1;
100 pp2_charset_token_first(r->prt, words, 0);
101 for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
108 while ((norm_str = pp2_charset_token_next(r->prt)))
111 int res = word_entry_match(r, norm_str, rank, &local_mult);
114 assert(res < r->vec_len);
115 mult[res] += local_mult + lead_mult;
122 for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
124 if (length == 0 || mult[i] == 0)
126 wrbuf_printf(w, "%s: field=%s vecf[%d] += mult(%d)",
127 e->display_str, name, i, mult[i]);
128 switch (r->length_divide)
131 wrbuf_printf(w, ";\n");
132 cluster->term_frequency_vecf[i] += (double) mult[i];
135 wrbuf_printf(w, " / log2(1+length(%d));\n", length);
136 cluster->term_frequency_vecf[i] +=
137 (double) mult[i] / log2(1 + length);
140 wrbuf_printf(w, " / length(%d);\n", length);
141 cluster->term_frequency_vecf[i] += (double) mult[i] / length;
143 cluster->term_frequency_vec[i] += mult[i];
146 cluster->term_frequency_vec[0] += length;
149 static void pull_terms(struct relevance *res, struct ccl_rpn_node *n)
162 pull_terms(res, n->u.p[0]);
163 pull_terms(res, n->u.p[1]);
166 nmem_strsplit(res->nmem, " ", n->u.t.term, &words, &numwords);
167 for (i = 0; i < numwords; i++)
169 const char *norm_str;
171 ccl_field = nmem_strdup_null(res->nmem, n->u.t.qual);
173 pp2_charset_token_first(res->prt, words[i], 0);
174 while ((norm_str = pp2_charset_token_next(res->prt)))
176 struct word_entry **e = &res->entries;
179 *e = nmem_malloc(res->nmem, sizeof(**e));
180 (*e)->norm_str = nmem_strdup(res->nmem, norm_str);
181 (*e)->ccl_field = ccl_field;
182 (*e)->termno = res->vec_len++;
183 (*e)->display_str = nmem_strdup(res->nmem, words[i]);
193 struct relevance *relevance_create_ccl(pp2_charset_fact_t pft,
194 struct ccl_rpn_node *query,
196 int follow_boost, int lead_boost,
199 NMEM nmem = nmem_create();
200 struct relevance *res = nmem_malloc(nmem, sizeof(*res));
206 res->rank_cluster = rank_cluster;
207 res->follow_boost = follow_boost;
208 res->lead_boost = lead_boost;
209 res->length_divide = length_divide;
210 res->prt = pp2_charset_token_create(pft, "relevance");
212 pull_terms(res, query);
214 res->doc_frequency_vec = nmem_malloc(nmem, res->vec_len * sizeof(int));
215 for (i = 0; i < res->vec_len; i++)
216 res->doc_frequency_vec[i] = 0;
219 res->term_frequency_vec_tmp =
220 nmem_malloc(res->nmem,
221 res->vec_len * sizeof(*res->term_frequency_vec_tmp));
225 void relevance_destroy(struct relevance **rp)
229 pp2_charset_token_destroy((*rp)->prt);
230 nmem_destroy((*rp)->nmem);
235 void relevance_newrec(struct relevance *r, struct record_cluster *rec)
237 if (!rec->term_frequency_vec)
241 // term frequency [1,..] . [0] is total length of all fields
242 rec->term_frequency_vec =
244 r->vec_len * sizeof(*rec->term_frequency_vec));
245 for (i = 0; i < r->vec_len; i++)
246 rec->term_frequency_vec[i] = 0;
248 // term frequency divided by length of field [1,...]
249 rec->term_frequency_vecf =
251 r->vec_len * sizeof(*rec->term_frequency_vecf));
252 for (i = 0; i < r->vec_len; i++)
253 rec->term_frequency_vecf[i] = 0.0;
257 void relevance_donerecord(struct relevance *r, struct record_cluster *cluster)
261 for (i = 1; i < r->vec_len; i++)
262 if (cluster->term_frequency_vec[i] > 0)
263 r->doc_frequency_vec[i]++;
265 r->doc_frequency_vec[0]++;
268 // Prepare for a relevance-sorted read
269 void relevance_prepare_read(struct relevance *rel, struct reclist *reclist)
272 float *idfvec = xmalloc(rel->vec_len * sizeof(float));
274 reclist_enter(reclist);
275 // Calculate document frequency vector for each term.
276 for (i = 1; i < rel->vec_len; i++)
278 if (!rel->doc_frequency_vec[i])
282 /* add one to nominator idf(t,D) to ensure a value > 0 */
283 idfvec[i] = log((float) (1 + rel->doc_frequency_vec[0]) /
284 rel->doc_frequency_vec[i]);
287 // Calculate relevance for each document
292 struct word_entry *e = rel->entries;
293 struct record_cluster *rec = reclist_read_record(reclist);
296 w = rec->relevance_explain2;
298 for (i = 1; i < rel->vec_len; i++)
300 float termfreq = (float) rec->term_frequency_vecf[i];
301 int add = 100000 * termfreq * idfvec[i];
303 wrbuf_printf(w, "idf[%d] = log(((1 + total(%d))/termoccur(%d));\n",
304 i, rel->doc_frequency_vec[0],
305 rel->doc_frequency_vec[i]);
306 wrbuf_printf(w, "%s: relevance += 100000 * vecf[%d](%f) * "
307 "idf[%d](%f) (%d);\n",
308 e->display_str, i, termfreq, i, idfvec[i], add);
312 if (!rel->rank_cluster)
314 struct record *record;
315 int cluster_size = 0;
317 for (record = rec->records; record; record = record->next)
320 wrbuf_printf(w, "score = relevance(%d)/cluster_size(%d);\n",
321 relevance, cluster_size);
322 relevance /= cluster_size;
326 wrbuf_printf(w, "score = relevance(%d);\n", relevance);
328 rec->relevance_score = relevance;
330 reclist_leave(reclist);
337 * c-file-style: "Stroustrup"
338 * indent-tabs-mode: nil
340 * vim: shiftwidth=4 tabstop=8 expandtab