blob: 77f71782e315a7f12be4afbb75ca91847c8a3bbf [file] [log] [blame]
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF 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.apache.lucene.search.similarities;
import java.util.ArrayList;
import java.util.List;
import org.apache.lucene.search.Explanation;
/**
* Axiomatic approaches for IR. From Hui Fang and Chengxiang Zhai 2005. An Exploration of Axiomatic
* Approaches to Information Retrieval. In Proceedings of the 28th annual international ACM SIGIR
* conference on Research and development in information retrieval (SIGIR '05). ACM, New York, NY,
* USA, 480-487.
*
* <p>There are a family of models. All of them are based on BM25, Pivoted Document Length
* Normalization and Language model with Dirichlet prior. Some components (e.g. Term Frequency,
* Inverted Document Frequency) in the original models are modified so that they follow some
* axiomatic constraints.
*
* @lucene.experimental
*/
public abstract class Axiomatic extends SimilarityBase {
/** hyperparam for the growth function */
protected final float s;
/** hyperparam for the primitive weighthing function */
protected final float k;
/** the query length */
protected final int queryLen;
/**
* Constructor setting all Axiomatic hyperparameters
*
* @param s hyperparam for the growth function
* @param queryLen the query length
* @param k hyperparam for the primitive weighting function
*/
public Axiomatic(float s, int queryLen, float k) {
if (Float.isFinite(s) == false || Float.isNaN(s) || s < 0 || s > 1) {
throw new IllegalArgumentException("illegal s value: " + s + ", must be between 0 and 1");
}
if (Float.isFinite(k) == false || Float.isNaN(k) || k < 0 || k > 1) {
throw new IllegalArgumentException("illegal k value: " + k + ", must be between 0 and 1");
}
if (queryLen < 0 || queryLen > Integer.MAX_VALUE) {
throw new IllegalArgumentException(
"illegal query length value: "
+ queryLen
+ ", must be larger 0 and smaller than MAX_INT");
}
this.s = s;
this.queryLen = queryLen;
this.k = k;
}
/**
* Constructor setting only s, letting k and queryLen to default
*
* @param s hyperparam for the growth function
*/
public Axiomatic(float s) {
this(s, 1, 0.35f);
}
/**
* Constructor setting s and queryLen, letting k to default
*
* @param s hyperparam for the growth function
* @param queryLen the query length
*/
public Axiomatic(float s, int queryLen) {
this(s, queryLen, 0.35f);
}
/** Default constructor */
public Axiomatic() {
this(0.25f, 1, 0.35f);
}
@Override
public double score(BasicStats stats, double freq, double docLen) {
double score =
tf(stats, freq, docLen)
* ln(stats, freq, docLen)
* tfln(stats, freq, docLen)
* idf(stats, freq, docLen)
- gamma(stats, freq, docLen);
score *= stats.boost;
// AxiomaticF3 similarities might produce negative scores due to their gamma component
return Math.max(0, score);
}
@Override
protected Explanation explain(BasicStats stats, Explanation freq, double docLen) {
List<Explanation> subs = new ArrayList<>();
double f = freq.getValue().doubleValue();
explain(subs, stats, f, docLen);
double score =
tf(stats, f, docLen) * ln(stats, f, docLen) * tfln(stats, f, docLen) * idf(stats, f, docLen)
- gamma(stats, f, docLen);
Explanation explanation =
Explanation.match(
(float) score,
"score("
+ getClass().getSimpleName()
+ ", freq="
+ freq.getValue()
+ "), computed from:",
subs);
if (stats.boost != 1f) {
explanation =
Explanation.match(
(float) (score * stats.boost),
"Boosted score, computed as (score * boost) from:",
explanation,
Explanation.match((float) stats.boost, "Query boost"));
}
if (score < 0) {
explanation =
Explanation.match(0, "max of:", Explanation.match(0, "Minimum legal score"), explanation);
}
return explanation;
}
@Override
protected void explain(List<Explanation> subs, BasicStats stats, double freq, double docLen) {
if (stats.getBoost() != 1.0d) {
subs.add(Explanation.match((float) stats.getBoost(), "boost, query boost"));
}
subs.add(Explanation.match(this.k, "k, hyperparam for the primitive weighting function"));
subs.add(Explanation.match(this.s, "s, hyperparam for the growth function"));
subs.add(Explanation.match(this.queryLen, "queryLen, query length"));
subs.add(tfExplain(stats, freq, docLen));
subs.add(lnExplain(stats, freq, docLen));
subs.add(tflnExplain(stats, freq, docLen));
subs.add(idfExplain(stats, freq, docLen));
subs.add(Explanation.match((float) gamma(stats, freq, docLen), "gamma"));
super.explain(subs, stats, freq, docLen);
}
/** Name of the axiomatic method. */
@Override
public abstract String toString();
/** compute the term frequency component */
protected abstract double tf(BasicStats stats, double freq, double docLen);
/** compute the document length component */
protected abstract double ln(BasicStats stats, double freq, double docLen);
/** compute the mixed term frequency and document length component */
protected abstract double tfln(BasicStats stats, double freq, double docLen);
/** compute the inverted document frequency component */
protected abstract double idf(BasicStats stats, double freq, double docLen);
/** compute the gamma component (only for F3EXp and F3LOG) */
protected abstract double gamma(BasicStats stats, double freq, double docLen);
/**
* Explain the score of the term frequency component for a single document
*
* @param stats the corpus level statistics
* @param freq number of occurrences of term in the document
* @param docLen the document length
* @return Explanation of how the tf component was computed
*/
protected abstract Explanation tfExplain(BasicStats stats, double freq, double docLen);
/**
* Explain the score of the document length component for a single document
*
* @param stats the corpus level statistics
* @param freq number of occurrences of term in the document
* @param docLen the document length
* @return Explanation of how the ln component was computed
*/
protected abstract Explanation lnExplain(BasicStats stats, double freq, double docLen);
/**
* Explain the score of the mixed term frequency and document length component for a single
* document
*
* @param stats the corpus level statistics
* @param freq number of occurrences of term in the document
* @param docLen the document length
* @return Explanation of how the tfln component was computed
*/
protected abstract Explanation tflnExplain(BasicStats stats, double freq, double docLen);
/**
* Explain the score of the inverted document frequency component for a single document
*
* @param stats the corpus level statistics
* @param freq number of occurrences of term in the document
* @param docLen the document length
* @return Explanation of how the idf component was computed
*/
protected abstract Explanation idfExplain(BasicStats stats, double freq, double docLen);
}