Lucene: Multi-word phrases as search terms

Question

I'm trying to make a searchable phone/local business directory using Apache Lucene.

I have fields for street name, business name, phone number etc. The problem that I'm having is that when I try to search by street where the street name has multiple words (e.g. 'the crescent'), no results are returned. But if I try to search with just one word, e.g 'crescent', I get all the results that I want.

I'm indexing the data with the following:

String LocationOfDirectory = "C:\\dir\\index";

StandardAnalyzer analyzer = new StandardAnalyzer(Version.LUCENE_34);
Directory Index = new SimpleFSDirectory(LocationOfDirectory);

IndexWriterConfig config = new IndexWriterConfig(Version.LUCENE.34, analyzer);
IndexWriter w = new IndexWriter(index, config);


Document doc = new Document();
doc.add(new Field("Street", "the crescent", Field.Store.YES, Field.Index.Analyzed);

w.add(doc);
w.close();

My searches work like this:

int numberOfHits = 200;
String LocationOfDirectory = "C:\\dir\\index";
TopScoreDocCollector collector = TopScoreDocCollector.create(numberOfHits, true);
Directory directory = new SimpleFSDirectory(new File(LocationOfDirectory));
IndexSearcher searcher = new IndexSearcher(IndexReader.open(directory);

WildcardQuery q = new WildcardQuery(new Term("Street", "the crescent");

searcher.search(q, collector);
ScoreDoc[] hits = collector.topDocs().scoreDocs;

I have tried swapping the wildcard query for a phrase query, first with the entire string and then splitting the string up on white space and wrapping them in a BooleanQuery like this:

String term = "the crescent";
BooleanQuery b = new BooleanQuery();
PhraseQuery p = new PhraseQuery();
String[] tokens = term.split(" ");
for(int i = 0 ; i < tokens.length ; ++i)
{
    p.add(new Term("Street", tokens[i]));
}
b.add(p, BooleanClause.Occur.MUST);

However, this didn't work. I tried using a KeywordAnalyzer instead of a StandardAnalyzer, but then all other types of search stopped working as well. I have tried replacing spaces with other characters (+ and @), and converting queries to and from this form, but that still doesn't work. I think it doesn't work because + and @ are special characters which are not indexed, but I can't seem to find a list anywhere of which characters are like that.

I'm beginning to go slightly mad, does anyone know what I'm doing wrong?

Answer 1

The reason why you don't get your documents back is that while indexing you're using StandardAnalyzer, which converts tokens to lowercase and removes stop words. So the only term that gets indexed for your example is 'crescent'. However, wildcard queries are not analyzed, so 'the' is included as mandatory part of the query. The same goes for phrase queries in your scenario.

KeywordAnalyzer is probably not very suitable for your use case, because it takes whole field content as a single token. You can use SimpleAnalyzer for the street field -- it will split the input on all non-letter characters and then convert them to lowercase. You can also consider using WhitespaceAnalyzer with LowerCaseFilter. You need to try different options and work out what works best for your data and users.

Also, you can use different analyzers per field (e.g. with PerFieldAnalyzerWrapper) if changing analyzer for that field breaks other searches.

Answer 2

I found that my attempt to generate a query without using a QueryParser was not working, so I stopped trying to create my own queries and used a QueryParser instead. All of the recomendations that I saw online showed that you should use the same Analyzer in the QueryParser that you use during indexing, so I used a StandardAnalyzer to build the QueryParser.

This works on this example because the StandardAnalyzer removes the word "the" from the street "the crescent" during indexing, and hence we can't search for it because it isn't in the index.

However, if we choose to search for "Grove Road", we have a problem with the out-of-the-box functionality, namely that the query will return all of the results containing either "Grove" OR "Road". This is easily fixed by setting up the QueryParser so that it's default operation is AND instead of OR.

In the end, the correct solution was the following:

int numberOfHits = 200;
String LocationOfDirectory = "C:\\dir\\index";
TopScoreDocCollector collector = TopScoreDocCollector.create(numberOfHits, true);
Directory directory = new SimpleFSDirectory(new File(LocationOfDirectory));
IndexSearcher searcher = new IndexSearcher(IndexReader.open(directory);

StandardAnalyzer analyzer = new StandardAnalyzer(Version.LUCENE_35);

//WildcardQuery q = new WildcardQuery(new Term("Street", "the crescent");
QueryParser qp = new QueryParser(Version.LUCENE_35, "Street", analyzer);
qp.setDefaultOperator(QueryParser.Operator.AND);

Query q = qp.parse("grove road");

searcher.search(q, collector);
ScoreDoc[] hits = collector.topDocs().scoreDocs;