July 1, 2008 7:15 AM   Subscribe

Pluribo is a way-cool Firefox extension that automagically summarises Amazon product reviews.
posted by matthewr (24 comments total) 4 users marked this as a favorite
The demo video on the home page doesn't work for me in either IE or Firefox. Doesn't give an error message, just blank while playing.
posted by Perplexity at 7:33 AM on July 1, 2008

Maybe a problem with your Quicktime install? Works fine for me.
posted by matthewr at 7:41 AM on July 1, 2008

Hm I thought this sounded quite exciting but so far I got:

Pluribo doesn't currently cover this category. More coming soon.
on a Wii game and an Anne Rice book


Pluribo doesn't currently cover this item. More coming soon.
on a Cyndi Lauper album and a DVB-T USB stick
posted by ClarissaWAM at 7:42 AM on July 1, 2008

It only works for some technology items now. I don't have the exhaustive list, but it seems to work for at least cameras, MP3 players, and TVs.

Full disclosure: tester and friend of developer.
posted by grouse at 7:48 AM on July 1, 2008

What happens when an impotent force, natural language processing, meets an illegible object, Amazon reviews?

posted by DU at 7:48 AM on July 1, 2008 [5 favorites]

Repost this when its actually useful. I looked at a bunch of items, and it couldn't help. Posting this plug in now is like posting a link to a blog that has no entries.
posted by Dave Faris at 7:51 AM on July 1, 2008 [1 favorite]

Despite my joke, this does look useful. But as others have said, it covers basically nothing. Also, the end result seems like it could be more informative. How does *it* know that something is "worth the extra cost" for me? Maybe I'm looking specifically for an unscratchable MP3 player, so a "tendency to scratch" makes it worthless.

Perhaps a pros/cons list, maybe weighted by how many reviewed listed each, culled from the comments would be better.
posted by DU at 7:57 AM on July 1, 2008

Oh, thank goodness, another source of unreliable information about expensive consumer products. And just in time, too.
posted by box at 8:32 AM on July 1, 2008

Interesting idea and it looks okay until you dig around ... here's a couple of gripes, from the first two items I looked at, if the developers are reading.

Canon Digital Rebel XTi (one of their test links). It displays examples of the positive reviews, but no examples of the negative reviews for overall reliability, which is quite low according to their numerical scores (and which would make me think twice).

Panasonic DMX-LX2. Reviews badly for weight, but 2 of the sample reviews say it is heavy, while the other 2 say it is lightweight.
posted by carter at 8:32 AM on July 1, 2008

This is a nice idea, although I am not sure its screen real-estate is worth the limited functionality, so I will probably uninstall it now. The puff piece before one downloads could explain that the key quality indicators are taken from their frequency of mention in reviews, to preclude unmet expectations like those expressed by DU above. (I am assuming this is how it works, from the operation and results).
I was disappointed that the app. won't summarize products that do not have a statistically significant number of reviews. It could still give a summary -- just provide a health warning that the number of reviews is too small to provide a meaningful assessment of the product. As it stands, this rules out about two-thirds of even those product categories that it does handle. So probably not worth keeping, just yet ... :-(
posted by Susurration at 8:35 AM on July 1, 2008

One aspect of Amazon reviews I often find helpful is references to alternate product / brand choices which this does not seem to help with.
posted by adamms222 at 9:42 AM on July 1, 2008

Hi, I'm one of the co-founders of Pluribo. I saw all of the really great discussion and comments on MetaFilter so I thought I'd join this community at long last. I'd been wanting to for quite a while, but I had been too cheap to cough up the $5... :-)

First, a small apology that our extension only displays summaries for major electronics items, and not every single product category on Amazon. We tried to make this explicit on the homepage prior to download, but it might have been easy to glaze over this important fact.

We actually have the ability to summarize much more than just electronics, but we're not releasing those right now. We're anal about quality control, so we're deliberately taking our time to tune the system for new categories. (If you were unimpressed by our lack of coverage right now, imagine how you'd feel if the summaries hadn't been accurate!)

The good news is that as we add new categories, you won't need to update the extension. You should get fed the new summaries automatically.

All of this leads me to pose two questions for the community:

1. Which categories would you want to see next? Appliances? Movies? Books? Etc?

2. In cases where we aren't calculating summaries, do you like the current tiny status message, or would you prefer the extension just to be silent altogether?

I'll respond to your other more specific feedback in separate comments.
posted by samidh at 9:46 AM on July 1, 2008

In reply to carter:

We actually have a much longer list of extracted quotations than can fit into the small amount of screen real estate we want to occupy. Perhaps we should add the ability to "page" through them? This might require the hovers to be "sticky" though. What is preferable?

Also, it's good that you noticed that sometimes the tone of the extracted quotations don't always match the quantitative dial. In cases where this happens, this is most often the correct behavior. The reason this happens is that we calculate all of the ratings *relative* to other products in that category. We have an FAQ exactly on this topic (it's the 7th one down, starting with "The summary text for a certain product sounds negative...").
posted by samidh at 10:14 AM on July 1, 2008

In reply to Susurration:

That's an interesting suggestion. I guess at the moment, we're just sticklers for accuracy, so we're shy about displaying a summary when the number of reviews is not statistically significant (i.e., the margin of error is relatively high). But in your view, you'd still prefer a potentially inaccurate summary over none at all?
posted by samidh at 10:19 AM on July 1, 2008

samidh, two ideas based off of how I found my landline phone.

First, in certain cases (such as phones) it was far better to search for best selling than highest average review. You might consider tossing that into the algorithm. Even Amazon suffers from a lack of reviews in certain areas and this was a good way to get beyond that problem.

Secondly, I found a base product that was selling very well and a sub version of it that had the features I wanted but only one review and it was sold only via Target (and the image was wrong). In other words, it would have been in the "gamble" or even "N/A" category if I hadn't known it was based on a rock solid product. It's been a few years and it's been great. I use this process all the time now. There might be a marker in the pages (maybe the "You might also like" section?) that you can mine for these scenarios.
posted by jwells at 10:40 AM on July 1, 2008

Thanks for the explanation Samidh. Sure, more/longer user comments could be useful - although you'd have to do some user/usability testing to figure that one out.

Re. 'correct behaviour' - yes it may be 'correct' from the point of view of the AI, but it's not necessarily so from the point of view of the users (who are probably a pretty varied bunch). I suspect that if you need a separate FAQ page (not accessible from the toolbar presumably?) to explain to users why the results may not necessarily match what they were expecting, then you probably need to do some more work ;)
posted by carter at 12:15 PM on July 1, 2008

Posting this plug in now is like posting a link to a blog that has no entries.

posted by davejay at 1:19 PM on July 1, 2008

In reply to samidh:
Although I can see where you are coming from, with the "statistically significant" filter, as someone who engages in a LOT of research for various purposes, I am equally skeptical about the significance of large numbers of subjective reviews ... Having read the market research on skew-rates for satisfied vs. unsatisfied responses, my take is that this data is probably biased whatever the number of reviews ... :-)
So yes, I would find it helpful to have a smaller number of reviews summarized (with a suitably phrased health warning about not having sufficient reviews to be confident that the views are not biased). For example, when Amazon have 36 reviews on a product, you do not consider that number significant (which seems to filter out the majority of items). But that is 4 additional webpages to be scoured. You have sufficient confidence to produce a summary of positive/negative sentiments at this number of reviews, but not the feature summary. Even if the latter is not statistically significant, it saves effort in navigating multiple webpages to get a feeling for the outcome -- and a summary would also help to remove my own sensitization bias, in remembering what people said from page to page of the reviews! Less than 10 reviews fit on just one additional webpage to be loaded -- not a big deal. More than 10 reviews means additional pages to be loaded, by which time I have forgotten what people said on page one of the reviews.
Psychology/HCI research shows that most people have a very poor ability to form a "big picture" from more than 4-5 complex informational items. So you might consider making that your cutoff point for the summary and provide a "significance" warning?
posted by Susurration at 8:52 AM on July 2, 2008

Anyone else notice in the demo video when he hovers over "tendency to scratch" one of the verbatim excerpts used to support this criticism is "impossible to scratch?" Hmm, that might cause some problems right there.
posted by arcticwoman at 9:30 AM on July 2, 2008

I didn't install the firefox extension, as don't do extensions, but I'm thinking there might be a good opportunity for Pluribo as a website.

The benefit being
1) only show products which have been rated - so users don't get the 'we don't have data on this yet'
2) no installs, so everybody can try it
3) revenue stream if a user buys from amazon via a/the pluribo site.

Though you may need another draw to get people to Pluribo site.
posted by pedalpete at 9:34 AM on July 2, 2008

Pluribo scans all of the reviews on any Amazon webpage...

Maybe it's just me, but it appears to work only with pages.
posted by mleonard at 11:26 AM on July 2, 2008

In reply to arcticwoman:

Yup, that's to be expected. Believe it or not, people are rarely unanimous in their opinions on just about anything ;-). That is why on the graphical dial, it is not at the extreme end of the scale. We also have an entry about this phenomenon on our FAQ (see "Why is it that sometimes the extracted quotes contradict each other?")
posted by samidh at 1:17 PM on July 2, 2008

In reply to mleonard:

Ahh good catch. We just support the USA localization of Amazon right now. It should be pretty easy for us to add other Amazon localizations, as long as they are English language (like We'll work on this. (Personal apologies since I went to grad school in the UK!)
posted by samidh at 1:21 PM on July 2, 2008

There is definitely potential in a service that can deliver me a feed for genuine customer reviews. I've made a few electronics purchases recently and I found that google wasn't particularly useful for gathering data of this sort.

Would it be possible for Pluribo to collate data from all Amazon customer reviews written in English? Maybe start by combining the UK and US pages. (Assuming that a given item is sold in both regions.)

But why stop there? Why limit yourself to Amazon feedback? For example, I've found newegg to be a very useful resource.

It seems to me that the real problem to be solved is this: how to incentivise end-users to make the effort to leave considered, useful (and literate) feedback for the good of other consumers. And preferably in one place.
posted by mleonard at 4:47 PM on July 2, 2008

« Older Hello friends and lovers -- Food Party   |   Is it jazz? Listen, bud... Newer »

This thread has been archived and is closed to new comments