from Hacker News

Ask HN: Would you use a review sentiment search engine?

by dopplesoldner on 4/7/14, 7:49 PM with 6 comments

Hi guys, wanted to get some opinion about an idea I've had. Constructive and honest feedback will be really appreciated.

A sentiment search engine. Suppose you type the name of a restaurant/camera/hotels/protein supplement etc. - you will get a condensed graphical view of with "sentiment scores" in different categories based on what real customers have been saying around the internet. E.g for an iphone, entities might be battery, display etc.

Data is scraped from individual reviews, shopping sites etc. Additionally you get information about the original source of information so you can carry further research if you want to.

So in short, it can be the default go to place for any kind of reviews.

  • by neuralk on 4/8/14, 2:30 AM

    This is a great idea. I would use it. In fact, I want to build it! I've been reading about sentiment analysis as part of my research, and I had this idea literally the other day: I was looking for laptop batteries and I did not have the patience to search through some N number of Amazon/eBay/other storefronts. The second issue was the fact reviews range from 1 star to 5 star on the same product BUT a crucial fact was that more recent reviews are more relevant (for instance, if everyone from the past 3 months is rating it 1 star, vs. only 5 star where all previous bad reviews were from 3+ years ago).

    So how would you define sentiment? A single score on a range from "good" to "bad"? Multidimensional, like a range of emotions? A major problem I've found in sentiment analysis literature that some assumptions/models yield limited/poor results even if algorithms are good.

    Have you started working on this? I'd be happy to discuss it or your ideas some more.

  • by ohashi on 4/8/14, 4:49 AM

    This sounds very similar to what I do for web hosting reviews. For instance, take a look at Digital Ocean's reviews: http://reviewsignal.com/webhosting/company/101/digitalocean/

    It calculates an overall (based on all pos/neg things said) but breaks down into components relevant to web hosting: support, uptime, price. You can also figure out who competitors are and see trends.

    If you can't tell, I like the idea. I also think it's an incredibly hard problem to solve. I've done it for a small niche. Scaling it to a general solution is something I haven't been able to figure out (yet).

  • by nateday on 4/7/14, 8:00 PM

    I agree with @manmeet that the algorithm for the sentiment analysis has to be great; the search algorithm also has to be tremendous to convince me to switch from Google or DuckDuckGo. To avoid the challenge of making a new search engine, maybe you could build a plugin?

    Somewhat similar sentiment analysis, for terms rather than unique entities: http://www.whatdoestheinternetthink.net

  • by manmeet on 4/7/14, 7:51 PM

    I think its a pretty good idea. All depends on how well u design your algorithm. I had a similar idea but wanted to use twitter to get sentiment regarding anything, from people to movies to devices or services.