4 ms·

Yelp extends academic dataset

41 points by Zephyr314 12y ago

minimaxir 12y ago
Note the restrictions in the terms of use, which prevent redistribution of the data or any summaries "on any website or other electronic media not covered by this agreement." Would this prohibit making a blog post on an analysis of the data?
- Zephyr314 12y ago
  A blog post analyzing the dataset is fine, and many have been published. Our limitation is on redistribution of Yelp's dataset itself.
  - izyda 12y ago
    What about visualizations of analysis of the data (say using an R Shiny app), ensuring of course, that the raw data could not be downloaded?
    - Zephyr314 12y ago
      Visualizations and analysis of the data are allowed and encouraged. If you are a student you should submit to the challenge as well!
  - minimaxir 12y ago
    Thanks for the clarification! I had done an analysis of Amazon review data of a similar size ( http://minimaxir.com/2014/06/reviewing-reviews/ http://minimaxir.com/2014/06/reviewing-reviews/ ) so it would be interesting to compare/contrast review data and behavior between industries.
rpedela 12y ago
Maybe I am reading it wrong, but the license seems to say that Yelp exclusively owns IP rights to all derivative works. Hopefully I am reading it wrong, but if not then definitely not cool in my opinion.
- Zephyr314 12y ago
  Yelp retains rights to all derivative works of the dataset itself. This would not apply to papers, articles, or algorithms analyzing the dataset.
  - rpedela 12y ago
    Okay, that sounds better. Thanks for clarification.
  - brendano 12y ago
    What does derivative work mean? Why is a graph or data summary not a derivative work?
    - Zephyr314 12y ago
      http://en.wikipedia.org/wiki/Derivative_work http://en.wikipedia.org/wiki/Derivative_work
danso 12y ago
This is great...I'm teaching a couple of classes next year on data analysis and it's incredibly helpful to have real-world data that is fun...things like Census/NOAA data are great, but too abstract (initially) for the average novice to really ask interesting questions of. But everyone knows what it's like to eat at a crappy/great restaurant, or where such places might be located, or how people might review as a cluster...and so everyone comes in with testable assumptions and hypotheses that are fun to explore. This item brought to mind yesterday's front page post on "Seven habits of highly fraudulent users" (https://news.ycombinator.com/item?id=8116047 https://news.ycombinator.com/item?id=8116047)...does Yelp do any extra processing of this academic set (beyond whatever regular cleaning they do of spam accounts)? It'd be interesting to test hypotheses on signals of spammy/fake accounts (OTOH, I imagine Yelp would probably prefer such trends not to be so apparent in bulk data)
- Zephyr314 12y ago
  You can imagine this as a scrape of our website for the specific businesses listed (and their associated recommended reviews and tips, and the associated users for that content) on the date the dataset was generated (which you can assume is the last timestamp). For more information about recommended reviews please check out: http://officialblog.yelp.com/2013/11/yelp-recommended-reviews.html http://officialblog.yelp.com/2013/11/yelp-recommended-review... Let me know if you have any other questions!
fiatjaf 12y ago
Less research is needed: http://blogs.plos.org/speakingofmedicine/2012/06/25/less-research-is-needed/ http://blogs.plos.org/speakingofmedicine/2012/06/25/less-res...
- izyda 12y ago
  I think this competition is a good examples of what the author advocates for - asking the right questions, not the tireless ones. There's no specific prompt and no cliche single purpose to the exercise. Further, the competition is aimed at students - which means in all likelihood, most projects will be relatively short, succinct examples of cool things that can be done with the data (the door is not open to more data collection, to studies that require a lot of time, and the dataset is small enough that computing time/infrastructure should not be an issue). The real challenge is having the creativity to come up with a question worth asking. From Yelp's perspective, I imagine the most value they get from hosting this competition is the (potentially monetizable) ideas for what types of business problems could be solved using their treasure trove of data. I'm sure they could have their data science team build an incredibly sophisticated model for some specific prediction task (probably more sophisticated than what any student could reasonably submit) - but if that specific prediction task turns out difficult or not monetizable, that's a huge investment lost. Having many students look at many different ideas is more likely to result in finding the right question.

41 points by Zephyr314 12y ago

Note the restrictions in the terms of use, which prevent redistribution of the data or any summaries "on any website or other electronic media not covered by this agreement." Would this prohibit making a blog post on an analysis of the data?

Zephyr314 12y ago
A blog post analyzing the dataset is fine, and many have been published. Our limitation is on redistribution of Yelp's dataset itself.
- izyda 12y ago
  What about visualizations of analysis of the data (say using an R Shiny app), ensuring of course, that the raw data could not be downloaded?
  - Zephyr314 12y ago
    Visualizations and analysis of the data are allowed and encouraged. If you are a student you should submit to the challenge as well!
- minimaxir 12y ago
  Thanks for the clarification! I had done an analysis of Amazon review data of a similar size ( http://minimaxir.com/2014/06/reviewing-reviews/ http://minimaxir.com/2014/06/reviewing-reviews/ ) so it would be interesting to compare/contrast review data and behavior between industries.

rpedela 12y ago

Maybe I am reading it wrong, but the license seems to say that Yelp exclusively owns IP rights to all derivative works. Hopefully I am reading it wrong, but if not then definitely not cool in my opinion.

Zephyr314 12y ago
Yelp retains rights to all derivative works of the dataset itself. This would not apply to papers, articles, or algorithms analyzing the dataset.
- rpedela 12y ago
  Okay, that sounds better. Thanks for clarification.
- brendano 12y ago
  What does derivative work mean? Why is a graph or data summary not a derivative work?
  - Zephyr314 12y ago
    http://en.wikipedia.org/wiki/Derivative_work http://en.wikipedia.org/wiki/Derivative_work

danso 12y ago

This is great...I'm teaching a couple of classes next year on data analysis and it's incredibly helpful to have real-world data that is fun...things like Census/NOAA data are great, but too abstract (initially) for the average novice to really ask interesting questions of. But everyone knows what it's like to eat at a crappy/great restaurant, or where such places might be located, or how people might review as a cluster...and so everyone comes in with testable assumptions and hypotheses that are fun to explore. This item brought to mind yesterday's front page post on "Seven habits of highly fraudulent users" (https://news.ycombinator.com/item?id=8116047 https://news.ycombinator.com/item?id=8116047)...does Yelp do any extra processing of this academic set (beyond whatever regular cleaning they do of spam accounts)? It'd be interesting to test hypotheses on signals of spammy/fake accounts (OTOH, I imagine Yelp would probably prefer such trends not to be so apparent in bulk data)

Zephyr314 12y ago
You can imagine this as a scrape of our website for the specific businesses listed (and their associated recommended reviews and tips, and the associated users for that content) on the date the dataset was generated (which you can assume is the last timestamp). For more information about recommended reviews please check out: http://officialblog.yelp.com/2013/11/yelp-recommended-reviews.html http://officialblog.yelp.com/2013/11/yelp-recommended-review... Let me know if you have any other questions!

fiatjaf 12y ago

Less research is needed: http://blogs.plos.org/speakingofmedicine/2012/06/25/less-research-is-needed/ http://blogs.plos.org/speakingofmedicine/2012/06/25/less-res...

izyda 12y ago
I think this competition is a good examples of what the author advocates for - asking the right questions, not the tireless ones. There's no specific prompt and no cliche single purpose to the exercise. Further, the competition is aimed at students - which means in all likelihood, most projects will be relatively short, succinct examples of cool things that can be done with the data (the door is not open to more data collection, to studies that require a lot of time, and the dataset is small enough that computing time/infrastructure should not be an issue). The real challenge is having the creativity to come up with a question worth asking. From Yelp's perspective, I imagine the most value they get from hosting this competition is the (potentially monetizable) ideas for what types of business problems could be solved using their treasure trove of data. I'm sure they could have their data science team build an incredibly sophisticated model for some specific prediction task (probably more sophisticated than what any student could reasonably submit) - but if that specific prediction task turns out difficult or not monetizable, that's a huge investment lost. Having many students look at many different ideas is more likely to result in finding the right question.