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;
44 struct word_entry *next;
47 static int word_entry_match(struct word_entry *entries, const char *norm_str,
48 const char *rank, int *mult)
50 for (; entries; entries = entries->next)
52 if (!strcmp(norm_str, entries->norm_str))
56 sscanf(rank, "%d%n", mult, &no_read);
60 if (no_read > 0 && (cp = strchr(rank, ' ')))
62 if ((cp - rank) == strlen(entries->ccl_field) &&
63 memcmp(entries->ccl_field, rank, cp - rank) == 0)
66 return entries->termno;
72 void relevance_countwords(struct relevance *r, struct record_cluster *cluster,
73 const char *words, const char *rank,
76 int *mult = cluster->term_frequency_vec_tmp;
79 pp2_charset_token_first(r->prt, words, 0);
80 for (i = 1; i < r->vec_len; i++)
84 while ((norm_str = pp2_charset_token_next(r->prt)))
87 int res = word_entry_match(r->entries, norm_str, rank, &local_mult);
90 assert(res < r->vec_len);
91 mult[res] += local_mult;
96 for (i = 1; i < r->vec_len; i++)
98 if (length > 0) /* only add if non-empty */
99 cluster->term_frequency_vecf[i] += (double) mult[i] / length;
100 cluster->term_frequency_vec[i] += mult[i];
103 cluster->term_frequency_vec[0] += length;
106 static void pull_terms(struct relevance *res, struct ccl_rpn_node *n)
119 pull_terms(res, n->u.p[0]);
120 pull_terms(res, n->u.p[1]);
123 nmem_strsplit(res->nmem, " ", n->u.t.term, &words, &numwords);
124 for (i = 0; i < numwords; i++)
126 const char *norm_str;
128 ccl_field = nmem_strdup_null(res->nmem, n->u.t.qual);
130 pp2_charset_token_first(res->prt, words[i], 0);
131 while ((norm_str = pp2_charset_token_next(res->prt)))
133 struct word_entry **e = &res->entries;
136 *e = nmem_malloc(res->nmem, sizeof(**e));
137 (*e)->norm_str = nmem_strdup(res->nmem, norm_str);
138 (*e)->ccl_field = ccl_field;
139 (*e)->termno = res->vec_len++;
149 struct relevance *relevance_create_ccl(pp2_charset_fact_t pft,
150 struct ccl_rpn_node *query)
152 NMEM nmem = nmem_create();
153 struct relevance *res = nmem_malloc(nmem, sizeof(*res));
159 res->prt = pp2_charset_token_create(pft, "relevance");
161 pull_terms(res, query);
163 res->doc_frequency_vec = nmem_malloc(nmem, res->vec_len * sizeof(int));
164 for (i = 0; i < res->vec_len; i++)
165 res->doc_frequency_vec[i] = 0;
169 void relevance_destroy(struct relevance **rp)
173 pp2_charset_token_destroy((*rp)->prt);
174 nmem_destroy((*rp)->nmem);
179 void relevance_newrec(struct relevance *r, struct record_cluster *rec)
181 if (!rec->term_frequency_vec)
185 // term frequency [1,..] . [0] is total length of all fields
186 rec->term_frequency_vec =
188 r->vec_len * sizeof(*rec->term_frequency_vec));
189 for (i = 0; i < r->vec_len; i++)
190 rec->term_frequency_vec[i] = 0;
192 // term frequency divided by length of field [1,...]
193 rec->term_frequency_vecf =
195 r->vec_len * sizeof(*rec->term_frequency_vecf));
196 for (i = 0; i < r->vec_len; i++)
197 rec->term_frequency_vecf[i] = 0.0;
199 // for relevance_countwords (so we don't have to xmalloc/xfree)
200 rec->term_frequency_vec_tmp =
202 r->vec_len * sizeof(*rec->term_frequency_vec_tmp));
206 void relevance_donerecord(struct relevance *r, struct record_cluster *cluster)
210 for (i = 1; i < r->vec_len; i++)
211 if (cluster->term_frequency_vec[i] > 0)
212 r->doc_frequency_vec[i]++;
214 r->doc_frequency_vec[0]++;
217 // Prepare for a relevance-sorted read
218 void relevance_prepare_read(struct relevance *rel, struct reclist *reclist)
221 float *idfvec = xmalloc(rel->vec_len * sizeof(float));
223 reclist_enter(reclist);
224 // Calculate document frequency vector for each term.
225 for (i = 1; i < rel->vec_len; i++)
227 if (!rel->doc_frequency_vec[i])
231 // This conditional may be terribly wrong
232 // It was there to address the situation where vec[0] == vec[i]
233 // which leads to idfvec[i] == 0... not sure about this
234 // Traditional TF-IDF may assume that a word that occurs in every
235 // record is irrelevant, but this is actually something we will
237 if ((idfvec[i] = log((float) rel->doc_frequency_vec[0] /
238 rel->doc_frequency_vec[i])) < 0.0000001)
242 // Calculate relevance for each document
247 struct record_cluster *rec = reclist_read_record(reclist);
250 for (t = 1; t < rel->vec_len; t++)
252 float termfreq = (float) rec->term_frequency_vecf[t];
253 relevance += 100000 * (termfreq * idfvec[t] + 0.0000005);
255 rec->relevance_score = relevance;
257 reclist_leave(reclist);
264 * c-file-style: "Stroustrup"
265 * indent-tabs-mode: nil
267 * vim: shiftwidth=4 tabstop=8 expandtab