Jun 19, 2015 · 27m · mad
Joseph Essas, OpenTable // Mining Diner Talk (Hosted by FirstMark Capital)
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
OpenTable CTO Joseph Essas presents how the company utilizes open-source big data architecture, natural language processing, and machine learning to analyze verified diner feedback. By extracting topics, culinary trends, and sentiment from reviews and reservation notes, OpenTable bridges the gap between restaurant marketing and diner expectations while powering personalized dining experiences.
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 1.2% of the talking time here. How this is scored →
Matt as informed peer 0.5 Guest teaching 0.5 Guest disagreement 0.0 Matt pushing back 0.1
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 0:00 Joking pushback on time limits
Joseph playfully resists host Matt Turck's strict 23-minute time limit before starting his presentation.
Hardest push from Matt ▶ 22:58 Host steering topic to weather modelingMatt interjects into Joseph's response to specifically direct the conversation toward OpenTable's weather prediction capabilities.
Biggest teaching moment ▶ 24:00 Explaining OpenTable's revenue breakdownJoseph educates the audience on OpenTable's dual revenue model based on subscription software and per-seat performance fees.
Matt holds his own ▶ 22:58 Citing OpenTable weather prediction dataMatt demonstrates informed knowledge of OpenTable's internal tech stack by highlighting their specific use of weather forecasting for reservation modeling.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Core Mission and Diner vs. Restaurant Perceptions | 0 | 0 | 0 | 0 | Joseph opens his solo presentation introducing OpenTable's scale and data architecture. The host does not participate during this monologue segment. | |
| Unique Value of Verified Diner Reviews | 0 | 0 | 0 | 0 | Joseph explains how OpenTable verifies reviews via seat status and details their topic modeling methodology. The host is absent during the presentation. | |
| Clustering Topics into Core Experiential Categories | 0 | 0 | 0 | 0 | Joseph demonstrates how topic clustering reveals blind spots in how restaurants view themselves versus what diners value. Host is absent during monologue. | |
| Cross-Continental Nuances and the Valentine's Day Steak Phenomenon | 0 | 0 | 0 | 0 | Joseph discusses regional differences and Valentine's Day review data insights regarding steak quality. Host is absent during monologue. | |
| Automated Attribute Extraction and Personalization Profiles | 0 | 0 | 0 | 0 | Joseph reviews dish tag extraction, notable features, and diner personalization profiles. Host remains absent during monologue. | |
| Analyzing Diner Reservation Notes and the Evolution of Hospitality | 0 | 0 | 0 | 0 | Joseph covers shift trends in user reservation notes and illustrates high-end personalized hospitality. Host is absent. | |
| Conclusion of Presentation and Stage Transition | 3 | 2 | 0 | 1 | Host Matt Turck steps in to ask about the data science team structure and prompts Joseph on weather prediction modeling. Joseph answers in detail. | |
| Q&A: OpenTable Revenue Model and OpenTable Pay Experiment | 1 | 2 | 0 | 0 | Matt moderates Q&A from audience members covering OpenTable's revenue model and the OpenTable Pay feature, which Joseph explains collaboratively. |
Statements from this episode (15)
OpenTable operates in about 32,000 restaurants as of June 2015
“It's about 32,000 restaurants, so we provide software that sits inside those restaurants, and allows restaurants to manage their inventory.”
OpenTable seats about 16 million diners per month as of June 2015
“We sit about sixteen million diners a month”
Essas: Restaurants often perceive themselves differently than how diners view them
“The way restaurants talk about themselves is not always how diners view them. And many times it's actually it's actually interesting that restaurants not even thinking about certain things, that diners who go there express themselves kind of on social media an…”
Essas: OpenTable routes event streams through Kafka, Cassandra, and Spark
“All of our events flowing through Kafka, they've been populated into Cassandra, which then we run Spark instances that kind of model on top of the data.”
Essas: OpenTable has 30 million reviews in its database
“We have thirty million reviews in, in our, kind of, in our database.”
Joseph Essas: OpenTable only allows reviews from verified diners
“Every review that a person leaves, they only allow to leave if they, we know for a fact that they actually dined in that restaurant.”
Essas: OpenTable used non-negative matrix factorization to find review topics
“Non-negative matrix factorization allowed us to break things into just make them much simpler and find topics easier.”
Diners in SF, NYC, and Chicago prioritize different restaurant scenery
“In San Francisco, people talk a lot about the view of the Bay, the view of the bridge. In New York, people talk about, ah, a view of the river, or view of the Hudson, ah, while in Chicago, ah, people talk about the city and the lake.”
OpenTable uses review topic modeling to advise restaurants on marketing strategies
“We also can teach our restaurants how to market on, on themselves on using words and descriptions that they wouldn't think about otherwise.”
Essas: Valentine's Day generates OpenTable's most polarized restaurant reviews
“So, our, literally, it's our most polarizing day on reviews basis, based on one-star review versus five-star reviews.”
Essas: OpenTable review data shows a major cauliflower trend in NYC
“For example, New York City, just looked, looked it up before I came to speak here there's a huge trend in cauliflower.”
Essas: OpenTable's model learned 'crispy' and 'moist' carry positive sentiment
“The model started to get smarter, so the model started to understand things like to die for, crispy, and moist are all positive words, which wouldn't, you wouldn't necessarily understand otherwise.”
Essas: OpenTable data shows diners have grown far more specific and demanding
“People used to be very generic. People used to say in our reviews, and that's the beauty of the fact that we have 15 years worth of them, people used to say, I just want a booth sorry, people used to say, I just want to be a romantic setting, or I want to have…”
OpenTable CTO Says Majority of Revenue Comes From Per-Diner Fees
“And so, so that's majority of our revenue comes from the dollar per person, which is, that's how we like it because that is per performance.”
Essas: Power user dining patterns cannot easily predict general consumer choices
“From what we looked at so far, it's pretty hard to make that, that extrapolation from that power diner population into the general population, which much more picky about when they would go and why would they go to certain places.”