The Berkeley Document Summarizer: Learning-Based, Single-Document Summarization

Re:The Berkeley Document Summarizer: Learning-Based, Single-... by fowlerpower on Sunday October 02, 2016 @12:31AM (#12621046)

So how does this compare to the stuff Google was doing with document summarization? Is the content unique, meaning does it summarize using brand new words?

It's unclear but still seems Really promising.

[ Parent | on HN ]

Re:The Berkeley Document Summarizer: Learning-Based, Single-... by deepGem on Sunday October 02, 2016 @11:20AM (#12622745)

Just going by the GitHub readme and the corresponding paper this tool does not generate new words, popularly known as the abstraction summary. This is doing an extraction task followed by a syntactic compression task.

Pages 3,4 of the paper http://www.cs.utexas.edu/~gdurrett/papers/durrett-berg-klein...

[ Parent | on HN ]

Re:The Berkeley Document Summarizer: Learning-Based, Single-... by thpalmear on Sunday October 02, 2016 @03:48AM (#12621440)

Google has been to busy trying to lift methods from Berkeley Lab and passing it off as their own, in particular Tomas Mikolov https://www.kaggle.com/c/word2vec-nlp-tutorial/forums/t/1234...

[ Parent | on HN ]

Re:The Berkeley Document Summarizer: Learning-Based, Single-... by wodenokoto on Sunday October 02, 2016 @09:11AM (#12622151)

You need a better argument than "somebody else has been working on word vectors before Google"

No shit.

[ Parent | on HN ]

Re:The Berkeley Document Summarizer: Learning-Based, Single-... by thpalmear on Monday October 03, 2016 @12:24AM (#12625509)

That's not what was said. It's how the feature attributes in the vectors are constructed, scored and ranked in addition to the calculations used to score vectors for similarity.

[ Parent | on HN ]