December 17, 2007

Summize: A Sentiment Engine For The “Reviewosphere”

Erick Schonfeld

24 comments »

summize-logo.pngSometimes reading online product reviews can be a real time sink. There are so many opinions out there—from CNet to Amazon to blogs. It can take a long time and a lot of sifting to figure out what people really think. But now there is another way. If you want a snap shot of popular sentiment about a book, movie, CD, or gadget, type its name into Summize.

summize-heatbar.png

Started by two AOL expats a year ago, and launched in a public beta last month, Summize does sentiment analysis across dozens of review sites and more than half-a-million blogs—what CEO Ajaipal Virdy calls the “reviewosphere.” These include review-heavy sites with ratings, reviews, and other structutred data like Amazon, CNet, IMDB, and Epinions, as well as active blogs filled with unstructured, unfiltered opinion. In that sense, it is a cross between CNet’s Metacritic and Blogcritics. Explains Virdy:


We algorithmically analyze the review text to learn the polarizing words that correlate with user-assigned star ratings. We use our knowledge of polarizing words to extract the opinions from unstructured blog posts, and algorithmically assign them ratings.

summize-radiohead1.png

Summize harvests all of this opinion and boils it down to a single heat map that it calls a “snip.” This is a horizontal bar with colors ranging from red (wretched) to yellow (so-so) to green (great). It attempts to capture the collective sentiment about any given product at a glance. For instance, out of 21,021 opinions about the band Radiohead, 67 percent are “great,”, and only 10 percent are “bad” or “wretched.” The band’s most reviewed album is Kid A, the most liked is OK Computer, and the most buzzed about is their recent online pay-what-you-want freebie In Rainbows. There are links and excerpts from recent blog posts about Radiohead. And, in a nice twist on Amazon-like collaborative filtering, every product has a list of related products. In this case, “bloggers who discuss Radiohead also discuss” Band of Horse, The Pixies, and Idelwild.

The site feels a little thin for some products, especially newer ones. But the approach is promising. Summize plans on adding other categories such as travel and restaurants in the future.

  • Sphere It

Trackbacks/Pings (Trackback URL)

  1. Summize: Web 2.0 Rankings Aggregator at blog.rotracker.net
  2. VelociPeek - Eric’s weblog on tech » Summize For iPhone

Comments

RSS feed for comments on this post.

  1. Paul

    Pretty good…

  2. Angela Hayden

    I like the site. It’s fun and easy to use. I luv a quickie’ opinion.

    sincerely,
    angela hayden
    art goddess

  3. Scott B

    There’s more than one horse in that stable, partner. Their name is “Band of Horses.” ;-)

  4. Adrian Keys

    I guess it’s just another way of representing vertical search results..

  5. Jeremy Olson

    I am impressed. For the ten plus minutes I’ve been perusing the site there has been lots of interesting content and a really nice interface. I really like the non-5 star interface. I think it presents the information in a very useful way. I also like the way the reviews are displayed (with green to red background based on their rating rather than displaying their star rating).

  6. Wes

    I did a few searches of products I know and I was quite impressed. Not sure whether if I will still be impressed if I were doing some serious searches.

  7. AnonTroll

    The results page is way too busy, talk about information overload. People with short attention span, meaning most of internet users, will take one look and leave. I did.

  8. Jones

    This is great! I am being driven crazy reading camera reviews lately. Finding them is bad enough. Then you have to parse them to figure out what is really goo/bad an dhow important these things are to the whole package. THis site looks like it could help me a bit.

  9. Rishi

    I was very excited until I realized that they seem to only be using Amazon reviews. Understandable since that’s how they are able to group opinions by product. Otherwise if you were simply crawling the web the problem is orders of magnitude more difficult since you’d have to not only identify what is review content but also figure out exactly which product the review is for.

    Amazon already does an OK job with helping a user filter thru reviews.

  10. CAR

    I agree with #9, i don’t like amazon services like alexa.

  11. rod / techwatching.com

    That’s some impressive parsing technology, but more importantly, it looks like someone there has come up with an intelligible and compact way to summarize thousands of bits of info - a very impressive feat of conceptual design - well done!

    http://techwatching.com/tag/summize

  12. Tampa WebSite Design

    Well, I think Amazon reviews simplified there work alot.

  13. livejamie

    i prefer metacritic.com

  14. Tyler Wright

    great logo and colors

  15. Stef

    This website will get sued in no matter of time by the brands.

    Other than that, there is no chance of it getting any success if it does not open up the database. Why? Because it’s too easy to put bias.

    Last but not least, sentiment analysis is very hard, except if you simply search for swear words.

  16. Parul Bindra

    This site certainly isn’t the first kid on the multi-site review block, but it’s looking to stand out with the sheer breadth of its coverage. Many of the reviews seem to come from database-type sites like Amazon, IMDB, and the like, but round up hundreds of blogs, dozens of newspapers and user comments on any topic, and you’ll never want for input.
    Summize does sentiment analysis across dozens of review sites and more than half-a-million blogs. Good Work !

    Parul
    http://www.bhopu.com

  17. 42mb.com

    I think they should add behavior analysis too with sentiment analysis.

  18. Shimi

    Omgili also has a very similar service for product reviews:

    http://reviews.omgili.com/

    It also has discussions from forums and overall buzz….

    Very interesting!

    P.S
    Many of the blog posts analysis for sentiment on summize are not accurate to say the least, how can I rely on them?

  19. Summize.com

    Sentiment analysis is indeed very difficult–especially for a consumer facing service like ours. Our job is to accurately detect sentiment even down to an individual blog post. However, just like with regular reviews, sentiment analysis is more reliable in the aggregate. After we analyze say 10-15 “reviews” in blog posts, a pattern starts to emerge that helps users determine at-a-glance the overall sentiment of a topic without having to read through a lot of unstructured text.

    To see examples of new types of derived experiences we can create using Sentiment Analysis, try “vs.” queries such as canon printers vs. epson printers

    http://www.summize.com/topic/c.....n/printers

    Also, take a peek at our trends page

    http://www.summize.com/trends

    to see blogger sentiment buildup on new products.

    Thank you all for the valuable feedback.

    Ajaipal “Jay” Virdy
    CEO
    Summize, Inc.

  20. Tom Peterson

    just search google for “PATENT RANKING RECOMMENDING PRODUCTS”
    you will see that it is pending PCT patent and they are in deep problem….

    this is realy funny becose i had the same idea and VC has searched for patent and this damn patent terminated everything…

    Tom

  21. chaching

    I think this type of tech is great… if it works. The ultimate use for this technology would be for stock picking =). But I’m sure summize already thought of that.

    Good Luck, and I’ll be sending you my resume soon.

  22. shelly

    the whole area of intermediaries on top of the mega sites reviews has gotten pretty solid momentum. http://www.reviewrave.com is another such example