MongoDB officially has offered full text search capabilities since its 2.6 release. The feature ranks among the Top 5 most voted features for the server component and ships in its current version with numerous stemmers and parsers, phrase matching, negation and per field weights. So it's about time to give it a little love and share what's cooking in Spring Data's kitchen to support that feature.

For text indexing and searching MongoDB defaults the language to English, normalizing the text by tokenizing, removing common stop words and reducing words to their base. There's support for several other languages as well.

Manually build text indexes

@Document
class CookingRecipe {
  String title;
  String content;
}

Here we've got a very simple entity CookingRecipe and want to have a text index based on its title and content fields, putting some weight of 2 to search hits in title. Attaching weights to fields allows you to influence relevancy of a document when being looked up. It defines the significance of a field relative to others, boosting the documents score. In this case it doubles the documents relevance when hitting a match in title. The raw MongoDB index definition would look something like this:

{
  title : "text",
  content : "text"
},
{
  weights: { title : 2 }
}

As of version 1.5 M1 of Spring Data MongoDB we can create a text index, capturing the fields we want to have full text search enabled on, manually.

TextIndexDefinition textIndex = new TextIndexDefinitionBuilder()
  .onField("title", 2F)
  .onField("content")
  .build();

MongoTemplate template = … // obtain MongoTemplate
template.indexOps(CookingRecipe.class).ensureIndex(textIndex);

Alternatively, let the index be created automatically using mapping annotations. All we need to do is adding a few hints on the domain class and we are good to go.

@Document
class CookingRecipe {
  @TextIndexed(weight=2) String title;
  @TextIndexed String content;
}

Please note that it is only possible to have one full text index per collection. Now that we've created the index we'll query the top 5 recipes matching "coffee" or "cake".

TextCriteria criteria = TextCriteria.forDefaultLanguage()
  .matchingAny("coffee", "cake");

Query query = TextQuery.queryText(criteria)
  .sortByScore()
  .with(new PageRequest(0, 5));

List<CookingRecipe> recipes = template.find(query, CookingRecipe);

Note that we provide dedicated types TextCriteria and TextQuery to express searches in full detail.

Scoring

As mentioned earlier, documents get scored while searching. The score value is not returned by default, but since this information is often helpful we can include it in the output by adding { score : { $meta : "textScore" } } to the projection, which is implicitly done by calling query.sortByScore(). To access the score in the resulting documents we add a @TextScore annotated property to CookingRecipe.

@Document
class CookingRecipe {
  @TextIndexed(weight=2) String title;
  @TextIndexed String content;
  @TextScore Float score;
}

The @TextScore annotation implicitly turns the score property into a read-only property due to the @ReadOnlyProperty annotation the annotation carries in turn. The latter can be used in other contexts as well where you'd like make sure fields from a document are only read but never written.

Additional resources on behavior and restrictions, as well as supported languages can be found in the MongoDB reference.

Moving forward with release train Evans we will be adding text search support to MongoRepositories.

If you want to learn more about Spring Data, be sure to register for this year's SpringOne conference. The schedule contains a lot of data-related talks to introduce you to the latest features we're going to ship with Evans.