我想为用户提供最相关,最好的结果。例如,我要奖励标题,说明,所附照片等较大的记录。对于上下文:记录是自行车路线,路线点(坐标)和元数据(例如照片,评论等)。
现在,我已经使用索引了这些记录Hibernate,然后使用Lucenein 在索引中搜索Hibernate Search。为了对结果进行评分,我根据文档属性构建查询,并boostedTo()在中使用增强查询should BooleanJunction clause:
Hibernate
Lucene
Hibernate Search
boostedTo()
should
BooleanJunction clause
bj.should(qb.range().onField("descriptionLength").above(3000).createQuery()).boostedTo(3.0f); bj.should(qb.range().onField("views.views").above(5000).createQuery()).boostedTo(3.0f); bj.should(qb.range().onField("nameLength").above(20).createQuery()).boostedTo(1.0f); bj.should(qb.range().onField("picturesLength").above(0).createQuery()).boostedTo(5.0f); bj.should(qb.keyword().onField("routePoints.poi.participant").matching("true").createQuery()).boostedTo(10.0f);
为了尝试禁用Lucene的评分,我重写了DefaultSimilarity该类,将所有比较设置为1.0f,并通过Hibernate配置启用了它:
DefaultSimilarity
public class IgnoreScoringSimilarity extends DefaultSimilarity { @Override public float idf(long docFreq, long numDocs) { return 1.0f; } @Override public float tf(float freq) { return 1.0f; } @Override public float coord(int overlap, int maxOverlap) { return 1.0f; } @Override public float lengthNorm(FieldInvertState state) { return 1.0f; } @Override public float queryNorm(float sumOfSquaredWeights) { return 1.0f; } }
hibernate配置:
<property name="hibernate.search.default.similarity" value="com.search.IgnoreScoringSimilarity"/>
这种方法在90%的时间内都有效,但是,我仍然看到一些奇怪的结果,看起来似乎不合适。我认识到的模式是这些路由(文档)的大小非常大。一条普通路线大约有20-30个路线点,但这些错位结果却有100-150。这使我相信默认的Lucene评分仍在发生(由于文档大小,评分更高)。
在禁用Lucene的评分时,我做错什么了吗?可能还有其他解释吗?
我可以建议另一种基于自定义结果排序的方法。您可以在答案中阅读它。这个答案有些过时,因此我根据Lucene API 4.10.1修改了此示例。比较器
public abstract class CustomComparator extends FieldComparator<Double> { double[] scoring; double bottom; double topValue; private FieldCache.Ints[] currentReaderValues; private String[] fields; protected abstract double getScore(int[] value); public CustomComparator(int hitNum, String[] fields) { this.fields = fields; scoring = new double[hitNum]; } int[] fromReaders(int doc) { int[] result = new int[currentReaderValues.length]; for (int i = 0; i < result.length; i++) { result[i] = currentReaderValues[i].get(doc); } return result; } @Override public int compare(int slot1, int slot2) { return Double.compare(scoring[slot1], scoring[slot2]); } @Override public void setBottom(int slot) { this.bottom = scoring[slot]; } @Override public void setTopValue(Double top) { topValue = top; } @Override public int compareBottom(int doc) throws IOException { double v2 = getScore(fromReaders(doc)); return Double.compare(bottom, v2); } @Override public int compareTop(int doc) throws IOException { double docValue = getScore(fromReaders(doc)); return Double.compare(topValue, docValue); } @Override public void copy(int slot, int doc) throws IOException { scoring[slot] = getScore(fromReaders(doc)); } @Override public FieldComparator<Double> setNextReader(AtomicReaderContext atomicReaderContext) throws IOException { currentReaderValues = new FieldCache.Ints[fields.length]; for (int i = 0; i < fields.length; i++) { currentReaderValues[i] = FieldCache.DEFAULT.getInts(atomicReaderContext.reader(), fields[i], null, false); } return this; } @Override public Double value(int slot) { return scoring[slot]; } }
搜索示例
public class SortExample { public static void main(String[] args) throws IOException { final String[] fields = new String[]{"descriptionLength", "views.views", "nameLength"}; Sort sort = new Sort( new SortField( "", new FieldComparatorSource() { public FieldComparator newComparator(String fieldname, int numHits, int sortPos, boolean reversed) throws IOException { return new CustomComparator(numHits, fields) { @Override protected double getScore(int[] value) { int descriptionLength = value[0]; int views = value[1]; int nameLength = value[2]; return -((descriptionLength > 2000.0 ? 5.0 : 0.0) + (views > 5000.0 ? 3.0 : 0.0) + (nameLength > 20.0 ? 1.0 : 0.0)); } }; } } ) ); IndexWriterConfig indexWriterConfig = new IndexWriterConfig(Version.LUCENE_4_10_4, new StandardAnalyzer()); Directory directory = new RAMDirectory(); IndexWriter indexWriter = new IndexWriter(directory, indexWriterConfig); addDoc(indexWriter, "score 0", 1000, 1000, 10); addDoc(indexWriter, "score 5", 3000, 1000, 10); addDoc(indexWriter, "score 3", 1000, 6000, 10); addDoc(indexWriter, "score 1", 1000, 1000, 30); addDoc(indexWriter, "score 4", 1000, 6000, 30); addDoc(indexWriter, "score 6", 5000, 1000, 30); addDoc(indexWriter, "score 9", 5000, 6000, 30); final IndexReader indexReader = DirectoryReader.open(indexWriter, false); IndexSearcher indexSearcher = new IndexSearcher(indexReader); Query query = new TermQuery(new Term("all", "all")); int nDocs = 100; final TopDocs search = indexSearcher.search(query, null, nDocs, sort); System.out.println("Max " + search.scoreDocs.length + " " + search.getMaxScore()); for (ScoreDoc sd : search.scoreDocs) { Document document = indexReader.document(sd.doc); System.out.println(document.getField("name").stringValue()); } } private static void addDoc(IndexWriter indexWriter, String name, int descriptionLength, int views, int nameLength) throws IOException { Document doc = new Document(); doc.add(new TextField("name", name, Field.Store.YES)); doc.add(new TextField("all", "all", Field.Store.YES)); doc.add(new IntField("descriptionLength", descriptionLength, Field.Store.YES)); doc.add(new IntField("views.views", views, Field.Store.YES)); doc.add(new IntField("nameLength", nameLength, Field.Store.YES)); indexWriter.addDocument(doc); } }
代码将输出
score 9 score 6 score 5 score 4 score 3 score 1 score 0