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;
35 struct word_entry *entries;
36 pp2_charset_token_t prt;
43 const char *display_str;
46 struct word_entry *next;
49 static int word_entry_match(struct word_entry *entries, const char *norm_str,
50 const char *rank, int *mult)
52 for (; entries; entries = entries->next)
54 if (*norm_str && !strcmp(norm_str, entries->norm_str))
58 sscanf(rank, "%d%n", mult, &no_read);
62 if (no_read > 0 && (cp = strchr(rank, ' ')))
64 if ((cp - rank) == strlen(entries->ccl_field) &&
65 memcmp(entries->ccl_field, rank, cp - rank) == 0)
68 return entries->termno;
74 void relevance_countwords(struct relevance *r, struct record_cluster *cluster,
75 const char *words, const char *rank,
78 int *mult = cluster->term_frequency_vec_tmp;
81 struct word_entry *e = r->entries;
82 WRBUF w = cluster->relevance_explain1;
84 pp2_charset_token_first(r->prt, words, 0);
85 for (i = 1; i < r->vec_len; i++)
89 while ((norm_str = pp2_charset_token_next(r->prt)))
92 int res = word_entry_match(r->entries, norm_str, rank, &local_mult);
95 assert(res < r->vec_len);
96 mult[res] += local_mult;
101 for (i = 1; i < r->vec_len; i++)
103 if (length > 0 && mult[i] > 0) /* only add if non-empty */
105 wrbuf_printf(w, "%s: field=%s vecf[%d] += mult(%d) / length(%d);\n",
106 e->display_str, name, i, mult[i], length);
107 cluster->term_frequency_vecf[i] += (double) mult[i] / length;
109 cluster->term_frequency_vec[i] += mult[i];
113 cluster->term_frequency_vec[0] += length;
116 static void pull_terms(struct relevance *res, struct ccl_rpn_node *n)
129 pull_terms(res, n->u.p[0]);
130 pull_terms(res, n->u.p[1]);
133 nmem_strsplit(res->nmem, " ", n->u.t.term, &words, &numwords);
134 for (i = 0; i < numwords; i++)
136 const char *norm_str;
138 ccl_field = nmem_strdup_null(res->nmem, n->u.t.qual);
140 pp2_charset_token_first(res->prt, words[i], 0);
141 while ((norm_str = pp2_charset_token_next(res->prt)))
143 struct word_entry **e = &res->entries;
146 *e = nmem_malloc(res->nmem, sizeof(**e));
147 (*e)->norm_str = nmem_strdup(res->nmem, norm_str);
148 (*e)->ccl_field = ccl_field;
149 (*e)->termno = res->vec_len++;
150 (*e)->display_str = nmem_strdup(res->nmem, words[i]);
160 struct relevance *relevance_create_ccl(pp2_charset_fact_t pft,
161 struct ccl_rpn_node *query,
164 NMEM nmem = nmem_create();
165 struct relevance *res = nmem_malloc(nmem, sizeof(*res));
171 res->rank_cluster = rank_cluster;
172 res->prt = pp2_charset_token_create(pft, "relevance");
174 pull_terms(res, query);
176 res->doc_frequency_vec = nmem_malloc(nmem, res->vec_len * sizeof(int));
177 for (i = 0; i < res->vec_len; i++)
178 res->doc_frequency_vec[i] = 0;
182 void relevance_destroy(struct relevance **rp)
186 pp2_charset_token_destroy((*rp)->prt);
187 nmem_destroy((*rp)->nmem);
192 void relevance_newrec(struct relevance *r, struct record_cluster *rec)
194 if (!rec->term_frequency_vec)
198 // term frequency [1,..] . [0] is total length of all fields
199 rec->term_frequency_vec =
201 r->vec_len * sizeof(*rec->term_frequency_vec));
202 for (i = 0; i < r->vec_len; i++)
203 rec->term_frequency_vec[i] = 0;
205 // term frequency divided by length of field [1,...]
206 rec->term_frequency_vecf =
208 r->vec_len * sizeof(*rec->term_frequency_vecf));
209 for (i = 0; i < r->vec_len; i++)
210 rec->term_frequency_vecf[i] = 0.0;
212 // for relevance_countwords (so we don't have to xmalloc/xfree)
213 rec->term_frequency_vec_tmp =
215 r->vec_len * sizeof(*rec->term_frequency_vec_tmp));
219 void relevance_donerecord(struct relevance *r, struct record_cluster *cluster)
223 for (i = 1; i < r->vec_len; i++)
224 if (cluster->term_frequency_vec[i] > 0)
225 r->doc_frequency_vec[i]++;
227 r->doc_frequency_vec[0]++;
230 // Prepare for a relevance-sorted read
231 void relevance_prepare_read(struct relevance *rel, struct reclist *reclist)
234 float *idfvec = xmalloc(rel->vec_len * sizeof(float));
236 reclist_enter(reclist);
237 // Calculate document frequency vector for each term.
238 for (i = 1; i < rel->vec_len; i++)
240 if (!rel->doc_frequency_vec[i])
244 /* add one to nominator idf(t,D) to ensure a value > 0 */
245 idfvec[i] = log((float) (1 + rel->doc_frequency_vec[0]) /
246 rel->doc_frequency_vec[i]);
249 // Calculate relevance for each document
254 struct word_entry *e = rel->entries;
255 struct record_cluster *rec = reclist_read_record(reclist);
258 w = rec->relevance_explain2;
260 for (i = 1; i < rel->vec_len; i++)
262 float termfreq = (float) rec->term_frequency_vecf[i];
263 int add = 100000 * termfreq * idfvec[i];
265 wrbuf_printf(w, "idf[%d] = log(((1 + total(%d))/termoccur(%d));\n",
266 i, rel->doc_frequency_vec[0],
267 rel->doc_frequency_vec[i]);
268 wrbuf_printf(w, "%s: relevance += 100000 * vecf[%d](%f) * "
269 "idf[%d](%f) (%d);\n",
270 e->display_str, i, termfreq, i, idfvec[i], add);
274 if (!rel->rank_cluster)
276 struct record *record;
277 int cluster_size = 0;
279 for (record = rec->records; record; record = record->next)
282 wrbuf_printf(w, "score = relevance(%d)/cluster_size(%d);\n",
283 relevance, cluster_size);
284 relevance /= cluster_size;
288 wrbuf_printf(w, "score = relevance(%d);\n", relevance);
290 rec->relevance_score = relevance;
292 reclist_leave(reclist);
299 * c-file-style: "Stroustrup"
300 * indent-tabs-mode: nil
302 * vim: shiftwidth=4 tabstop=8 expandtab