Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Teaching a Computer to Read: NLP Hacking in Python

blog.scripted.com

Post reply on HN
Print view View original

11–20 of 25 posts

Re: Teaching a Computer to Read: NLP Hacking in Python

#11

Post by drakaal »

adpreese wrote: post #9
drakaal wrote: post #5

The CPU cost to do use this approach is terribly high. I don't think this approach is going to give better results than a few simple rules and NLTK would. This API will do a better job telling you what an article is about. https://www.mashape.com/stremor/stremor-noun-phrase-and-part... That said, the approach we use for our TLDR software and search rankings doesn't rely on just frequency, the adjectives that amplify…

I tried to test your example with your API, but it requires a credit card even for the freemium plan. Is there any way you can make a rate limited API that never charges to avoid that? I'm not familiar with mashape so it may not be possible.

I can't easily. The free quota is pretty high on their. (5000 calls), For development I'd work off a few test cases that you save the response on, then if you get something working decide if 5000 calls a month will fill your needs.

Re: Teaching a Computer to Read: NLP Hacking in Python

#12

Post by drakaal »

adpreese wrote: post #9
drakaal wrote: post #5

The CPU cost to do use this approach is terribly high. I don't think this approach is going to give better results than a few simple rules and NLTK would. This API will do a better job telling you what an article is about. https://www.mashape.com/stremor/stremor-noun-phrase-and-part... That said, the approach we use for our TLDR software and search rankings doesn't rely on just frequency, the adjectives that amplify…

I tried to test your example with your API, but it requires a credit card even for the freemium plan. Is there any way you can make a rate limited API that never charges to avoid that? I'm not familiar with mashape so it may not be possible.

Saw you are trying it out. Awesome! Sorry the documentation is a bit weak right now, we had people wanting it so we got it out, rather than getting all the docs complete.

Re: Teaching a Computer to Read: NLP Hacking in Python

#13

Post by msalahi »

drakaal wrote: post #7
msalahi wrote: post #6

Earlier quoted context omitted.

i've actually found the performance of gensim (the topic modeling python module i use here) to be pretty great. we're not at a scale where CPU performance is make or break just yet, so i haven't done any comprehensive testing of performance. but i've definitely not run into any performance issues worth complaining about. however, gensim is 100% based on lazy evaluation where it can be, so it's relatively light on the…

The issue with Genism is you have to know what you are trying to analyze before you analyze it. It doesn't do well if you use the wrong corpus or if like you mention start with a million word corpus. If you were analyzing emails in a single organization all day you could probably sort out topics really well. Doing all of the web it breaks down because it gets less accurate the larger the variety of content.

"doing all of the web" will cause pretty much any approach to AI/machine learning/NLP to break down. i'm a big believer in it being the responsibility of the engineer employing these techniques to take stock of the problem at hand and find out what constraints you can take advantage of to achieve better performance/accuracy/prettiness of code. there's not really a silver bullet that you can just release on the internet with the task of bringing back incredibly useful information without "knowing what you're trying to analyze before you analyze it."

Re: Teaching a Computer to Read: NLP Hacking in Python

#14

Post by drakaal »

msalahi wrote: post #13
drakaal wrote: post #7

Earlier quoted context omitted.

The issue with Genism is you have to know what you are trying to analyze before you analyze it. It doesn't do well if you use the wrong corpus or if like you mention start with a million word corpus. If you were analyzing emails in a single organization all day you could probably sort out topics really well. Doing all of the web it breaks down because it gets less accurate the larger the variety of content.

"doing all of the web" will cause pretty much any approach to AI/machine learning/NLP to break down. i'm a big believer in it being the responsibility of the engineer employing these techniques to take stock of the problem at hand and find out what constraints you can take advantage of to achieve better performance/accuracy/prettiness of code. there's not really a silver bullet that you can just release on the intern…

Wait an hour. We decided to push that bit of code live in Alpha. :-)

Re: Teaching a Computer to Read: NLP Hacking in Python

#15

Post by msalahi »

drakaal wrote: post #14
msalahi wrote: post #13

Earlier quoted context omitted.

"doing all of the web" will cause pretty much any approach to AI/machine learning/NLP to break down. i'm a big believer in it being the responsibility of the engineer employing these techniques to take stock of the problem at hand and find out what constraints you can take advantage of to achieve better performance/accuracy/prettiness of code. there's not really a silver bullet that you can just release on the intern…

Wait an hour. We decided to push that bit of code live in Alpha. :-)

SCIENCE!

Re: Teaching a Computer to Read: NLP Hacking in Python

#16

Post by adpreese »

drakaal wrote: post #12
adpreese wrote: post #9

Earlier quoted context omitted.

I tried to test your example with your API, but it requires a credit card even for the freemium plan. Is there any way you can make a rate limited API that never charges to avoid that? I'm not familiar with mashape so it may not be possible.

Saw you are trying it out. Awesome! Sorry the documentation is a bit weak right now, we had people wanting it so we got it out, rather than getting all the docs complete.

I did try it out. It does a good job of pulling out different bits and categorizing them. I went ahead and ran the example you had and put it up to continue the conversation(https://gist.github.com/adpreese/6722561). If you want me to take it down, I will certainly respect that but I thought it'd be convenient for anyone else paying attention.

The noun phrases part of the response gave a concise list of things, including the word thing(hijacked, NLP thread, stuff, My NLP, thing, Drakaal, cooler, heuristics, Just). That's maybe good at picking out the nouns, but it's not really actionable yet. It might be great as the bag of words to use for trying to classify something, but by itself the best it's giving me is NLP/heuristics if I had the concepts grouped together, somehow. I think that's a reasonable takeaway from your example, but I'd be curious what your thoughts are on it.

PS. I tried the TLDR API on a copy and paste of the original article with commas, periods, and single and double quotes removed but it returned a 500 error. I'm probably doing something wrong, but I'd love to see what it spits out if you can help me out.

Re: Teaching a Computer to Read: NLP Hacking in Python

#17

Post by jlees »

msalahi wrote: post #4
msalahi wrote: post #3

Earlier quoted context omitted.

As with most successful applications of machine learning, it's about finessing your approach based on the problem at hand. In our case, we have classes divided on the level of "Medicine," "Real Estate," etc. So, we could throw away lots of words that only occurred once or twice in the massive corpus we crawled to build the language model and still have a pretty robust representation of the subject you're trying to re…

In fact, if your training corpus is sufficiently large, you'd be shocked how many words you can eliminate right away for a term frequency of one or two. I went from millions of words in the vocabulary to something like 60k just by ignoring words that happen once or twice in the corpus. Plus, you probably won't learn much about the relationships between words if they only occur a few times in the corpus.

Yeah, but consider that some rare words are much stronger indicators of topic than more common ones. Even more so if you look at n-grams. If you use something like wordnet you can get a lot of meaning out of low-frequency words and throw away the meaningless higher-frequency ones that occur in too many categories to be useful.

Re: Teaching a Computer to Read: NLP Hacking in Python

#18

Post by samuizo »

drakaal wrote: post #5

The CPU cost to do use this approach is terribly high. I don't think this approach is going to give better results than a few simple rules and NLTK would. This API will do a better job telling you what an article is about. https://www.mashape.com/stremor/stremor-noun-phrase-and-part... That said, the approach we use for our TLDR software and search rankings doesn't rely on just frequency, the adjectives that amplify…

While CPU cost is a concern, often memory and correspondingly IO is the bottleneck in vector space approaches. Practitioners can leverage highly-optimized libraries for performing the matrix decompositions, so random disk seeks become more of a concern than CPU time. It's in iterative SVD that Gensim really shines in my opinion.

Turning to your example, any model based on term frequencies, vector space treatments included, would have trouble identifying 'Drakaal' as the most important term. But, this can be mitigated to some extent by preprocessing. In particular, naive coreference resolution would simply assign 'Drakaal' to every occurrence of 'he'/'his' in the sentence (since there are no other candidates). In which case, the count of 'Drakaal' jumps from 1 to 5. Just taking the comments in this thread as the corpus, that's a pretty high frequency for a single document, which might indeed get it to stand out on that basis alone.

Now whether we could get even more nuanced and determine that it's not just about 'Drakaal' but also a certain disposition toward him really depends on the task. If it's important to uncover those sorts of patterns then I would incorporate some documents that are illustrative of the distinction. In this sense, vector space approaches can be both purely exploratory as well as guided toward the divisions you aim for.

Re: Teaching a Computer to Read: NLP Hacking in Python

#19

Post by samuizo »

adpreese wrote: post #2

How do you deal with keeping your super rare words list sensible? For many forms of technical writing I could see things getting out of hand where you have lots of tiny dense clusters not really close to anything else if you didn't manage the list well.

You can often strike a balance between rare words that appear in only a couple of documents and very frequent words that occur all over the place by employing both a term frequency and a document frequency weighting scheme; 'tf-idf' in the nomenclature [1].

The basic idea is that you keep track of counts both within documents and among documents. For English, word like 'the' will be frequent in each document it occurs in. It will also occur in every document. The high document frequency counteracts the high term frequency. On the other hand, 'motherboard' might be infrequent overall (but not extremely so), but its low document frequency boosts its importance.

The scheme is commonly employed and works quite well, sometimes obviating the need for careful vocabulary pruning. FWIW, scikit-learn implements it in their feature extraction library [2].

[1] http://en.wikipedia.org/wiki/Tf–idf‎ [2] http://scikit-learn.org/stable/modules/generated/sklearn.fea...

Re: Teaching a Computer to Read: NLP Hacking in Python

#20

Post by agibsonccc »

msalahi wrote: post #13
drakaal wrote: post #7

Earlier quoted context omitted.

The issue with Genism is you have to know what you are trying to analyze before you analyze it. It doesn't do well if you use the wrong corpus or if like you mention start with a million word corpus. If you were analyzing emails in a single organization all day you could probably sort out topics really well. Doing all of the web it breaks down because it gets less accurate the larger the variety of content.

"doing all of the web" will cause pretty much any approach to AI/machine learning/NLP to break down. i'm a big believer in it being the responsibility of the engineer employing these techniques to take stock of the problem at hand and find out what constraints you can take advantage of to achieve better performance/accuracy/prettiness of code. there's not really a silver bullet that you can just release on the intern…

Web Developers are a neat bunch. It's amazing what kind of inference you can do with exploiting document structure to make different kind of inferences alongside more traditional approaches like word frequency analysis, LDA, or even deep learning/word distributional inference. NLP on the web especially question answering and search, can still be greatly expanded upon.
Post reply on HN