A Curious Case Of Incorrect Spelling
Maa ka laadla bigad gaya
Photo by Daniel Mingook Kim on Unsplash
Yesterday my mom was searching for a Gujarati movie on Amazon Prime Video in India.
She said, "I just saw it in search results and now I can't find it."
On probing, I realised she was searching with incorrect spelling.
This got me curious because fuzzy matching is like the basic feature of a search engine. It's not even hard to implement.
Fuzzy matching negatively affects latency and precision by some amount but has a positive effect on top-k recall. Imagine if Google failed at incorrect spelling. It will be mayhem and will lead to a loss of millions.
I tried some wrong spelling queries on prime videos and this is what I found
My thoughts
It seems prime does support fuzzy queries up to edit distance of 1, but the behaviour is not consistent amongst Hollywood and non-Hollywood content.
One of the main reason I use Prime videos is its superior regional collection compared to Netflix. I think they need to do everything to get a high recall on regional movie queries.
They can consider phonetic tokenization to match queries like 'bhut' to ‘bhoot’ as Indian users have a habit of writing in Hinglish.