1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
|
// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include "text_similarity_feature.h"
#include <vespa/searchlib/fef/itermdata.h>
#include <vespa/searchlib/fef/featurenamebuilder.h>
#include <vespa/vespalib/util/stash.h>
namespace search::features {
using CollectionType = fef::FieldInfo::CollectionType;
namespace {
struct Term {
fef::TermFieldHandle handle;
int weight;
Term(fef::TermFieldHandle handle_in, int weight_in)
: handle(handle_in), weight(weight_in) {}
};
struct State {
uint32_t field_length;
uint32_t matched_terms;
int sum_term_weight;
uint32_t last_pos;
double sum_proximity_score;
uint32_t last_idx;
uint32_t num_in_order;
State(uint32_t length, uint32_t first_pos, int32_t first_weight, uint32_t first_idx)
: field_length(length),
matched_terms(1), sum_term_weight(first_weight),
last_pos(first_pos), sum_proximity_score(0.0),
last_idx(first_idx), num_in_order(0) {}
double proximity_score(uint32_t dist) {
return (dist > 8) ? 0 : (1.0 - (((dist-1)/8.0) * ((dist-1)/8.0)));
}
bool want_match(uint32_t pos) {
return (pos > last_pos);
}
void addMatch(uint32_t pos, int32_t weight, uint32_t idx) {
sum_proximity_score += proximity_score(pos - last_pos);
num_in_order += (idx > last_idx) ? 1 : 0;
last_pos = pos;
last_idx = idx;
++matched_terms;
sum_term_weight += weight;
}
void calculateScore(size_t num_query_terms, int total_term_weight,
double &score_out,
double &proximity_out, double &order_out,
double &query_coverage_out, double &field_coverage_out)
{
double matches = std::min(field_length, matched_terms);
if (matches < 2) {
proximity_out = proximity_score(field_length);
order_out = (num_query_terms == 1) ? 1.0 : 0.0;
} else {
proximity_out = sum_proximity_score / (matches - 1);
order_out = num_in_order / (double) (matches - 1);
}
query_coverage_out = sum_term_weight / (double) total_term_weight;
field_coverage_out = matches / (double) field_length;
score_out = (0.35 * proximity_out) + (0.15 * order_out)
+ (0.30 * query_coverage_out) + (0.20 * field_coverage_out);
}
};
} // namespace search::features::<unnamed>
//-----------------------------------------------------------------------------
TextSimilarityExecutor::TextSimilarityExecutor(const fef::IQueryEnvironment &env,
uint32_t field_id)
: _handles(),
_weights(),
_total_term_weight(0),
_queue(),
_md(nullptr)
{
std::vector<Term> terms;
for (uint32_t i = 0; i < env.getNumTerms(); ++i) {
const fef::ITermData *termData = env.getTerm(i);
if (termData->getWeight().percent() != 0) { // only consider query terms with contribution
using FRA = fef::ITermFieldRangeAdapter;
for (FRA iter(*termData); iter.valid(); iter.next()) {
const fef::ITermFieldData &tfd = iter.get();
if (tfd.getFieldId() == field_id) {
int term_weight = termData->getWeight().percent();
_total_term_weight += term_weight;
terms.push_back(Term(tfd.getHandle(), term_weight));
}
}
}
}
_handles.reserve(terms.size());
_weights.reserve(terms.size());
for (size_t i = 0; i < terms.size(); ++i) {
_handles.push_back(terms[i].handle);
_weights.push_back(terms[i].weight);
}
}
void
TextSimilarityExecutor::execute(uint32_t docId)
{
for (size_t i = 0; i < _handles.size(); ++i) {
const fef::TermFieldMatchData *tfmd = _md->resolveTermField(_handles[i]);
if (tfmd->getDocId() == docId) {
Item item(i, tfmd->begin(), tfmd->end());
if (item.pos != item.end) {
_queue.push(item);
}
}
}
if (_queue.empty()) {
outputs().set_number(0, 0.0);
outputs().set_number(1, 0.0);
outputs().set_number(2, 0.0);
outputs().set_number(3, 0.0);
outputs().set_number(4, 0.0);
return;
}
const Item &first = _queue.front();
State state(first.pos->getElementLen(),
first.pos->getPosition(),
_weights[first.idx],
first.idx);
_queue.pop_front();
while (!_queue.empty()) {
Item &item = _queue.front();
if (state.want_match(item.pos->getPosition())) {
state.addMatch(item.pos->getPosition(),
_weights[item.idx],
item.idx);
_queue.pop_front();
} else {
++item.pos;
if (item.pos == item.end) {
_queue.pop_front();
} else {
_queue.adjust();
}
}
}
state.calculateScore(_handles.size(), _total_term_weight,
*outputs().get_number_ptr(0),
*outputs().get_number_ptr(1),
*outputs().get_number_ptr(2),
*outputs().get_number_ptr(3),
*outputs().get_number_ptr(4));
}
void
TextSimilarityExecutor::handle_bind_match_data(const fef::MatchData &md)
{
_md = &md;
}
//-----------------------------------------------------------------------------
const vespalib::string TextSimilarityBlueprint::score_output("score");
const vespalib::string TextSimilarityBlueprint::proximity_output("proximity");
const vespalib::string TextSimilarityBlueprint::order_output("order");
const vespalib::string TextSimilarityBlueprint::query_coverage_output("queryCoverage");
const vespalib::string TextSimilarityBlueprint::field_coverage_output("fieldCoverage");
TextSimilarityBlueprint::TextSimilarityBlueprint()
: Blueprint("textSimilarity"), _field_id(fef::IllegalHandle) {}
TextSimilarityBlueprint::~TextSimilarityBlueprint() = default;
void
TextSimilarityBlueprint::visitDumpFeatures(const fef::IIndexEnvironment &env,
fef::IDumpFeatureVisitor &visitor) const
{
for (uint32_t i = 0; i < env.getNumFields(); ++i) {
const fef::FieldInfo &field = *env.getField(i);
if (field.type() == fef::FieldType::INDEX) {
if (!field.isFilter() && field.collection() == CollectionType::SINGLE) {
fef::FeatureNameBuilder fnb;
fnb.baseName(getBaseName()).parameter(field.name());
visitor.visitDumpFeature(fnb.output(score_output).buildName());
visitor.visitDumpFeature(fnb.output(proximity_output).buildName());
visitor.visitDumpFeature(fnb.output(order_output).buildName());
visitor.visitDumpFeature(fnb.output(query_coverage_output).buildName());
visitor.visitDumpFeature(fnb.output(field_coverage_output).buildName());
}
}
}
}
fef::Blueprint::UP
TextSimilarityBlueprint::createInstance() const
{
return std::make_unique<TextSimilarityBlueprint>();
}
bool
TextSimilarityBlueprint::setup(const fef::IIndexEnvironment &env,
const fef::ParameterList ¶ms)
{
const fef::FieldInfo *field = params[0].asField();
_field_id = field->id();
describeOutput(score_output, "default normalized combination of other outputs");
describeOutput(proximity_output, "normalized match proximity score");
describeOutput(order_output, "normalized match order score");
describeOutput(query_coverage_output, "normalized query match coverage");
describeOutput(field_coverage_output, "normalized field match coverage");
env.hintFieldAccess(field->id());
return true;
}
fef::FeatureExecutor &
TextSimilarityBlueprint::createExecutor(const fef::IQueryEnvironment &env, vespalib::Stash &stash) const
{
return stash.create<TextSimilarityExecutor>(env, _field_id);
}
}
|
