Comments

• 

  about 7 years ago by ericp | Reply

  Mixed results on the hack.

  On one hand, I learned how to use Python's nltk library to do natural language processing on the reviews, to the point that I could lex each word. nltk has parsing abilities as well, but I decided to see how far I could get with just lexing and statistical analysis. Also, the flask and sqlalchemy libraries were straightforward Python analogs of Ruby's sinatra and ActiveRecord/Sequel ORMs, nothing new here. And the Yelp API was straightforward to work with.

  The downside was that the Yelp API only exposed the first 10 words of three reviews for each business. If we assume the average business has 100 reviews of about 200 words each, this wasn't going to give me the data I needed. However, each review in the resource returned byhttps://api.yelp.com/v3/businessesBUSINESSID/reviews also contained a URL, and following that URL gave me the full text of 19 reviews.

  And apparently following that URL violated Yelp's general terms of service, but not the developers's ToS, and I was cut off after pulling down reviews for 350 restaurants. At least I randomized my selection procedure, so I ended up with a smattering of Mexican, Chinese, Japanese and American-style restaurants.

  The best generated sentence might have been one of the first: "I travel the bone tender." I also liked "My wife had the chipotle pancakes." But most of the sentences were grammatically incorrect, or made no sense, or both. I did try tweaking the Markov generator to use a mix of single-word and double-word prefixes, but given the lack of data, I ended the hack and went back to work.

×
Edit Comment 2514
# A First Level Header
## A Second Level Header

Use one asterisk to *emphasize*

Use two asterisks for **strong emphasis**

* Use asterisks
* for lists

This is an [example link](http://example.com/)

This is an ![example image](http://paste.opensuse.org/view/raw/68957446)

This is a user link @hans

This is a project link hw#some-cool-title
More Complex Markdown Help
Formatting Help
• Edit
• Preview
Mixed results on the hack. On one hand, I learned how to use Python's `nltk` library to do natural language processing on the reviews, to the point that I could lex each word. `nltk` has parsing abilities as well, but I decided to see how far I could get with just lexing and statistical analysis. Also, the `flask` and `sqlalchemy` libraries were straightforward Python analogs of Ruby's `sinatra` and `ActiveRecord`/`Sequel` ORMs, nothing new here. And the Yelp API was straightforward to work with. The downside was that the Yelp API only exposed the first 10 words of three reviews for each business. If we assume the average business has 100 reviews of about 200 words each, this wasn't going to give me the data I needed. However, each review in the resource returned by`https://api.yelp.com/v3/businessesBUSINESSID/reviews` also contained a URL, and following that URL gave me the full text of 19 reviews. And apparently following that URL violated Yelp's general terms of service, but not the developers's ToS, and I was cut off after pulling down reviews for 350 restaurants. At least I randomized my selection procedure, so I ended up with a smattering of Mexican, Chinese, Japanese and American-style restaurants. The best generated sentence might have been one of the first: "I travel the bone tender." I also liked "My wife had the chipotle pancakes." But most of the sentences were grammatically incorrect, or made no sense, or both. I did try tweaking the Markov generator to use a mix of single-word and double-word prefixes, but given the lack of data, I ended the hack and went back to work.
 
×
Reply to ericp
Mixed results on the hack.
On one hand, I learned how to use Python's nltk library to do natural language processing on the reviews, to the point that I could lex each word. nltk has parsing abilities as well, but I decided to see how far I could get with just lexing and statistical analysis. Also, the flask and sqlalchemy libraries were straightforward Python analogs of Ruby's sinatra and ActiveRecord/Sequel ORMs, nothing new here. And the Yelp API was straightforward to work with.
The downside was that the Yelp API only exposed the first 10 words of three reviews for each business. If we assume the average business has 100 reviews of about 200 words each, this wasn't going to give me the data I needed. However, each review in the resource returned byhttps://api.yelp.com/v3/businessesBUSINESSID/reviews also contained a URL, and following that URL gave me the full text of 19 reviews.
And apparently following that URL violated Yelp's general terms of service, but not the developers's ToS, and I was cut off after pulling down reviews for 350 restaurants. At least I randomized my selection procedure, so I ended up with a smattering of Mexican, Chinese, Japanese and American-style restaurants.
The best generated sentence might have been one of the first: "I travel the bone tender." I also liked "My wife had the chipotle pancakes." But most of the sentences were grammatically incorrect, or made no sense, or both. I did try tweaking the Markov generator to use a mix of single-word and double-word prefixes, but given the lack of data, I ended the hack and went back to work.
---
# A First Level Header
## A Second Level Header

Use one asterisk to *emphasize*

Use two asterisks for **strong emphasis**

* Use asterisks
* for lists

This is an [example link](http://example.com/)

This is an ![example image](http://paste.opensuse.org/view/raw/68957446)

This is a user link @hans

This is a project link hw#some-cool-title
More Complex Markdown Help
Formatting Help
• Edit
• Preview