1 /* This file is part of Pazpar2.
2 Copyright (C) 2006-2013 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"
32 #define log2(x) (log(x)/log(2))
37 int *doc_frequency_vec;
38 int *term_frequency_vec_tmp;
41 struct word_entry *entries;
42 pp2_charset_token_t prt;
52 const char *display_str;
55 struct word_entry *next;
58 static struct word_entry *word_entry_match(struct relevance *r,
60 const char *rank, int *weight)
63 struct word_entry *entries = r->entries;
64 for (; entries; entries = entries->next, i++)
66 if (*norm_str && !strcmp(norm_str, entries->norm_str))
70 sscanf(rank, "%d%n", weight, &no_read);
74 if (no_read > 0 && (cp = strchr(rank, ' ')))
76 if ((cp - rank) == strlen(entries->ccl_field) &&
77 memcmp(entries->ccl_field, rank, cp - rank) == 0)
78 *weight = atoi(cp + 1);
86 int relevance_snippet(struct relevance *r,
87 const char *words, const char *name,
94 pp2_charset_token_first(r->prt, words, 0);
95 while ((norm_str = pp2_charset_token_next(r->prt)))
97 size_t org_start, org_len;
98 struct word_entry *entries = r->entries;
101 pp2_get_org(r->prt, &org_start, &org_len);
102 for (; entries; entries = entries->next, i++)
104 if (*norm_str && !strcmp(norm_str, entries->norm_str))
112 wrbuf_puts(w_snippet, "<match>");
121 wrbuf_puts(w_snippet, "</match>");
124 wrbuf_xmlputs_n(w_snippet, words + org_start, org_len);
127 wrbuf_puts(w_snippet, "</match>");
130 yaz_log(YLOG_DEBUG, "SNIPPET match: %s", wrbuf_cstr(w_snippet));
135 void relevance_countwords(struct relevance *r, struct record_cluster *cluster,
136 const char *words, const char *rank,
139 int *w = r->term_frequency_vec_tmp;
140 const char *norm_str;
142 double lead_decay = r->lead_decay;
143 struct word_entry *e;
144 WRBUF wr = cluster->relevance_explain1;
145 int printed_about_field = 0;
147 pp2_charset_token_first(r->prt, words, 0);
148 for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
155 while ((norm_str = pp2_charset_token_next(r->prt)))
157 int local_weight = 0;
158 e = word_entry_match(r, norm_str, rank, &local_weight);
164 if (!printed_about_field)
166 printed_about_field = 1;
167 wrbuf_printf(wr, "field=%s content=", name);
168 if (strlen(words) > 50)
170 wrbuf_xmlputs_n(wr, words, 49);
171 wrbuf_puts(wr, " ...");
174 wrbuf_xmlputs(wr, words);
175 wrbuf_puts(wr, ";\n");
177 assert(res < r->vec_len);
178 w[res] += local_weight / (1 + log2(1 + lead_decay * length));
179 wrbuf_printf(wr, "%s: w[%d] += w(%d) / "
180 "(1+log2(1+lead_decay(%f) * length(%d)));\n",
181 e->display_str, res, local_weight, lead_decay, length);
183 if (j > 0 && r->term_pos[j])
185 int d = length + 1 - r->term_pos[j];
186 wrbuf_printf(wr, "%s: w[%d] += w[%d](%d) * follow(%f) / "
188 e->display_str, res, res, w[res],
189 r->follow_factor, d);
190 w[res] += w[res] * r->follow_factor / (1 + log2(d));
192 for (j = 0; j < r->vec_len; j++)
193 r->term_pos[j] = j < res ? 0 : length + 1;
198 for (e = r->entries, i = 1; i < r->vec_len; i++, e = e->next)
200 if (length == 0 || w[i] == 0)
202 wrbuf_printf(wr, "%s: tf[%d] += w[%d](%d)", e->display_str, i, i, w[i]);
203 switch (r->length_divide)
206 cluster->term_frequency_vecf[i] += (double) w[i];
209 wrbuf_printf(wr, " / log2(1+length(%d))", length);
210 cluster->term_frequency_vecf[i] +=
211 (double) w[i] / log2(1 + length);
214 wrbuf_printf(wr, " / length(%d)", length);
215 cluster->term_frequency_vecf[i] += (double) w[i] / length;
217 cluster->term_frequency_vec[i] += w[i];
218 wrbuf_printf(wr, " (%f);\n", cluster->term_frequency_vecf[i]);
221 cluster->term_frequency_vec[0] += length;
224 static void pull_terms(struct relevance *res, struct ccl_rpn_node *n)
237 pull_terms(res, n->u.p[0]);
238 pull_terms(res, n->u.p[1]);
241 nmem_strsplit(res->nmem, " ", n->u.t.term, &words, &numwords);
242 for (i = 0; i < numwords; i++)
244 const char *norm_str;
246 ccl_field = nmem_strdup_null(res->nmem, n->u.t.qual);
248 pp2_charset_token_first(res->prt, words[i], 0);
249 while ((norm_str = pp2_charset_token_next(res->prt)))
251 struct word_entry **e = &res->entries;
254 *e = nmem_malloc(res->nmem, sizeof(**e));
255 (*e)->norm_str = nmem_strdup(res->nmem, norm_str);
256 (*e)->ccl_field = ccl_field;
257 (*e)->termno = res->vec_len++;
258 (*e)->display_str = nmem_strdup(res->nmem, words[i]);
267 void relevance_clear(struct relevance *r)
272 for (i = 0; i < r->vec_len; i++)
273 r->doc_frequency_vec[i] = 0;
277 struct relevance *relevance_create_ccl(pp2_charset_fact_t pft,
278 struct ccl_rpn_node *query,
280 double follow_factor, double lead_decay,
283 NMEM nmem = nmem_create();
284 struct relevance *res = nmem_malloc(nmem, sizeof(*res));
289 res->rank_cluster = rank_cluster;
290 res->follow_factor = follow_factor;
291 res->lead_decay = lead_decay;
292 res->length_divide = length_divide;
293 res->prt = pp2_charset_token_create(pft, "relevance");
295 pull_terms(res, query);
297 res->doc_frequency_vec = nmem_malloc(nmem, res->vec_len * sizeof(int));
300 res->term_frequency_vec_tmp =
301 nmem_malloc(res->nmem,
302 res->vec_len * sizeof(*res->term_frequency_vec_tmp));
305 nmem_malloc(res->nmem, res->vec_len * sizeof(*res->term_pos));
307 relevance_clear(res);
311 void relevance_destroy(struct relevance **rp)
315 pp2_charset_token_destroy((*rp)->prt);
316 nmem_destroy((*rp)->nmem);
321 void relevance_newrec(struct relevance *r, struct record_cluster *rec)
323 if (!rec->term_frequency_vec)
327 // term frequency [1,..] . [0] is total length of all fields
328 rec->term_frequency_vec =
330 r->vec_len * sizeof(*rec->term_frequency_vec));
331 for (i = 0; i < r->vec_len; i++)
332 rec->term_frequency_vec[i] = 0;
334 // term frequency divided by length of field [1,...]
335 rec->term_frequency_vecf =
337 r->vec_len * sizeof(*rec->term_frequency_vecf));
338 for (i = 0; i < r->vec_len; i++)
339 rec->term_frequency_vecf[i] = 0.0;
343 void relevance_donerecord(struct relevance *r, struct record_cluster *cluster)
347 for (i = 1; i < r->vec_len; i++)
348 if (cluster->term_frequency_vec[i] > 0)
349 r->doc_frequency_vec[i]++;
351 r->doc_frequency_vec[0]++;
354 // Prepare for a relevance-sorted read
355 void relevance_prepare_read(struct relevance *rel, struct reclist *reclist)
358 float *idfvec = xmalloc(rel->vec_len * sizeof(float));
360 reclist_enter(reclist);
361 // Calculate document frequency vector for each term.
362 for (i = 1; i < rel->vec_len; i++)
364 if (!rel->doc_frequency_vec[i])
368 /* add one to nominator idf(t,D) to ensure a value > 0 */
369 idfvec[i] = log((float) (1 + rel->doc_frequency_vec[0]) /
370 rel->doc_frequency_vec[i]);
373 // Calculate relevance for each document
378 struct word_entry *e = rel->entries;
379 struct record_cluster *rec = reclist_read_record(reclist);
382 w = rec->relevance_explain2;
384 wrbuf_puts(w, "relevance = 0;\n");
385 for (i = 1; i < rel->vec_len; i++)
387 float termfreq = (float) rec->term_frequency_vecf[i];
388 int add = 100000 * termfreq * idfvec[i];
390 wrbuf_printf(w, "idf[%d] = log(((1 + total(%d))/termoccur(%d));\n",
391 i, rel->doc_frequency_vec[0],
392 rel->doc_frequency_vec[i]);
393 wrbuf_printf(w, "%s: relevance += 100000 * tf[%d](%f) * "
394 "idf[%d](%f) (%d);\n",
395 e->display_str, i, termfreq, i, idfvec[i], add);
399 if (!rel->rank_cluster)
401 struct record *record;
402 int cluster_size = 0;
404 for (record = rec->records; record; record = record->next)
407 wrbuf_printf(w, "score = relevance(%d)/cluster_size(%d);\n",
408 relevance, cluster_size);
409 relevance /= cluster_size;
413 wrbuf_printf(w, "score = relevance(%d);\n", relevance);
415 rec->relevance_score = relevance;
417 reclist_leave(reclist);
424 * c-file-style: "Stroustrup"
425 * indent-tabs-mode: nil
427 * vim: shiftwidth=4 tabstop=8 expandtab