“We can’t show you 6,000 places in Paris—you won’t have a successful booking experience,” Elena Grewal, data science manager with Airbnb, told data scientists and business leaders attending the Machine Learning Conference in San Francisco on November 11, 2016. “And so throughout our product we’re always thinking about how can we personalize, how can we make this better, how can we use data to improve the user experience? We use machine learning at all steps in the booking process.”
In addition to illuminating how Airbnb designs machine-learning–based products, the event provided a peek into how Netflix optimizes its video assets and how Amazon refines its search rankings. It also offered other lessons from more than a dozen presenters about applying cutting-edge machine-learning techniques to achieve business goals. (Videos or slides of the presentations are available from MLconf 2016). Below are some of the attendees’ tweeted highlights.
A New Way of Thinking
Anna Novakovska ML helps us to understand what image would make us watch the movie!
Lauren Different color meanings in different cultures. Important for choosing how to represent shows in Netflix.
Joachim How Amazon recommends books. @MLconf
Lauren Metrics are important! Which Breaking Bad image had the best CTR [click through ratio] on Netflix? Probably not the one you think. (Answer: the bus)
Define the Goal, Measure Results
Ashwin Baskaran Summarizing @Airbnb Elena Grewal’s intro: it’s the question, stupid! :-)
Rupinder Singh Most important step in any ML problem… well chosen Target. Interesting talk by Elena Grewal from AirBnB
Dara Strauss-Albee Elena Grewal, Airbnb: regression model for price —> classification model for bookings w/price variable. Problem definition is key
Lauren Which metric to use for Amazon search rankings -CTR or purchases? Users who searched for “iPhone” most often bought the cable.
Joachim Don’t predict user interests in 2016 based on their behaviours in 2008.
Dara Strauss-Albee Anjuli Kannan @GoogleBrain – RNN [Recurrent Neural Networks] for Gmail auto reply now generate 10% of mobile responses
Details Matter
Lauren #machinelearning workflow Step 1. Data Step 2. ??? Step 3. ML algorithms Step 4. ??? Step 5. Profit!
Adarsh Heard @MLconf “When you know what you want to learn it’s hard, when you don’t it’s harder.” @JFPuget D!