Add Reciprocal Rank query evaluation metric

This adds a second query evaluation metric alongside precision_at. Reciprocal
Rank is defined as 1/rank, where rank is the position of the first relevant
document in the search result. The results are averaged across all queries
across the sample of queries, according to
https://en.wikipedia.org/wiki/Mean_reciprocal_rank
This commit is contained in:
Christoph Büscher 2016-07-01 16:22:35 +02:00
parent 4162582ee8
commit ad87bacf91
5 changed files with 246 additions and 4 deletions

View File

@ -21,9 +21,11 @@ package org.elasticsearch.index.rankeval;
import java.util.Collection;
/** Returned for each search specification. Summarizes the measured quality metric for this search request
* and adds the document ids found that were in the search result but not annotated in the original request.
* */
/**
* Returned for each search specification. Summarizes the measured quality
* metric for this search request and adds the document ids found that were in
* the search result but not annotated in the original request.
*/
public class EvalQueryQuality {
private double qualityLevel;

View File

@ -58,6 +58,9 @@ public abstract class RankedListQualityMetric implements NamedWriteable {
case PrecisionAtN.NAME:
rc = PrecisionAtN.fromXContent(parser, context);
break;
case ReciprocalRank.NAME:
rc = ReciprocalRank.fromXContent(parser, context);
break;
default:
throw new ParsingException(parser.getTokenLocation(), "[_na] unknown query metric name [{}]", metricName);
}

View File

@ -0,0 +1,98 @@
/*
* Licensed to Elasticsearch under one or more contributor
* license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright
* ownership. Elasticsearch licenses this file to you under
* the Apache License, Version 2.0 (the "License"); you may
* not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package org.elasticsearch.index.rankeval;
import org.elasticsearch.common.ParseFieldMatcherSupplier;
import org.elasticsearch.common.io.stream.StreamOutput;
import org.elasticsearch.common.xcontent.ObjectParser;
import org.elasticsearch.common.xcontent.XContentParser;
import org.elasticsearch.index.rankeval.PrecisionAtN.Rating;
import org.elasticsearch.index.rankeval.PrecisionAtN.RatingMapping;
import org.elasticsearch.search.SearchHit;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import java.util.HashSet;
import java.util.List;
import java.util.Set;
import javax.naming.directory.SearchResult;
/**
* Evaluate reciprocal rank.
* */
public class ReciprocalRank extends RankedListQualityMetric {
public static final String NAME = "reciprocal_rank";
// the rank to use if the result list does not contain any relevant document
// TODO decide on better default or make configurable
private static final int RANK_IF_NOT_FOUND = Integer.MAX_VALUE;
@Override
public String getWriteableName() {
return NAME;
}
/**
* Compute ReciprocalRank based on provided relevant document IDs.
* @return reciprocal Rank for above {@link SearchResult} list.
**/
@Override
public EvalQueryQuality evaluate(SearchHit[] hits, List<RatedDocument> ratedDocs) {
Set<String> relevantDocIds = new HashSet<>();
Set<String> irrelevantDocIds = new HashSet<>();
for (RatedDocument doc : ratedDocs) {
if (Rating.RELEVANT.equals(RatingMapping.mapTo(doc.getRating()))) {
relevantDocIds.add(doc.getDocID());
} else if (Rating.IRRELEVANT.equals(RatingMapping.mapTo(doc.getRating()))) {
irrelevantDocIds.add(doc.getDocID());
}
}
Collection<String> unknownDocIds = new ArrayList<String>();
int firstRelevant = RANK_IF_NOT_FOUND;
boolean found = false;
for (int i = 0; i < hits.length; i++) {
String id = hits[i].getId();
if (relevantDocIds.contains(id) && found == false) {
firstRelevant = i + 1; // add one because rank is not 0-based
found = true;
continue;
} else if (irrelevantDocIds.contains(id) == false) {
unknownDocIds.add(id);
}
}
double reciprocalRank = 1.0d / firstRelevant;
return new EvalQueryQuality(reciprocalRank, unknownDocIds);
}
@Override
public void writeTo(StreamOutput out) throws IOException {
}
private static final ObjectParser<ReciprocalRank, ParseFieldMatcherSupplier> PARSER = new ObjectParser<>(
"reciprocal_rank", () -> new ReciprocalRank());
public static ReciprocalRank fromXContent(XContentParser parser, ParseFieldMatcherSupplier matcher) {
return PARSER.apply(parser, matcher);
}
}

View File

@ -0,0 +1,68 @@
/*
* Licensed to Elasticsearch under one or more contributor
* license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright
* ownership. Elasticsearch licenses this file to you under
* the Apache License, Version 2.0 (the "License"); you may
* not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package org.elasticsearch.action.quality;
import org.elasticsearch.common.text.Text;
import org.elasticsearch.index.rankeval.EvalQueryQuality;
import org.elasticsearch.index.rankeval.PrecisionAtN.Rating;
import org.elasticsearch.index.rankeval.RatedDocument;
import org.elasticsearch.index.rankeval.ReciprocalRank;
import org.elasticsearch.search.SearchHit;
import org.elasticsearch.search.internal.InternalSearchHit;
import org.elasticsearch.test.ESTestCase;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
public class ReciprocalRankTests extends ESTestCase {
public void testEvaluationOneRelevantInResults() {
ReciprocalRank reciprocalRank = new ReciprocalRank();
SearchHit[] hits = new SearchHit[10];
for (int i = 0; i < 10; i++) {
hits[i] = new InternalSearchHit(i, Integer.toString(i), new Text("type"), Collections.emptyMap());
}
List<RatedDocument> ratedDocs = new ArrayList<>();
// mark one of the ten docs relevant
int relevantAt = randomIntBetween(0, 9);
for (int i = 0; i <= 20; i++) {
if (i == relevantAt) {
ratedDocs.add(new RatedDocument(Integer.toString(i), Rating.RELEVANT.ordinal()));
} else {
ratedDocs.add(new RatedDocument(Integer.toString(i), Rating.IRRELEVANT.ordinal()));
}
}
EvalQueryQuality evaluation = reciprocalRank.evaluate(hits, ratedDocs);
assertEquals(1.0 / (relevantAt + 1), evaluation.getQualityLevel(), Double.MIN_VALUE);
}
public void testEvaluationNoRelevantInResults() {
ReciprocalRank reciprocalRank = new ReciprocalRank();
SearchHit[] hits = new SearchHit[10];
for (int i = 0; i < 10; i++) {
hits[i] = new InternalSearchHit(i, Integer.toString(i), new Text("type"), Collections.emptyMap());
}
List<RatedDocument> ratedDocs = new ArrayList<>();
EvalQueryQuality evaluation = reciprocalRank.evaluate(hits, ratedDocs);
assertEquals(1.0 / Integer.MAX_VALUE, evaluation.getQualityLevel(), Double.MIN_VALUE);
}
}

View File

@ -1,6 +1,12 @@
---
"Response format":
- do:
indices.create:
index: foo
body:
settings:
index:
number_of_shards: 1
- do:
index:
index: foo
@ -55,3 +61,68 @@
- match: {rank_eval.quality_level: 1}
- match: {rank_eval.unknown_docs.0.amsterdam_query: [ "doc4"]}
- match: {rank_eval.unknown_docs.1.berlin_query: [ "doc4"]}
---
"Reciprocal Rank":
- do:
indices.create:
index: foo
body:
settings:
index:
number_of_shards: 1
- do:
index:
index: foo
type: bar
id: doc1
body: { "text": "berlin" }
- do:
index:
index: foo
type: bar
id: doc2
body: { "text": "amsterdam" }
- do:
index:
index: foo
type: bar
id: doc3
body: { "text": "amsterdam" }
- do:
index:
index: foo
type: bar
id: doc4
body: { "text": "something about amsterdam and berlin" }
- do:
indices.refresh: {}
- do:
rank_eval:
body: {
"spec_id" : "cities_qa_queries",
"requests" : [
{
"id": "amsterdam_query",
"request": { "query": { "match" : {"text" : "amsterdam" }}},
# doc4 should be returned in third position, so reciprocal rank is 1/3
"ratings": [{ "doc4": 1}]
},
{
"id" : "berlin_query",
"request": { "query": { "match" : { "text" : "berlin" } }, "size" : 10 },
# doc1 should be returned in first position, doc3 in second, so reciprocal rank is 1/2
"ratings": [{"doc4": 1}]
}
],
"metric" : { "reciprocal_rank": {} }
}
- match: {rank_eval.spec_id: "cities_qa_queries"}
# average is (1/3 + 1/2)/2 = 5/12 ~ 0.41666666666666663
- match: {rank_eval.quality_level: 0.41666666666666663}