autosuggest

Autosuggest

Generate autocomplete suggestions based on what your users search

:tangerine: Battle-tested at Instacart

Installation

Add this line to your application’s Gemfile:

gem "autosuggest"

Getting Started

Prepare your data

Start with a hash of queries and their popularity, like the number of users who have searched it.

top_queries = {
  "bananas" => 353,
  "apples"  => 213,
  "oranges" => 140
}

With Searchjoy, you can do:

top_queries = Searchjoy::Search.group(:normalized_query)
  .having("COUNT(DISTINCT user_id) >= 5").distinct.count(:user_id)

Then pass them to Autosuggest.

autosuggest = Autosuggest::Generator.new(top_queries)

Filter duplicates

Stemming is used to detect duplicates like apple and apples.

Specify the stemming language (defaults to english) with:

autosuggest = Autosuggest::Generator.new(top_queries, language: "spanish")

The most popular query is preferred by default. To override this, use:

autosuggest.prefer ["apples"]

To fix false positives, use:

autosuggest.not_duplicates [["straws", "straus"]]

Filter misspellings

We tried open-source libraries like Aspell and Hunspell but quickly realized we needed to build a corpus specific to our application.

There are two ways to build the corpus, which can be used together.

1. Add words

autosuggest.parse_words Product.pluck(:name)

Use the min option to only add words that appear multiple times.

1. Add concepts

autosuggest.add_concept "brand", Brand.pluck(:name)

Filter words

Profanity is blocked by default. Add custom words with:

autosuggest.block_words ["boom"]

Generate suggestions

Generate suggestions with:

suggestions = autosuggest.suggestions

Save suggestions

Save suggestions in your database or another data store.

With Rails, you can generate a simple model with:

rails generate autosuggest:suggestions
rails db:migrate

And update suggestions with:

now = Time.now
records = suggestions.map { |s| s.slice(:query, :score).merge(updated_at: now) }
Autosuggest::Suggestion.transaction do
  Autosuggest::Suggestion.upsert_all(records, unique_by: :query)
  Autosuggest::Suggestion.where("updated_at < ?", now).delete_all
end

Leave out unique_by for MySQL, and use activerecord-import for upserts with Rails < 6.

Show suggestions

Use a JavaScript autocomplete library like typeahead.js to show suggestions in the UI.

If you only have a few thousand suggestions, it’s much faster to load them all at once instead of as a user types (eliminates network requests).

With Rails, you can load all suggestions with:

Autosuggest::Suggestion.order(score: :desc).pluck(:query)

And suggestions matching user input with:

input = params[:query]
Autosuggest::Suggestion
  .order(score: :desc)
  .where("query LIKE ?", "%#{Autosuggest::Suggestion.sanitize_sql_like(input.downcase)}%")
  .pluck(:query)

You can also cache suggestions for performance.

Rails.cache.fetch("suggestions", expires_in: 5.minutes) do
  Autosuggest::Suggestion.order(score: :desc).pluck(:query)
end

Additional considerations

You may want to have someone manually approve suggestions:

Autosuggest::Suggestion.where(status: "approved")

Or filter suggestions without results:

Autosuggest::Suggestion.find_each do |suggestion|
  suggestion.results_count = Product.search(suggestion.query, load: false).count
  suggestion.save! if suggestion.changed?
end

Autosuggest::Suggestion.where("results_count > 0")

You can add additional fields to your model/data store to accomplish this.

Example

top_queries = Searchjoy::Search.group(:normalized_query)
  .having("COUNT(DISTINCT user_id) >= 5").distinct.count(:user_id)
product_names = Product.pluck(:name)
brand_names = Brand.pluck(:name)

autosuggest = Autosuggest::Generator.new(top_queries)
autosuggest.parse_words product_names
autosuggest.add_concept "brand", brand_names
autosuggest.prefer brand_names
autosuggest.not_duplicates [["straws", "straus"]]
autosuggest.block_words ["boom"]

suggestions = autosuggest.suggestions

now = Time.now
records = suggestions.map { |s| s.slice(:query, :score).merge(updated_at: now) }
Autosuggest::Suggestion.transaction do
  Autosuggest::Suggestion.upsert_all(records, unique_by: :query)
  Autosuggest::Suggestion.where("updated_at < ?", now).delete_all
end

Upgrading

0.2.0

Suggestions are now filtered by default, and only the query and score are returned. Get all queries and fields with:

autosuggest.suggestions(filter: false)

History

View the changelog

Contributing

Everyone is encouraged to help improve this project. Here are a few ways you can help:

• Report bugs
• Fix bugs and submit pull requests
• Write, clarify, or fix documentation
• Suggest or add new features

To get started with development:

git clone https://github.com/ankane/autosuggest.git
cd autosuggest
bundle install
bundle exec rake test