Package org.apache.lucene.collation

Unicode collation support.

See: Description

Class Summary

Class	Description
CollationAttributeFactory	Converts each token into its CollationKey, and then encodes the bytes as an index term.
CollationDocValuesField	Indexes collation keys as a single-valued SortedDocValuesField.
CollationKeyAnalyzer	Configures KeywordTokenizer with CollationAttributeFactory.

Package org.apache.lucene.collation Description

Unicode collation support.

Collation converts each token into its binary CollationKey using the provided Collator, allowing it to be stored as an index term.

Use Cases

• Efficient sorting of terms in languages that use non-Unicode character orderings. (Lucene Sort using a Locale can be very slow.)
• Efficient range queries over fields that contain terms in languages that use non-Unicode character orderings. (Range queries using a Locale can be very slow.)
• Effective Locale-specific normalization (case differences, diacritics, etc.). (LowerCaseFilter and ASCIIFoldingFilter provide these services in a generic way that doesn't take into account locale-specific needs.)

Example Usages

Farsi Range Queries

   // "fa" Locale is not supported by Sun JDK 1.4 or 1.5
   Collator collator = Collator.getInstance(new Locale("ar"));
   CollationKeyAnalyzer analyzer = new CollationKeyAnalyzer(collator);
   RAMDirectory ramDir = new RAMDirectory();
   IndexWriter writer = new IndexWriter(ramDir, new IndexWriterConfig(analyzer));
   Document doc = new Document();
   doc.add(new TextField("content", "ساب", Field.Store.YES));
   writer.addDocument(doc);
   writer.close();
   IndexReader ir = DirectoryReader.open(ramDir);
   IndexSearcher is = new IndexSearcher(ir);
 
   QueryParser aqp = new QueryParser("content", analyzer);
   aqp.setAnalyzeRangeTerms(true);
     
   // Unicode order would include U+0633 in [ U+062F - U+0698 ], but Farsi
   // orders the U+0698 character before the U+0633 character, so the single
   // indexed Term above should NOT be returned by a ConstantScoreRangeQuery
   // with a Farsi Collator (or an Arabic one for the case when Farsi is not
   // supported).
   ScoreDoc[] result
     = is.search(aqp.parse("[ د TO ژ ]"), null, 1000).scoreDocs;
   assertEquals("The index Term should not be included.", 0, result.length);
 

Danish Sorting

   Analyzer analyzer 
     = new CollationKeyAnalyzer(Collator.getInstance(new Locale("da", "dk")));
   RAMDirectory indexStore = new RAMDirectory();
   IndexWriter writer = new IndexWriter(indexStore, new IndexWriterConfig(analyzer));
   String[] tracer = new String[] { "A", "B", "C", "D", "E" };
   String[] data = new String[] { "HAT", "HUT", "HÅT", "HØT", "HOT" };
   String[] sortedTracerOrder = new String[] { "A", "E", "B", "D", "C" };
   for (int i = 0 ; i < data.length ; ++i) {
     Document doc = new Document();
     doc.add(new StoredField("tracer", tracer[i]));
     doc.add(new TextField("contents", data[i], Field.Store.NO));
     writer.addDocument(doc);
   }
   writer.close();
   IndexReader ir = DirectoryReader.open(indexStore);
   IndexSearcher searcher = new IndexSearcher(ir);
   Sort sort = new Sort();
   sort.setSort(new SortField("contents", SortField.STRING));
   Query query = new MatchAllDocsQuery();
   ScoreDoc[] result = searcher.search(query, null, 1000, sort).scoreDocs;
   for (int i = 0 ; i < result.length ; ++i) {
     Document doc = searcher.doc(result[i].doc);
     assertEquals(sortedTracerOrder[i], doc.getValues("tracer")[0]);
   }
 

Turkish Case Normalization

   Collator collator = Collator.getInstance(new Locale("tr", "TR"));
   collator.setStrength(Collator.PRIMARY);
   Analyzer analyzer = new CollationKeyAnalyzer(collator);
   RAMDirectory ramDir = new RAMDirectory();
   IndexWriter writer = new IndexWriter(ramDir, new IndexWriterConfig(analyzer));
   Document doc = new Document();
   doc.add(new TextField("contents", "DIGY", Field.Store.NO));
   writer.addDocument(doc);
   writer.close();
   IndexReader ir = DirectoryReader.open(ramDir);
   IndexSearcher is = new IndexSearcher(ir);
   QueryParser parser = new QueryParser("contents", analyzer);
   Query query = parser.parse("dıgy");   // U+0131: dotless i
   ScoreDoc[] result = is.search(query, null, 1000).scoreDocs;
   assertEquals("The index Term should be included.", 1, result.length);
 

Caveats and Comparisons

WARNING: Make sure you use exactly the same Collator at index and query time -- CollationKeys are only comparable when produced by the same Collator. Since RuleBasedCollators are not independently versioned, it is unsafe to search against stored CollationKeys unless the following are exactly the same (best practice is to store this information with the index and check that they remain the same at query time):

1. JVM vendor
2. JVM version, including patch version
3. The language (and country and variant, if specified) of the Locale used when constructing the collator via Collator.getInstance(java.util.Locale).
4. The collation strength used - see Collator.setStrength(int)

ICUCollationKeyAnalyzer, available in the icu analysis module, uses ICU4J's Collator, which makes its version available, thus allowing collation to be versioned independently from the JVM. ICUCollationKeyAnalyzer is also significantly faster and generates significantly shorter keys than CollationKeyAnalyzer. See http://site.icu-project.org/charts/collation-icu4j-sun for key generation timing and key length comparisons between ICU4J and java.text.Collator over several languages.

CollationKeys generated by java.text.Collators are not compatible with those those generated by ICU Collators. Specifically, if you use CollationKeyAnalyzer to generate index terms, do not use ICUCollationKeyAnalyzer on the query side, or vice versa.