Automatic web article summarizer

#5 Product of the DayAugust 29, 2016

Would you recommend this product?
No reviews yet
Very cool. Here's an accurate 5 sentence summary of my recent blog post: This could be a useful Chrome extension built on Twitter. Instead of clicking through to read articles, simply hover over a shared article in your timeline to get the summary. #savedyouaclick 😊
Upvote (15)Share
@rrhoover this is an interesting idea. So the next thing that I'm currently building on top of this is a Facebook bot. You can send it links and it'll parse and send you the summary. Chrome extension would probably be easier to build though
@rrhoover Absolute +1 to the Chrome extension. I went through the site expecting to find one. Gimme gimme! :D
@jjangsangy @rrhoover this would be cool, the latency of the heroku free plan may make it a bit slow, though. @jjangsangy Btw, I can build that, if you want. Any chance to allow cors requests and return response as json or something?
@oelmekki @rrhoover sure! I can modify the source code to return a json in response to something like POST {"url": "http://summarize-me.com"} to the url endpoint 'http://site-url/api/v1.0/summary'
@jjangsangy Cool, let's do that! I'll have a look tomorrow, after contract time :)
have seen a few of these summarization tools on PH - anybody done a collection yet?
@passingnotes added some more to the list.
This looks amazing. Can you elaborate on what TextRank is? Does it use a neural network, or is it an other kind of NLP method?
@oelmekki thanks for your enthusiasm. TextRank is actually a much simpler than neural networks. It creates a simple language graph of documents and calculates the centroid. Each word in a way is recommends other words in a similar way to PageRank recommending web pages. Here is the original paper written about TextRank, it's very readable. http://web.eecs.umich.edu/%7Emih...
@jjangsangy Thanks :) Skimming through it, I see it uses quite similar method than 5-gram neural networks are using to generate word embeddings. The reconstruction/extraction process looks super cool, though. Will have a deeper look, thanks!
This is awesome! Definitely a time saver! Is there an API?
@parkhmp7 just made one. `https://explaintome.herokuapp.co...` with a JSON containing the `url` field
@jjangsangy I'm getting a 404 error :(
@parkhmp7 what page did the 404 occur for?
@mynhpark smmry has a decent API
Grabs a totally different URL for processing in my tests.