StarZero All shows Try the podcast clipper →

Nov 23, 2015 · 25m · mad

Forecasting Content with Mashable's Chief Data Scientist

Haile Wusu · 20m spoken Matt Turck · 34s 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 DataDrivenNYC event, Mashable's Chief Data Scientist Dr. Haile Owusu presents how historical weather forecasting methodologies inspire modern predictive analytics and Mashable's proprietary Velocity engine to forecast digital content virality.

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 2.6% of the talking time here. How this is scored →

Matt as informed peer 0.5 Guest teaching 0.3 Guest disagreement 0.3 Matt pushing back 0.2
05100:0010:0020:000:00–2:58 · Matt as informed peer 0/10 Event Title Card and Speaker Introductions Haile Wusu opens his presentation discussing the history of weather forecasting and colloquially dismissing early techniques as bullshit. Host Matt Turck does not participate in this presentation segment.2:58–9:25 · Matt as informed peer 0/10 Admiral Robert FitzRoy and Early Weather Reports Haile Wusu delivers a monologue detailing Admiral Robert FitzRoy's pioneering weather reports and ultimate suicide. Host Matt Turck remains absent during the talk.9:25–13:02 · Matt as informed peer 0/10 Evolution of Modern Meteorological Predictive Models Haile Wusu connects meteorological chaos theory to content prediction challenges at Mashable. The host does not speak during this segment.13:02–15:10 · Matt as informed peer 0/10 Network Graph Topology and Social Content Diffusion Haile Wusu presents graph topologies and research on social content diffusion. Host Matt Turck is completely silent.15:10–17:26 · Matt as informed peer 0/10 Evaluating Velocity Engine Performance and Accuracy Metrics Haile Wusu reviews accuracy metrics for Mashable's Velocity system compared against Facebook internal benchmarks. Host Matt Turck does not intervene.17:26–25:37 · Matt as informed peer 3/10 Mashable Knowledge Graph and Presentation Conclusion Host Matt Turck moderates Q&A, asking a structured question about build versus buy infrastructure. Haile Wusu then answers audience questions collaboratively.0:00–2:58 · Guest teaching 0/10 Event Title Card and Speaker Introductions Haile Wusu opens his presentation discussing the history of weather forecasting and colloquially dismissing early techniques as bullshit. Host Matt Turck does not participate in this presentation segment.2:58–9:25 · Guest teaching 0/10 Admiral Robert FitzRoy and Early Weather Reports Haile Wusu delivers a monologue detailing Admiral Robert FitzRoy's pioneering weather reports and ultimate suicide. Host Matt Turck remains absent during the talk.9:25–13:02 · Guest teaching 0/10 Evolution of Modern Meteorological Predictive Models Haile Wusu connects meteorological chaos theory to content prediction challenges at Mashable. The host does not speak during this segment.13:02–15:10 · Guest teaching 0/10 Network Graph Topology and Social Content Diffusion Haile Wusu presents graph topologies and research on social content diffusion. Host Matt Turck is completely silent.15:10–17:26 · Guest teaching 0/10 Evaluating Velocity Engine Performance and Accuracy Metrics Haile Wusu reviews accuracy metrics for Mashable's Velocity system compared against Facebook internal benchmarks. Host Matt Turck does not intervene.17:26–25:37 · Guest teaching 2/10 Mashable Knowledge Graph and Presentation Conclusion Host Matt Turck moderates Q&A, asking a structured question about build versus buy infrastructure. Haile Wusu then answers audience questions collaboratively.0:00–2:58 · Guest disagreement 1/10 Event Title Card and Speaker Introductions Haile Wusu opens his presentation discussing the history of weather forecasting and colloquially dismissing early techniques as bullshit. Host Matt Turck does not participate in this presentation segment.2:58–9:25 · Guest disagreement 0/10 Admiral Robert FitzRoy and Early Weather Reports Haile Wusu delivers a monologue detailing Admiral Robert FitzRoy's pioneering weather reports and ultimate suicide. Host Matt Turck remains absent during the talk.9:25–13:02 · Guest disagreement 0/10 Evolution of Modern Meteorological Predictive Models Haile Wusu connects meteorological chaos theory to content prediction challenges at Mashable. The host does not speak during this segment.13:02–15:10 · Guest disagreement 0/10 Network Graph Topology and Social Content Diffusion Haile Wusu presents graph topologies and research on social content diffusion. Host Matt Turck is completely silent.15:10–17:26 · Guest disagreement 0/10 Evaluating Velocity Engine Performance and Accuracy Metrics Haile Wusu reviews accuracy metrics for Mashable's Velocity system compared against Facebook internal benchmarks. Host Matt Turck does not intervene.17:26–25:37 · Guest disagreement 1/10 Mashable Knowledge Graph and Presentation Conclusion Host Matt Turck moderates Q&A, asking a structured question about build versus buy infrastructure. Haile Wusu then answers audience questions collaboratively.0:00–2:58 · Matt pushing back 0/10 Event Title Card and Speaker Introductions Haile Wusu opens his presentation discussing the history of weather forecasting and colloquially dismissing early techniques as bullshit. Host Matt Turck does not participate in this presentation segment.2:58–9:25 · Matt pushing back 0/10 Admiral Robert FitzRoy and Early Weather Reports Haile Wusu delivers a monologue detailing Admiral Robert FitzRoy's pioneering weather reports and ultimate suicide. Host Matt Turck remains absent during the talk.9:25–13:02 · Matt pushing back 0/10 Evolution of Modern Meteorological Predictive Models Haile Wusu connects meteorological chaos theory to content prediction challenges at Mashable. The host does not speak during this segment.13:02–15:10 · Matt pushing back 0/10 Network Graph Topology and Social Content Diffusion Haile Wusu presents graph topologies and research on social content diffusion. Host Matt Turck is completely silent.15:10–17:26 · Matt pushing back 0/10 Evaluating Velocity Engine Performance and Accuracy Metrics Haile Wusu reviews accuracy metrics for Mashable's Velocity system compared against Facebook internal benchmarks. Host Matt Turck does not intervene.17:26–25:37 · Matt pushing back 1/10 Mashable Knowledge Graph and Presentation Conclusion Host Matt Turck moderates Q&A, asking a structured question about build versus buy infrastructure. Haile Wusu then answers audience questions collaboratively.

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

0:00 · Matt 0% · guest 100%0:00 · Matt 0% · guest 100%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 0% · guest 100%12:00 · Matt 0% · guest 100%15:00 · Matt 0% · guest 100%15:00 · Matt 0% · guest 100%18:00 · Matt 15.2% · guest 84.8%18:00 · Matt 15.2% · guest 84.8%21:00 · Matt 7% · guest 93%21:00 · Matt 7% · guest 93%24:00 · Matt 2.4% · guest 97.6%24:00 · Matt 2.4% · guest 97.6%
Sharpest disagreement ▶ 2:36 Dismissing traditional weather tricks

Haile Wusu forcefully rejects historical weather prediction methods such as putting frogs in jars, calling the concept complete bullshit.

 Hardest push from Matt ▶ 18:18 Questioning build versus buy strategy

Matt Turck steps in post-presentation to probe whether Mashable built their prediction engine and underlying infrastructure in-house or relied on vendors.

 Biggest teaching moment ▶ 20:10 Clarifying network graph edge definitions

Haile Wusu gently corrects an audience member's assumption about binomial relationships, explaining downstream viewing metrics and graph node edges.

 Matt holds his own ▶ 18:21 Framing technical build vs buy question

Matt Turck demonstrates domain awareness by asking a targeted question on machine learning infrastructure development versus vendor licensing.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Event Title Card and Speaker Introductions 0010 Haile Wusu opens his presentation discussing the history of weather forecasting and colloquially dismissing early techniques as bullshit. Host Matt Turck does not participate in this presentation segment.
Admiral Robert FitzRoy and Early Weather Reports 0000 Haile Wusu delivers a monologue detailing Admiral Robert FitzRoy's pioneering weather reports and ultimate suicide. Host Matt Turck remains absent during the talk.
Evolution of Modern Meteorological Predictive Models 0000 Haile Wusu connects meteorological chaos theory to content prediction challenges at Mashable. The host does not speak during this segment.
Network Graph Topology and Social Content Diffusion 0000 Haile Wusu presents graph topologies and research on social content diffusion. Host Matt Turck is completely silent.
Evaluating Velocity Engine Performance and Accuracy Metrics 0000 Haile Wusu reviews accuracy metrics for Mashable's Velocity system compared against Facebook internal benchmarks. Host Matt Turck does not intervene.
Mashable Knowledge Graph and Presentation Conclusion 3211 Host Matt Turck moderates Q&A, asking a structured question about build versus buy infrastructure. Haile Wusu then answers audience questions collaboratively.

Statements from this episode (8)

Assertion Not checkable as stated
Mashable crawls about a million pieces of content daily
“So we crawl a, on the order of about a million pieces of content daily.”
Haile Wusu Nov 23, 2015 ▶ 12:25
Disclosure
Mashable models graph growth across tens of thousands of URLs daily
“And then do this replicated over tens of thousands of URLs daily so that we can get an aggregation of which are the better performers.”
Haile Wusu Nov 23, 2015 ▶ 14:00
Assertion Not checkable as stated
Mashable's Velocity engine reaches 75% content prediction accuracy within five minutes
“We are able over within about five minutes to get about 75% accurate with Our predictions for overall sharing behavior with increasing accuracy over time.”
Haile Owusu Nov 23, 2015 ▶ 16:58
Assertion Not checkable as stated
Mashable content predictions compare favorably to Facebook despite lacking granular graph data
“And these results actually compare favorably with the results found by folks at Facebook, for example, who actually have all of the granular information that we don't have about the actual graph structure on which these cascades happen.”
Haile Owusu Nov 23, 2015 ▶ 17:10
Disclosure
Mashable plans to expand content modeling using Knowledge Graph data
“We are looking to expand our ability to model this complicated problem out using information from something we call Knowledge Graph, where we collect very, very granular information about sharing behavior on just our content.”
Haile Wusu Nov 23, 2015 ▶ 17:27
Disclosure
Owusu says Mashable built its Velocity predictive engine completely in-house
“The entire structure of this Of this app Velocity was built in-house. I mean, really we built it from the ground up. I had the great pleasure of working with exceedingly talented people in, in particular, our chief chief architect built this from the ground up…”
Haile Owusu Nov 23, 2015 ▶ 18:43
Disclosure
Owusu: Mashable licensed Velocity out and built models without pre-existing frameworks
“Because we obviously have an interest in, and have actually licensed this license velocity out, we haven't used any pre-existing models for predictions.”
Haile Owusu Nov 23, 2015 ▶ 19:17
Assertion Not checkable as stated
Velocity prediction engine sources the overwhelming majority of Mashable's top content
“And we actually find that the top end of our content, those stories that are viewed very widely and shared very, very widely, the overwhelming majority of those are sourced from Velocity by our editorial staff.”
Haile Owusu Nov 23, 2015 ▶ 22:52
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