Hacker News Reader

Comment by brador | Hacker News Reader

Any insight into the tech this is using?

Opening the ipa file, and browsing the contents, there are a couple of plain text files - containing stop words, dictionaries and others on natural language processing.

The summarised results from the app are very random - pick any article other than the test case in the video and you won't get any useful output, 3 random sentences usually.

From DataInfo.rtf in the iap:

StopWords: common words in languages: Swedish, English, Mandarin, Italian, French, German, Spanish Japanese, Russian, Finnish, Dutch.

Special Word: discard pronouns/trigger keywords in all of above languages.

Currently removes all pronoun in sentences

Stop Words From Where:

http://www.ranks.nl/resources/stopwords.html http://www.lextek.com/manuals/onix/stopwords1.html http://www.lextek.com/manuals/onix/stopwords2.html http://www.link-assistant.com/seo-stop-words.html English

http://members.unine.ch/jacques.savoy/clef/index.html http://wiki.apache.org/solr/LanguageAnalysis Other languages

http://hi.baidu.com/zhaocy0113/blog/item/146b5c346a738c4d251... http://tribes.tribe.net/geographicalife/thread/98f6abb6-71bf... Chinese

http://dnnspeedblog.com/SpeedBlog/PostID/3187/Japanese-Stop-...

http://www.ranks.nl/stopwords/japanese.html Japanese

THEN ALL TRANSLATE ENGLISH FIRST STUFF

LENGTH STOP: 19,070

LENGTH GLOBAL: 8059

LENGTH SPECIAL: 362

Good means no change/Draft little change/Bad 000s

I thought it was a stretch that D'Aloiso taught himself the linear algebra, statistics, probability theory, NLP, and machine learning required to implement such an algorithm. Especially since he doesn't take a deep interest in programming.

It appears to be an iteration of his previous app Trimit: http://www.fastcompany.com/1772823/the-15-year-old-brain-beh... http://techcrunch.com/2011/07/15/trimit-summarizes-emails-bl...

I am guessing that he just outsourced the AI/algorithm part based on the articles content: 'we' and 'The Trimit team' and Nick just designed the UI. Overall, great concept, but poor execution (in terms of accuracy) and journalists seem to be buying the hype.

Is there any legal and simple way of obtaining the ipa file without using iTunes (i.e. on Linux)?

http://www.multiupload.com/RWD3XZMOW2

"Summly’s patent pending technology has been evaluated by MIT Researchers and proven state-of-the-art. We utilise ontological detection and machine learning techniques in our summarisation processes.

Our core technology is language independent and produces great results out of the box for any language. In addition, specific optimisations have been made for English, French, Spanish, Italian, Dutch, German, Swedish, Mandarin, Russian, Japanese and Finnish."

http://www.summly.com/en/technology.html

It says that, but to what extend is it the truth? I can put that on my website too if my friend from MIT tells me something I did is state-of-the-art just from hearing me describe it (aka not digging into the code etc).

I'm not saying it isn't true, but from the results the app gives on non demo cases and from the discussion in this thread i'm not sure?

Reply on news.ycombinator.com