Hacker News Reader

Comment by rahooligan | Hacker News Reader

In the case of Moonwalking with Einstein over 70% of critics gave it a positive.

If 7 out of 10 critic reviews are positive, the rating is 70%. We recommend a book if it scores over 70%. More info here: http://idreambooks.com/toread_or_nottoread

Thumbs up/down doesn't personalize yet...

View on HN

Sorry, I think you misunderstood me, I am specifically referring to the New York Times review of it, which is one of the 4 negative reviews of the book, but if you read it is in actual fact positive. My mistake though, because I accidentally linked the same page twice when I wanted to link this the second time: http://www.nytimes.com/2011/03/13/books/review/book-review-m...

We have a strict set of criteria when rating a review as positive or negative. In general, if the review is even slightly negative we rate it as such...in the case of Moonwalking, NYT expresses a few disappointments about the book. This is to help surface only the very best of books in each genre.

I'm not sure how giving such a binary 'good' or 'bad' to a review is helpful for books. Reviews could be on a continuum (hence the star rating system that many folks use) and calling a review negative just because of a few negative opinions voiced in a review could potentially miss the point of the review.

Any review of sufficient length and thoroughness is likely to include at least some criticism, if for no other reason than the reviewer not wanting to look like a cheerleader. The NYT often selects reviewers with relevant domain knowledge, or who are themselves authors in a related genre. So opinions will be expressed, hobbyhorses ridden, and nits picked, even in glowing reviews.

You should be careful to avoid setting up a business rule that could overemphasize a few negative raisins in a positive pudding.

The idea of a binary positive vs negative is really troubling to me as far as an algorithm goes. It should very much be on that sliding scale that Metacritic or Rotten Tomatoes uses--sentiment analysis shouldn't return just a 1 or a 0.

Reply on news.ycombinator.com