Search is incredibly hard to do well. It can take a weeks of any developer’s time to set up and tune. After going through this at Instacart, we decided to open source our efforts to help others bootstrap their search.
Here’s what you get with Searchkick.
1. Works Out-of-the-Box
Searchkick handles:
- stemming -
tomatoesmatchestomato - special characters -
jalapenomatchesjalapeño - extra whitespace -
dishwashermatchesdish washer - misspellings -
zuchinimatcheszucchini - custom synonyms -
qtipmatchescotton swab
Searchkick packages some standard Elasticsearch token filters to accomplish this, including snowball, shingle, and synonym.
2. Self-Improving Searches
Searchkick uses conversion data to learn what your users are looking for. If a user searches for “ice cream” and adds Ben & Jerry’s Chunky Monkey to the cart (the conversion metric), that item gets a little more weight for similar searches. Results get better as more people search.
Here’s how the converted search result position (mean) changed over the past 12 weeks.
3. Zero Downtime Re-Indexing
Searchkick allows you to re-index your documents without any downtime. Queries are run against the products_production index, which an is alias. When you reindex, Searchkick creates a new index suffixed with the current time - products_production_20130730120000. Once the index is built, it’s aliased to products_production in one atomic operation.
4. Query Like SQL
Query like SQL so there’s no need to learn a new query language.
Product.search "2% Milk", where: {in_stock: true}, limit: 10What’s next?
We’d like Searchkick to work with any language, since much of the project is Elasticsearch configuration.
Also, we’d like to prepopulate synonyms from WordNet so scallion matches green onion out of the box.
Give your users an incredible search experience.
Acknowledgements
- Jaroslav Kalistsuk for zero downtime reindexing