Tuesday, April 19, 2016

How to handle Stop Words in Hibernate Search 5.5.2 / Apache Lucene 5.4.x?

The Stop Words like ["a", "an", "and", "are", "as", "at", "be", "but", "by", "for", "if", "in", "into", "is", "it", "no", "not", "of", "on", "or", "such", "that", "the", "their", "then", "there", "these", "they", "this", "to", "was", "will", "with"] and the existence of them in terms or database or files that are to be indexed/searched by lucene can lead to any of the following:

1. Stop Words being Ignored/Filtered during the Lucene Indexing Process
2. Stop Words being Ignored/Filtered during the Lucene Querying Process
3. No Result for Queries that Include, Start With or End With any Stop Word

The way to solve this problem or to handle them during both indexing and searching process is as follows. The method explained here is specially suitable if you are using Hibernate Search 5.5.2 which in turn is using Apache Lucene 5.3.x/5.4.x

1. Define your Custom Analyzer, Adapted from the Standard Analyzer
You need to include only the two filters - 'LowerCaseFilterFactory' and 'StandardFilterFactory' as part of the Tokenizer definition. The filter factory that we have not included here is the 'StopFilter'. This allows Stop Words to be considered as other normal English Words and they are indexed.

@Table(name="table_name", catalog="catalog_name") 
@AnalyzerDef(name = "fedexTextAnalyzer",
   tokenizer = @TokenizerDef(factory = StandardTokenizerFactory.class),
   filters = {
     @TokenFilterDef(factory = LowerCaseFilterFactory.class),
     @TokenFilterDef(factory = StandardFilterFactory.class)

2. Mark the Field with Relevant Annotations (@Analyzer on @Field)
Along with the @Field Annotation on every Entity's or Table's Column Field, declare the Analyzer that we have defined above.

@Column(name="fedex_cs_product_name", nullable=false, length=100)
@Field(index=Index.YES,analyze=Analyze.YES, store=Store.NO, analyzer=@Analyzer(definition = "fedexTextAnalyzer"))
public String getFedexCsItemName() {
   return this.fedexCsItemName;

3. Use WhitespaceAnalyzer to Query so that Stop Words are 'Processed' by Default
Although the official documentation says that if we use 'StandardAnalyzer' by passing in the argument for Stop Words as CharArraySet.EMPTY_SET I found that the Query was still not able to retrreve any result. On Analysis with Luke, I found that for Queries such as 'Computer Science Books for Beginners', the 'for' was being ignored. Strange! I replaced it with WhitespaceAnalyzer, I found that it works for all 'Stop Words' and all 'Cases'.

I have found that the above is the best/minimal way to fix this issue. Also, our QA has verified that it works for all 'Stop Word' cases! Hope this helps you.

1 comment:

sanket said...

I tried that with CharArraySet.EMPTY_SET but never got a chance to test it. Thanks for sharing.