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;
37 struct word_entry *entries;
38 pp2_charset_token_t prt;
48 const char *display_str;
51 struct word_entry *next;
54 static struct word_entry *word_entry_match(struct relevance *r,
56 const char *rank, int *mult)
59 struct word_entry *entries = r->entries;
60 for (; entries; entries = entries->next, i++)
62 if (*norm_str && !strcmp(norm_str, entries->norm_str))
66 sscanf(rank, "%d%n", mult, &no_read);
70 if (no_read > 0 && (cp = strchr(rank, ' ')))
72 if ((cp - rank) == strlen(entries->ccl_field) &&
73 memcmp(entries->ccl_field, rank, cp - rank) == 0)
82 void relevance_countwords(struct relevance *r, struct record_cluster *cluster,
83 const char *words, const char *rank,
86 int *mult = r->term_frequency_vec_tmp;
89 double lead_decay = r->lead_decay;
91 WRBUF w = cluster->relevance_explain1;
93 pp2_charset_token_first(r->prt, words, 0);
94 for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
101 while ((norm_str = pp2_charset_token_next(r->prt)))
104 e = word_entry_match(r, norm_str, rank, &local_mult);
110 assert(res < r->vec_len);
111 mult[res] += local_mult / (1 + log2(1 + lead_decay * length));
112 wrbuf_printf(w, "%s: mult[%d] += local_mult(%d) / "
113 "(1+log2(1+lead_decay(%f) * length(%d)));\n",
114 e->display_str, res, local_mult, lead_decay, length);
116 if (j > 0 && r->term_pos[j])
118 int d = length + 1 - r->term_pos[j];
119 mult[res] += mult[res] * r->follow_factor / (1 + log2(d));
120 wrbuf_printf(w, "%s: mult[%d] += mult[%d](%d) * follow(%f) / "
122 e->display_str, res, res, mult[res],
123 r->follow_factor, d);
125 for (j = 0; j < r->vec_len; j++)
126 r->term_pos[j] = j < res ? 0 : length + 1;
131 for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
133 if (length == 0 || mult[i] == 0)
135 wrbuf_printf(w, "%s: field=%s vecf[%d] += mult[%d](%d)",
136 e->display_str, name, i, i, mult[i]);
137 switch (r->length_divide)
140 wrbuf_printf(w, ";\n");
141 cluster->term_frequency_vecf[i] += (double) mult[i];
144 wrbuf_printf(w, " / log2(1+length(%d));\n", length);
145 cluster->term_frequency_vecf[i] +=
146 (double) mult[i] / log2(1 + length);
149 wrbuf_printf(w, " / length(%d);\n", length);
150 cluster->term_frequency_vecf[i] += (double) mult[i] / length;
152 cluster->term_frequency_vec[i] += mult[i];
155 cluster->term_frequency_vec[0] += length;
158 static void pull_terms(struct relevance *res, struct ccl_rpn_node *n)
171 pull_terms(res, n->u.p[0]);
172 pull_terms(res, n->u.p[1]);
175 nmem_strsplit(res->nmem, " ", n->u.t.term, &words, &numwords);
176 for (i = 0; i < numwords; i++)
178 const char *norm_str;
180 ccl_field = nmem_strdup_null(res->nmem, n->u.t.qual);
182 pp2_charset_token_first(res->prt, words[i], 0);
183 while ((norm_str = pp2_charset_token_next(res->prt)))
185 struct word_entry **e = &res->entries;
188 *e = nmem_malloc(res->nmem, sizeof(**e));
189 (*e)->norm_str = nmem_strdup(res->nmem, norm_str);
190 (*e)->ccl_field = ccl_field;
191 (*e)->termno = res->vec_len++;
192 (*e)->display_str = nmem_strdup(res->nmem, words[i]);
202 struct relevance *relevance_create_ccl(pp2_charset_fact_t pft,
203 struct ccl_rpn_node *query,
205 double follow_factor, double lead_decay,
208 NMEM nmem = nmem_create();
209 struct relevance *res = nmem_malloc(nmem, sizeof(*res));
215 res->rank_cluster = rank_cluster;
216 res->follow_factor = follow_factor;
217 res->lead_decay = lead_decay;
218 res->length_divide = length_divide;
219 res->prt = pp2_charset_token_create(pft, "relevance");
221 pull_terms(res, query);
223 res->doc_frequency_vec = nmem_malloc(nmem, res->vec_len * sizeof(int));
224 for (i = 0; i < res->vec_len; i++)
225 res->doc_frequency_vec[i] = 0;
228 res->term_frequency_vec_tmp =
229 nmem_malloc(res->nmem,
230 res->vec_len * sizeof(*res->term_frequency_vec_tmp));
233 nmem_malloc(res->nmem, res->vec_len * sizeof(*res->term_pos));
238 void relevance_destroy(struct relevance **rp)
242 pp2_charset_token_destroy((*rp)->prt);
243 nmem_destroy((*rp)->nmem);
248 void relevance_newrec(struct relevance *r, struct record_cluster *rec)
250 if (!rec->term_frequency_vec)
254 // term frequency [1,..] . [0] is total length of all fields
255 rec->term_frequency_vec =
257 r->vec_len * sizeof(*rec->term_frequency_vec));
258 for (i = 0; i < r->vec_len; i++)
259 rec->term_frequency_vec[i] = 0;
261 // term frequency divided by length of field [1,...]
262 rec->term_frequency_vecf =
264 r->vec_len * sizeof(*rec->term_frequency_vecf));
265 for (i = 0; i < r->vec_len; i++)
266 rec->term_frequency_vecf[i] = 0.0;
270 void relevance_donerecord(struct relevance *r, struct record_cluster *cluster)
274 for (i = 1; i < r->vec_len; i++)
275 if (cluster->term_frequency_vec[i] > 0)
276 r->doc_frequency_vec[i]++;
278 r->doc_frequency_vec[0]++;
281 // Prepare for a relevance-sorted read
282 void relevance_prepare_read(struct relevance *rel, struct reclist *reclist)
285 float *idfvec = xmalloc(rel->vec_len * sizeof(float));
287 reclist_enter(reclist);
288 // Calculate document frequency vector for each term.
289 for (i = 1; i < rel->vec_len; i++)
291 if (!rel->doc_frequency_vec[i])
295 /* add one to nominator idf(t,D) to ensure a value > 0 */
296 idfvec[i] = log((float) (1 + rel->doc_frequency_vec[0]) /
297 rel->doc_frequency_vec[i]);
300 // Calculate relevance for each document
305 struct word_entry *e = rel->entries;
306 struct record_cluster *rec = reclist_read_record(reclist);
309 w = rec->relevance_explain2;
311 for (i = 1; i < rel->vec_len; i++)
313 float termfreq = (float) rec->term_frequency_vecf[i];
314 int add = 100000 * termfreq * idfvec[i];
316 wrbuf_printf(w, "idf[%d] = log(((1 + total(%d))/termoccur(%d));\n",
317 i, rel->doc_frequency_vec[0],
318 rel->doc_frequency_vec[i]);
319 wrbuf_printf(w, "%s: relevance += 100000 * vecf[%d](%f) * "
320 "idf[%d](%f) (%d);\n",
321 e->display_str, i, termfreq, i, idfvec[i], add);
325 if (!rel->rank_cluster)
327 struct record *record;
328 int cluster_size = 0;
330 for (record = rec->records; record; record = record->next)
333 wrbuf_printf(w, "score = relevance(%d)/cluster_size(%d);\n",
334 relevance, cluster_size);
335 relevance /= cluster_size;
339 wrbuf_printf(w, "score = relevance(%d);\n", relevance);
341 rec->relevance_score = relevance;
343 reclist_leave(reclist);
350 * c-file-style: "Stroustrup"
351 * indent-tabs-mode: nil
353 * vim: shiftwidth=4 tabstop=8 expandtab