Mongoid Fulltext Search

Full-text search using n-gram matching for the Mongoid ODM. Tested on MongoDB 1.6 and above, but probably works on earlier versions as well.

MongoDB currently has no native full-text search capabilities, so this gem is a good fit for cases where you want something a little less than a full-blown indexing service like Solr. mongoid_fulltext lets you do a fuzzy string search across relatively short strings, which makes it good for populating autocomplete boxes based on the display names of your Rails models but not appropriate for, say, indexing hundreds of thousands of HTML documents.

Install

gem 'mongoid_fulltext'

Some examples:

Suppose you have an Artist model and want to index each artist's name:

class Artist
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :first_name
  field :last_name

  def name
    [first_name, last_name].join(' ')
  end

  fulltext_search_in :name
end

The fulltext_search_in directive will index the full name of the artist, so now you can call:

Artist.fulltext_search("vince vangogh")

which will return an array of the Artist instances that best match the search string. Most likely, Vincent van Gogh will be included in the results. You can index multiple fields with the same index, so we can get the same effect of our Artist index above using:

class Artist
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :first_name
  field :last_name

  fulltext_search_in :first_name, :last_name
end

To restrict the number of results returned, pass the :max_results parameter to fulltext_search:

Artist.fulltext_search("vince vangogh", { :max_results => 5 })

To return a pair of [ result, score ] instead of an array of results, pass the :return_scores parameter to fulltext_search:

Artist.fulltext_search("vince vangogh", { :return_scores => true })

The larger a score is, the better mongoid_fulltext thinks the match is. The scores have the following rough interpretation that you can use to make decisions about whether the match is good enough:

If a prefix of your query matches something indexed, or if your query matches a prefix of something indexed (for example, searching for "foo" finds "myfoo" or searching for "myfoo" finds "foo"), you can expect a score of at least 1 for the match.
If an entire word in your query matches an entire word that's indexed and you have the index_full_words option turned on (it's turned on by default), you can expect a score of at least 2 for the match.
If neither of the above criteria are met, you can expect a score less than one.

If you don't specify a field to index, the default is the result of to_s called on the object. The following definition will index the first and last name of an artist:

class Artist
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :first_name
  field :last_name

  def to_s
    '%s %s' % [first_name, last_name]
  end

  fulltext_search_in
end

The full-text index is stored in a separate MongoDB collection in the same database as the models you're indexing. By default, the name of this collection is generated for you. Above, a collection named something like mongoid_fulltext.index_artist_0 will be created to hold the index data. You can override this naming and provide your own collection name with the :index_name parameter:

class Artwork
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :title
  fulltext_search_in :title, :index_name => 'mongoid_fulltext.foobar'
end

You can also create multiple indexes on a single model, in which case you'll want to provide index names:

class Artwork
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :title
  field :artist_name
  field :gallery_name
  filed :gallery_address

  fulltext_search_in :title, :index_name => 'title_index'
  fulltext_search_in :artist_name, :index_name => 'artist_name_index'
  fulltext_search_in :gallery_name, :gallery_address, :index_name => 'gallery_index'
end

The index names are helpful now because you'll have to specify which one you want to use when you call fulltext_search:

Artwork.fulltext_search('warhol', :index => 'artist_name_index')

If you have multiple indexes specified and you don't supply a name to fulltext_search, the method call will raise an exception.

If you're indexing multiple models, you may find that you need to combine results to create a single result set. For example, if both the Artist model and the Artwork model are indexed for full-text search, then to get results from both, you'd have to call Artist.fulltext_search and Artwork.fulltext_search and combine the results yourself. If your intention is instead to get the top k results from both Artists and Artworks, you can merge both into a single index by using the same :external_index parameter:

class Artwork
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :title
  fulltext_search_in :title, :index_name => 'artwork_and_artists'
end

class Artist
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :name
  fulltext_search_in :name, :index_name => 'artwork_and_artists'
end

Now that these two models share the same external index collection, we can search them both through either model's fulltext_search method:

Artwork.fulltext_search('picasso')  # returns same results as Artist.fulltext_search('picasso')

If you want to filter the results from full-text search, you set up filters when the indexes are defined. For example, suppose that in addition to wanting to use the artwork_and_artists index defined above to search for Artworks or Artists, we want to be able to run full-text searches for artists only and for artworks priced above $10,000. Instead of creating two new indexes or attempting to filter the results after the query is run, we can specify the filter predicates at the time of index definition:

class Artwork
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :title
  field :price
  fulltext_search_in :title, :index_name => 'artwork_and_artists',
                     :filters => { :is_expensive => lambda { |x| x.price > 10000 },
                                   :has_long_name => lambda { |x| x.title.length > 20 }}
end

class Artist
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :name
  field :birth_year
  fulltext_search_in :name, :index_name => 'artwork_and_artists',
                     :filters => { :born_before_1900 => lambda { |x| x.birth_year < 1900 },
                                   :has_long_name => lambda { |x| x.name.length > 20}}
end

After defining filters, you can query for results that match particular values of filters:

# Only return artists born before 1900 that match 'foobar'
Artist.fulltext_search('foobar', :born_before_1900 => true)

# Return artists or artworks that match 'foobar' and have short names
Artist.fulltext_search('foobar', :has_long_name => false)

# Only return artworks with prices over 10000 that match 'mona lisa'
Artwork.fulltext_search('mona lisa', :is_expensive => true)

# Only return artworks with prices less than 10000 that match 'mona lisa'
Artwork.fulltext_search('mona lisa', :is_expensive => false)

Note that in all of the example queries above, supplying a filter that is defined on exactly one of the models will restrict the search to results from that model only. For example, since :is_expensive is defined only on Artworks, a call to fulltext_search with either :is_expensive => true or :is_expensive => false will return only Artwork results.

You can specify multiple filters per index and per model. Each filter is a predicate that will be called on objects as they're inserted into the full-text index (any time the model is saved.) Filters are only called on instances of models they're defined on, so in the example above, the is_expensive filter is only applied to instances of Artwork and the born_before_1900 filter is only applied to instances of Artist, although both filters can be used when querying from either model. The has_long_name filter, on the other hand, will return instances of both Artwork and Artist since it's defined on each model.

Filters shouldn't ever throw, but if they do, the filter is just ignored. If you apply filters to indexes that are on multiple fields, the filter is applied to each field and the filter result is the AND of all of the individual results for each of the fields. Finally, if a filter is defined but criteria for that filter aren't passed to fulltext_search, the result is as if the filter had never been defined - you see both models that both pass and fail the filter in the results.

Indexing Options

Additional indexing/query options can be used as parameters to fulltext_search_in.

alphabet: letters to index, default is abcdefghijklmnopqrstuvwxyz0123456789 .
word_separators: word separators, default is the space character.
ngram_width: ngram width, default is 3.
index_full_words: index full words, which improves exact matches, default is true.
index_short_prefixes: index a prefix of each full word of length (ngram_width-1). Useful if you use a larger ngram_width than the default of 3. Default for this option is false.
stop_words: a hash of words to avoid indexing as full words. Used only if index_full_words is set to true. Defaults to a hash containing a list of common English stop words.
apply_prefix_scoring_to_all_words: score n-grams at beginning of words higher, default is true.
max_ngrams_to_search: maximum number of ngrams to query at any given time, default is 6.
max_candidate_set_size: maximum number of candidate ngrams to examine for a given query. Defaults to 1000. If you're seeing poor results, you can try increasing this value to consider more ngrams per query (changing this parameter does not require a re-index.) The amount of time a search takes is directly proportional to this parameter's value.
remove_accents: remove accents on accented characters, default is true. We strip the accents using NFKD normalization using an external library, unicode_utils.
update_if: controls whether or not the index will be updated. This can be set to a symbol, string, or proc. If the result of evaluating the value is true, the index will be updated.
- When set to a symbol, the symbol is sent to the document.
- When set to a string, the string is evaluated within the document's instance.
- When set to a proc, the proc is called, and the document is given to the proc as the first arg.
- When set to any other type of object, the document's index will not be updated.
reindex_immediately: whether models will be reindexed automatically upon saves or updates. Defaults to true. When set to false, the class-level update_ngram_index method can be called to perform reindexing.

Array filters

A filter may also return an Array. Consider the following example.

class Artist
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :name
  field :exhibitions, as: Array, default: []

  fulltext_search_in :name, :index_name => 'exhibited_artist',
    :filters => { 
      :exhibitions => lambda { |artist| artist.exhibitions }
    }
end

You can now find all artists that are at the Art Basel exhibition or all artists that have exhibited at both the Art Basel and the New York Armory exhibition.

# All artists
Artist.fulltext_search('foobar')

# Artists at the Art Basel exhibition only
Artist.fulltext_search('foobar', :exhibitions => [ "Art Basel" ])

# Artists at both the Art Basel and the New York Armory exhibition
Artist.fulltext_search('foobar', :exhibitions => [ "Art Basel", "New York Armory" ])

# Note that the following explicit syntax may be used to achieve the
# same result as above
Artist.fulltext_search('foobar', :exhibitions => {:all => [ "Art Basel", "New York Armory" ]})

If you want to find all artists that are at either the Art Basel or the New York Armory exhibition, then you may specify the :any operator in the filter.

# Artists at either the Art Basel or the New York Armory exhibition
Artist.fulltext_search('foobar', :exhibitions => {:any => [ "Art Basel", "New York Armory" ]})

Note that :all and :any are currently the only supported operators for the array filters.

Building the index

The fulltext index is built and maintained incrementally by hooking into before_save and before_destroy callbacks on each model that's being indexed. If you want to build an index on existing models, you can call the update_ngram_index method on the class or each instance:

Artwork.update_ngram_index
Artwork.find(id).update_ngram_index

You can remove all or individual instances from the index with the remove_from_ngram_index method:

Artwork.remove_from_ngram_index
Artwork.find(id).remove_from_ngram_index

The methods on the model level perform bulk removal operations and are therefore faster that updating or removing records individually.

If you need to control when the index is updated, provide the update_if option to fulltext_search_in, and set it to a symbol, string, or proc. Eg:

# Only update the "age" index if the "age" field has changed.
fulltext_search_in :age,    :update_if => :age_changed?

# Only update the "names" index if the "firstname" or "lastname" field has changed.
fulltext_search_in :names,  :update_if => "firstname_changed? || lastname_changed?"

# Only update the "email" index if the "email" field ends with "gmail.com".
fulltext_search_in :email,  :update_if => Proc.new { |doc| doc.email.match /gmail.com\Z/ }

Mongo Database Indexes

Mongoid provides an indexing mechanism on its models triggered by the create_indexes method. Mongoid_fulltext will hook into that behavior and create appropriate database indexes on its collections. These indexes are required for an efficient full text search.

Creating database indexes is typically done with the db:mongoid:create_indexes task.

rake db:mongoid:create_indexes

Running the specs

To run the specs, execute rake spec. You need a local MongoDB instance to run the specs.

Contributing

Fork the project. Make your feature addition or bug fix with tests. Send a pull request. Bonus points for topic branches.

Copyright and License

MIT License, see LICENSE for details.

Xenofex/mongoid_fulltext