5 ms·

How big data can help you pick better wine

65 points by erehweb 9y ago

simonsarris 9y ago
Part of the appeal of good wine for me is the social aspect of discovery, and I don't think I'd want to trade that in for a technical one. I don't want a perfect wine so much as I want to drink a sincere recommendation from a friend or family member, and hopefully have that same wine again with them, and have something to talk about. In fact when I'm with certain people, even bad wine is fondly remembered. "What was that chalk we were drinking last month... I crave it when I think of you."* The funny thing about all this is that his big data quest for more good wine was launched because of the same reason I enjoy wine: It was his brother-in-law's recommendation that found him his favorite, not crunching the #cassis #vanilla #persistentfinish numbers. And he's doing this to find bottles similar to the one from a beloved family member, most of which are far more expensive (eyeballing the graph), but if I were him, I don't know if I could ever find a better bottle than "my brother-in-law's wine." Isn't sentimentality worth more than the data, especially when working with the most fickle data-set of all? Also, come on, don't make a post like this and not give us the list! * (We were drinking Troublemaker, a Paso Robles blend, which I swear is 50% syrah and 50% limestone.)
- gingerlime 9y ago
  You have to start somewhere, but I'd love to be able to get wine recommendations based on my personal favourites + some heuristics that come up with something less obvious or that I wouldn't pick otherwise. I don't care so much about his list, but being able to use the same engine he did with my own starting point would be great. On a side note, I wonder how the recommendations are influenced by the audience. i.e. wine.com is presumably used by US customers, so perhaps it could explain why French wines are described in certain ways compared to California. But you might get the opposite if you look at a European website.
- ghostDancer 9y ago
  I always ask friends and sellers for similar wines , the recommendations are usually good and sometimes with surprises (bad and good).
daxorid 9y ago
"better" wine? https://www.theguardian.com/lifeandstyle/2013/jun/23/wine-tasting-junk-science-analysis https://www.theguardian.com/lifeandstyle/2013/jun/23/wine-ta... http://www.realclearscience.com/blog/2014/08/the_most_infamous_study_on_wine_tasting.html http://www.realclearscience.com/blog/2014/08/the_most_infamo...
- schwap 9y ago
  In the context of the article "better" = tastes most pleasing (to the author)
  - ghostDancer 9y ago
    That's the best definition of better in this case. I have a friend who is a wine maker(enologist) and he always says the same, the best wine is the one you enjoy not the one some magazine/person/web says. You can but the most expensive wine in the world and do not like it then it doesn't matter if it's good or not.
- graeme 9y ago
  Have you seen this? http://www.economist.com/blogs/graphicdetail/2017/05/daily-chart-11 http://www.economist.com/blogs/graphicdetail/2017/05/daily-c... It defies common sense to think all wine is equal quality. The fact that there's much nonsense in the wine world doesn't mean it's all nonsense. We'd need to see much stronger evidence than what has been presented.
  - NikolaeVarius 9y ago
    They were trying to figure out Grape type and origin country not "quality". And even then was less than 50% accurate.
dejv 9y ago
The article display some cool use for NLP, but as a winemaker and wine judge I find methodology funny and flawed. You have to understand that wine descriptions are nothing more then marketing message that have very little resemblance to what is inside of the bottle. It would be interesting to see how those description change during the time as this message is changing all the time to catch the current trend. Wine market is actually quite simple: you have few tiers of wine/wineries: from your basic mass market wines to mid tier to premium and ultra premium. Each category have some price point and higher the price better fruit (should) come to it. There are differences in between the regions, but lower level wines are usually same no matter the marketing message and premium wines are usually very different and special in each own way anyway.
- jsm386 9y ago
  Agree that using marketing descriptions provided by the winery isn't the best approach. It would be much more useful if the data used stuck to just the reviews (though _which_ reviewers are used is its own issue): > A Quid network shows 3,380 nodes, where each node represents a description and review of a single bottle of Bordeaux-style red wine from Wine.com. FWIW, this is a topic we've explored in a few ways at VinePair: * https://vinepair.com/wine-blog/how-much-do-the-labels-on-wine-bottles-influence-what-we-buy/ https://vinepair.com/wine-blog/how-much-do-the-labels-on-win... * https://vinepair.com/articles/wine-industry-touch-consumers/ https://vinepair.com/articles/wine-industry-touch-consumers/ * And a more humorous look at these descriptions @ https://vinepair.com/robert-robot/ https://vinepair.com/robert-robot/
  - dejv 9y ago
    Using user review are not very helpful as well. Each taster use different vocabulary and there are even different way to analyse wine: some people describe the flavours (this wine is strawberries, apple, lime) some, including me, are more of a structural tester (comparing "length" of a wine, width and character of the taste and after taste). Second problem is (and there is even statistical proof) that after wine pass some threshold of kinda good wine, the resulting score is random. This is caused by few things: your mood, what you drink before it, anticipation, the meal and even music. Scoring (and recommending) wine and selecting what you like is very complicated subject and probably best way is still ask some good wine buyer, describe what you like and he might recommend wines he likes that loosely fits your description.
gjem97 9y ago
I actually like this analysis. I think there's something there. But man-oh-man, this is about as "small data" as it gets. Has "big data" become just a short name for an assorted bag of statistical tools?
- socalnate1 9y ago
  >>>Has "big data" become just a short name for an assorted bag of statistical tools? Yes. Or as @mhoye on twitter put it: "FACT: most "big data" is actually two small datas standing on each other shoulders wearing a trenchcoat and sunglasses."
tlanc 9y ago
Does it do better than Ashenfelter's simple formula: https://www.betterment.com/resources/investment-strategy/behavioral-finance-investing-strategy/human-vs-algorithm-investing-a-lesson-from-wine-country/ https://www.betterment.com/resources/investment-strategy/beh...
sin7 9y ago
In 2013, I got taste and aroma for around 1100 wines. I used the data to create flavor profiles and predict the price based on the chemical composition. In general, more expensive wines are more complex and the predicted price matches the actual price quite well. This breaks down on the extremes. Boxed wines such as Franzia and Vella are under priced. Not by much, but their juice is closer to 4-5 dollar bottled wines. At the time of the calculation a bottle of Franzia was around 2 dollars. So, it's a pretty good deal. On the other side, most new world wines over 35 dollars are over priced. There's nothing chemically different to the the wines just below 35 dollars. There were a few wines in the middle that were way off. I remember a Clos Du Bois was under priced by 7 dollars, but most of them were in the right range. The biggest offenders were wines that built their reputation in a well known appellation and moved a level higher. It's very common for wines to start in Sonoma Coast or Carneros and end up with a California appellation.
- cityhall 9y ago
  Did you try the underpriced ones, and were they good? A particular wine can be complex but have some specific quality that makes it unappealing. I'd be surprised if complexity is the main variable between different vintages of the same wine that have very different prices.
  - sin7 9y ago
    I didn't try any. I drink very little wine. I did show the list around and no one was surprised. It wasn't Kendall-Jackson that was under priced. It was smaller wineries that had been getting solid reviews. There must have been some surprises, but I knew very little of wine back then. I also knew very little about machine learning, R or anything else. I knew enough to use Random Forests, R for graphing and the opinion of others to judge the results. I tried finding the file with the list of all the wines and the predicted prices. I didn't even know enough to name the file correctly. I think the large companies are very consistent with their product. They have many sources for their juice and because of that their product will be more average. The small companies can have years where they do everything right and create a great product, but it's not like they can just change their price to reflect the better quality.
kumarski 9y ago
Or make it.... http://avawinery.com http://avawinery.com

65 points by erehweb 9y ago

Part of the appeal of good wine for me is the social aspect of discovery, and I don't think I'd want to trade that in for a technical one. I don't want a perfect wine so much as I want to drink a sincere recommendation from a friend or family member, and hopefully have that same wine again with them, and have something to talk about. In fact when I'm with certain people, even bad wine is fondly remembered. "What was that chalk we were drinking last month... I crave it when I think of you."* The funny thing about all this is that his big data quest for more good wine was launched because of the same reason I enjoy wine: It was his brother-in-law's recommendation that found him his favorite, not crunching the #cassis #vanilla #persistentfinish numbers. And he's doing this to find bottles similar to the one from a beloved family member, most of which are far more expensive (eyeballing the graph), but if I were him, I don't know if I could ever find a better bottle than "my brother-in-law's wine." Isn't sentimentality worth more than the data, especially when working with the most fickle data-set of all? Also, come on, don't make a post like this and not give us the list! * (We were drinking Troublemaker, a Paso Robles blend, which I swear is 50% syrah and 50% limestone.)

gingerlime 9y ago
You have to start somewhere, but I'd love to be able to get wine recommendations based on my personal favourites + some heuristics that come up with something less obvious or that I wouldn't pick otherwise. I don't care so much about his list, but being able to use the same engine he did with my own starting point would be great. On a side note, I wonder how the recommendations are influenced by the audience. i.e. wine.com is presumably used by US customers, so perhaps it could explain why French wines are described in certain ways compared to California. But you might get the opposite if you look at a European website.
ghostDancer 9y ago
I always ask friends and sellers for similar wines , the recommendations are usually good and sometimes with surprises (bad and good).

daxorid 9y ago

"better" wine? https://www.theguardian.com/lifeandstyle/2013/jun/23/wine-tasting-junk-science-analysis https://www.theguardian.com/lifeandstyle/2013/jun/23/wine-ta... http://www.realclearscience.com/blog/2014/08/the_most_infamous_study_on_wine_tasting.html http://www.realclearscience.com/blog/2014/08/the_most_infamo...

schwap 9y ago
In the context of the article "better" = tastes most pleasing (to the author)
- ghostDancer 9y ago
  That's the best definition of better in this case. I have a friend who is a wine maker(enologist) and he always says the same, the best wine is the one you enjoy not the one some magazine/person/web says. You can but the most expensive wine in the world and do not like it then it doesn't matter if it's good or not.
graeme 9y ago
Have you seen this? http://www.economist.com/blogs/graphicdetail/2017/05/daily-chart-11 http://www.economist.com/blogs/graphicdetail/2017/05/daily-c... It defies common sense to think all wine is equal quality. The fact that there's much nonsense in the wine world doesn't mean it's all nonsense. We'd need to see much stronger evidence than what has been presented.
- NikolaeVarius 9y ago
  They were trying to figure out Grape type and origin country not "quality". And even then was less than 50% accurate.

dejv 9y ago

The article display some cool use for NLP, but as a winemaker and wine judge I find methodology funny and flawed. You have to understand that wine descriptions are nothing more then marketing message that have very little resemblance to what is inside of the bottle. It would be interesting to see how those description change during the time as this message is changing all the time to catch the current trend. Wine market is actually quite simple: you have few tiers of wine/wineries: from your basic mass market wines to mid tier to premium and ultra premium. Each category have some price point and higher the price better fruit (should) come to it. There are differences in between the regions, but lower level wines are usually same no matter the marketing message and premium wines are usually very different and special in each own way anyway.

jsm386 9y ago
Agree that using marketing descriptions provided by the winery isn't the best approach. It would be much more useful if the data used stuck to just the reviews (though _which_ reviewers are used is its own issue): > A Quid network shows 3,380 nodes, where each node represents a description and review of a single bottle of Bordeaux-style red wine from Wine.com. FWIW, this is a topic we've explored in a few ways at VinePair: * https://vinepair.com/wine-blog/how-much-do-the-labels-on-wine-bottles-influence-what-we-buy/ https://vinepair.com/wine-blog/how-much-do-the-labels-on-win... * https://vinepair.com/articles/wine-industry-touch-consumers/ https://vinepair.com/articles/wine-industry-touch-consumers/ * And a more humorous look at these descriptions @ https://vinepair.com/robert-robot/ https://vinepair.com/robert-robot/
- dejv 9y ago
  Using user review are not very helpful as well. Each taster use different vocabulary and there are even different way to analyse wine: some people describe the flavours (this wine is strawberries, apple, lime) some, including me, are more of a structural tester (comparing "length" of a wine, width and character of the taste and after taste). Second problem is (and there is even statistical proof) that after wine pass some threshold of kinda good wine, the resulting score is random. This is caused by few things: your mood, what you drink before it, anticipation, the meal and even music. Scoring (and recommending) wine and selecting what you like is very complicated subject and probably best way is still ask some good wine buyer, describe what you like and he might recommend wines he likes that loosely fits your description.

gjem97 9y ago

I actually like this analysis. I think there's something there. But man-oh-man, this is about as "small data" as it gets. Has "big data" become just a short name for an assorted bag of statistical tools?

socalnate1 9y ago
>>>Has "big data" become just a short name for an assorted bag of statistical tools? Yes. Or as @mhoye on twitter put it: "FACT: most "big data" is actually two small datas standing on each other shoulders wearing a trenchcoat and sunglasses."

tlanc 9y ago

Does it do better than Ashenfelter's simple formula: https://www.betterment.com/resources/investment-strategy/behavioral-finance-investing-strategy/human-vs-algorithm-investing-a-lesson-from-wine-country/ https://www.betterment.com/resources/investment-strategy/beh...

sin7 9y ago

In 2013, I got taste and aroma for around 1100 wines. I used the data to create flavor profiles and predict the price based on the chemical composition. In general, more expensive wines are more complex and the predicted price matches the actual price quite well. This breaks down on the extremes. Boxed wines such as Franzia and Vella are under priced. Not by much, but their juice is closer to 4-5 dollar bottled wines. At the time of the calculation a bottle of Franzia was around 2 dollars. So, it's a pretty good deal. On the other side, most new world wines over 35 dollars are over priced. There's nothing chemically different to the the wines just below 35 dollars. There were a few wines in the middle that were way off. I remember a Clos Du Bois was under priced by 7 dollars, but most of them were in the right range. The biggest offenders were wines that built their reputation in a well known appellation and moved a level higher. It's very common for wines to start in Sonoma Coast or Carneros and end up with a California appellation.

cityhall 9y ago
Did you try the underpriced ones, and were they good? A particular wine can be complex but have some specific quality that makes it unappealing. I'd be surprised if complexity is the main variable between different vintages of the same wine that have very different prices.
- sin7 9y ago
  I didn't try any. I drink very little wine. I did show the list around and no one was surprised. It wasn't Kendall-Jackson that was under priced. It was smaller wineries that had been getting solid reviews. There must have been some surprises, but I knew very little of wine back then. I also knew very little about machine learning, R or anything else. I knew enough to use Random Forests, R for graphing and the opinion of others to judge the results. I tried finding the file with the list of all the wines and the predicted prices. I didn't even know enough to name the file correctly. I think the large companies are very consistent with their product. They have many sources for their juice and because of that their product will be more average. The small companies can have years where they do everything right and create a great product, but it's not like they can just change their price to reflect the better quality.

kumarski 9y ago

Or make it.... http://avawinery.com http://avawinery.com