Welcome to 2020! For our first event of this year, we've got something special - Thomas Wolf! Just because we're all about underpromising and over-delivering, we've also got a rather interesting talk by Ash Fernando from Canva lined up.
We're really excited about having such a great line-up and hope you'll all come along! -- Team Sydney NLP
Thomas is the Chief Science Officer of Huggingface, which we're sure you're all aware of by now. Huggingface - and Thomas - are on a mission to build the definitive NLP library and have recently taken another step in this direction with the release of their python tokenizers package. Prior to Huggingface, Thomas gained a Ph.D. in quantum physics and a legal practitioner's license to work on patenting Deep Learning and Machine Learning IP.
Ashwanth is a backend engineering lead at Canva. Before joining the Presentations team, he was instrumental in building spell correction with a team of 3 people for Canva’s public facing search. In a past life he worked in the SF Bay Area for close to a decade at iconic companies such as Apple, eBay and WalmartLabs (The research arm of Walmart). His current work involves building the next generation of presentation products for Canva. His primary interests lie in software engineering, but he has a side interest in data systems and ML.
Search Query Spell Corrections at Canva:
There is no silver bullet for spell correction in public facing search. Solutions involve rudimentary systems like a simple hard coded table that corrects the top X spelling mistakes all the way to sophisticated language models that can detect from spelling to grammatical mistakes.
How do you right size an appropriate spell correction system that will not bet the farm but still provide a reasonable ROI to the business is always a top consideration for many websites that have public facing search.
This talk will go through some of the considerations of building a data-driven, continuously improving, spell correction system for Canva’s public facing search.