StarZero All shows Try the podcast clipper →

Dec 5, 2013 · 20m · mad

Blake Shaw, Foursquare // Data Driven NYC 20 // Nov 2013

Blake Shaw · 15m spoken Matt Turck · 1m spoken Sydney Beveridge · 29s spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

At a Data Driven NYC event, Foursquare data scientist Blake Shaw explains how the company transforms billions of check-ins and passive location signals into predictive machine learning models and context-aware venue recommendations.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 6.8% of the talking time here. How this is scored →

Matt as informed peer 1.0 Guest teaching 2.0 Guest disagreement 0.2 Matt pushing back 0.5
05100:0010:0020:000:23–3:54 · Matt as informed peer 1/10 Foursquare Platform Overview and User Capabilities Blake introduces Foursquare's platform scale and maps check-in density across major cities. Host Matt Turck briefly chimes in to help name the Istanbul Bridge during a pause, but otherwise lets Blake present uninterrupted.3:54–5:57 · Matt as informed peer 0/10 Data Exhaust and Contextual Awareness Blake introduces contextual awareness products and announces Foursquare's newly shipped contextual notifications feature. This is a pure monologue presentation segment with zero host interaction.5:57–9:15 · Matt as informed peer 0/10 Location Accuracy and Spatiotemporal Place Matching Blake details technical spatial matching challenges, explaining GPS noise and spatiotemporal models using venue check-in distributions. Host scores remain zero as this is a technical presentation monologue.9:15–12:23 · Matt as informed peer 0/10 Predictive Machine Learning via History and Social Data Blake outlines combining social history and temporal signals into a Random Forest predictor, along with NLP sentiment extraction on tips. The segment is an uninterrupted expert monologue.12:23–14:31 · Matt as informed peer 0/10 Real-World Behavioral Impact and Presentation Conclusion Blake explains measuring real-world walk-through conversions rather than web click-throughs to conclude his presentation. Host host-side scores remain zero for this presentation conclusion.14:31–20:26 · Matt as informed peer 5/10 Fireside Q&A Session with Matt Turck and Audience Matt Turck leads Q&A, probing location privacy terms and bringing up industry competitors like Factual to question if Foursquare's bootstrapping model remains unique. Blake responds collaboratively while explaining Foursquare's live scale data moat.0:23–3:54 · Guest teaching 1/10 Foursquare Platform Overview and User Capabilities Blake introduces Foursquare's platform scale and maps check-in density across major cities. Host Matt Turck briefly chimes in to help name the Istanbul Bridge during a pause, but otherwise lets Blake present uninterrupted.3:54–5:57 · Guest teaching 2/10 Data Exhaust and Contextual Awareness Blake introduces contextual awareness products and announces Foursquare's newly shipped contextual notifications feature. This is a pure monologue presentation segment with zero host interaction.5:57–9:15 · Guest teaching 2/10 Location Accuracy and Spatiotemporal Place Matching Blake details technical spatial matching challenges, explaining GPS noise and spatiotemporal models using venue check-in distributions. Host scores remain zero as this is a technical presentation monologue.9:15–12:23 · Guest teaching 2/10 Predictive Machine Learning via History and Social Data Blake outlines combining social history and temporal signals into a Random Forest predictor, along with NLP sentiment extraction on tips. The segment is an uninterrupted expert monologue.12:23–14:31 · Guest teaching 2/10 Real-World Behavioral Impact and Presentation Conclusion Blake explains measuring real-world walk-through conversions rather than web click-throughs to conclude his presentation. Host host-side scores remain zero for this presentation conclusion.14:31–20:26 · Guest teaching 3/10 Fireside Q&A Session with Matt Turck and Audience Matt Turck leads Q&A, probing location privacy terms and bringing up industry competitors like Factual to question if Foursquare's bootstrapping model remains unique. Blake responds collaboratively while explaining Foursquare's live scale data moat.0:23–3:54 · Guest disagreement 0/10 Foursquare Platform Overview and User Capabilities Blake introduces Foursquare's platform scale and maps check-in density across major cities. Host Matt Turck briefly chimes in to help name the Istanbul Bridge during a pause, but otherwise lets Blake present uninterrupted.3:54–5:57 · Guest disagreement 0/10 Data Exhaust and Contextual Awareness Blake introduces contextual awareness products and announces Foursquare's newly shipped contextual notifications feature. This is a pure monologue presentation segment with zero host interaction.5:57–9:15 · Guest disagreement 0/10 Location Accuracy and Spatiotemporal Place Matching Blake details technical spatial matching challenges, explaining GPS noise and spatiotemporal models using venue check-in distributions. Host scores remain zero as this is a technical presentation monologue.9:15–12:23 · Guest disagreement 0/10 Predictive Machine Learning via History and Social Data Blake outlines combining social history and temporal signals into a Random Forest predictor, along with NLP sentiment extraction on tips. The segment is an uninterrupted expert monologue.12:23–14:31 · Guest disagreement 0/10 Real-World Behavioral Impact and Presentation Conclusion Blake explains measuring real-world walk-through conversions rather than web click-throughs to conclude his presentation. Host host-side scores remain zero for this presentation conclusion.14:31–20:26 · Guest disagreement 1/10 Fireside Q&A Session with Matt Turck and Audience Matt Turck leads Q&A, probing location privacy terms and bringing up industry competitors like Factual to question if Foursquare's bootstrapping model remains unique. Blake responds collaboratively while explaining Foursquare's live scale data moat.0:23–3:54 · Matt pushing back 0/10 Foursquare Platform Overview and User Capabilities Blake introduces Foursquare's platform scale and maps check-in density across major cities. Host Matt Turck briefly chimes in to help name the Istanbul Bridge during a pause, but otherwise lets Blake present uninterrupted.3:54–5:57 · Matt pushing back 0/10 Data Exhaust and Contextual Awareness Blake introduces contextual awareness products and announces Foursquare's newly shipped contextual notifications feature. This is a pure monologue presentation segment with zero host interaction.5:57–9:15 · Matt pushing back 0/10 Location Accuracy and Spatiotemporal Place Matching Blake details technical spatial matching challenges, explaining GPS noise and spatiotemporal models using venue check-in distributions. Host scores remain zero as this is a technical presentation monologue.9:15–12:23 · Matt pushing back 0/10 Predictive Machine Learning via History and Social Data Blake outlines combining social history and temporal signals into a Random Forest predictor, along with NLP sentiment extraction on tips. The segment is an uninterrupted expert monologue.12:23–14:31 · Matt pushing back 0/10 Real-World Behavioral Impact and Presentation Conclusion Blake explains measuring real-world walk-through conversions rather than web click-throughs to conclude his presentation. Host host-side scores remain zero for this presentation conclusion.14:31–20:26 · Matt pushing back 3/10 Fireside Q&A Session with Matt Turck and Audience Matt Turck leads Q&A, probing location privacy terms and bringing up industry competitors like Factual to question if Foursquare's bootstrapping model remains unique. Blake responds collaboratively while explaining Foursquare's live scale data moat.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 1.6% · guest 98.4%0:00 · Matt 1.6% · guest 98.4%3:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%12:00 · Matt 14.7% · guest 85.3%12:00 · Matt 14.7% · guest 85.3%15:00 · Matt 13.6% · guest 86.4%15:00 · Matt 13.6% · guest 86.4%18:00 · Matt 21.4% · guest 78.6%18:00 · Matt 21.4% · guest 78.6%
Sharpest disagreement ▶ 17:25 Blake reframes generic data competitors

Blake rejects the idea that static directories or competitors can match Foursquare, arguing generic solutions lack the live scale behavioral data sauce.

 Hardest push from Matt ▶ 14:34 Matt presses on continuous location tracking privacy

Matt challenges Blake on whether using Foursquare implies consent to continuous background location tracking, forcing Blake to address privacy policies directly.

 Biggest teaching moment ▶ 16:24 Blake explains bootstrapping via gamification

Blake educates the audience on how Foursquare solved its initial data cold-start problem by using badges and mayorship games before pivoting into a big data engine.

 Matt holds his own ▶ 17:09 Matt cites competitor Factual and questions bootstrapping

Matt demonstrates strong industry market knowledge by bringing up competitors like Factual and asking if modern location apps still need a gamification bootstrap phase.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Foursquare Platform Overview and User Capabilities 1100 Blake introduces Foursquare's platform scale and maps check-in density across major cities. Host Matt Turck briefly chimes in to help name the Istanbul Bridge during a pause, but otherwise lets Blake present uninterrupted.
Data Exhaust and Contextual Awareness 0200 Blake introduces contextual awareness products and announces Foursquare's newly shipped contextual notifications feature. This is a pure monologue presentation segment with zero host interaction.
Location Accuracy and Spatiotemporal Place Matching 0200 Blake details technical spatial matching challenges, explaining GPS noise and spatiotemporal models using venue check-in distributions. Host scores remain zero as this is a technical presentation monologue.
Predictive Machine Learning via History and Social Data 0200 Blake outlines combining social history and temporal signals into a Random Forest predictor, along with NLP sentiment extraction on tips. The segment is an uninterrupted expert monologue.
Real-World Behavioral Impact and Presentation Conclusion 0200 Blake explains measuring real-world walk-through conversions rather than web click-throughs to conclude his presentation. Host host-side scores remain zero for this presentation conclusion.
Fireside Q&A Session with Matt Turck and Audience 5313 Matt Turck leads Q&A, probing location privacy terms and bringing up industry competitors like Factual to question if Foursquare's bootstrapping model remains unique. Blake responds collaboratively while explaining Foursquare's live scale data moat.

Statements from this episode (10)

Assertion Not checkable as stated
Foursquare reports 40 million registered users and 4.5 billion total check-ins
“We're, we have over forty million registered users. We're aware of fifty-five million places all over the world. We've collected 4.5 billion check-ins. And we get about six million check-ins per day.”
Blake Shaw Dec 5, 2013 ▶ 1:12
Disclosure
Foursquare launches automated location recommendations that require no manual check-ins
“The idea, right, is you walk into a venue, and without having to do anything, you get a notification like this on your phone.”
Blake Shaw Dec 5, 2013 ▶ 5:00
Disclosure
Foursquare triggers automated notifications only when users enter unfamiliar territory
“You know, walk into a neighborhood or a place, and we make sure that you don't miss the most important information about the place that you're at, but only if you're in unfamiliar territory.”
Blake Shaw Dec 5, 2013 ▶ 5:43
Assertion Not publicly verifiable
Foursquare claims its background contextual notifications use under 1% battery hourly
“Less than one percent per hour is the, is what it does.”
Blake Shaw Dec 5, 2013 ▶ 6:44
Assertion Not checkable as stated
Foursquare data reveals 73% of venue visits are to previously visited places
“We've seen that 73% of the time a person is going to a place that they've already been to, or 62% of the time they're going to a place that someone in their social network has been to before.”
Blake Shaw Dec 5, 2013 ▶ 9:19
Insight
Effective location machine learning requires mapping raw coordinates to contextual venues
“You can't do machine learning with lat-longs.”
Blake Shaw Dec 5, 2013 ▶ 10:14
Disclosure
Foursquare built NLP models to extract venue recommendations and sentiment from tips
“We've built, you know, sort of NLP models that can read You know, text and extract these key recommendations and figure out their sentiment.”
Blake Shaw Dec 5, 2013 ▶ 11:48
Assertion Not checkable as stated
18% of users shown a Foursquare Explore recommendation check in within days
“With our recommendation product, Explore, 18% of people who see a result in Explore check in there within the first three days.”
Blake Shaw Dec 5, 2013 ▶ 13:21
Disclosure
Foursquare commits to not sharing raw user location data with third parties
“We're not going to share your data with anybody without, you know we're just not going to share your data.”
Blake Shaw Dec 5, 2013 ▶ 15:08
Assertion Not checkable as stated
Faking Foursquare accounts with verified GPS trails and social graphs remains difficult
“It's really hard to fake a user with a, like, verified GPS trail going to a business, like, and having these real friends on a social network. You know, like, that's just, it's a very hard thing to fake at the moment, although I, I'm upset to say that, because…”
Blake Shaw Dec 5, 2013 ▶ 19:52
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more