Original URL: http://www.theregister.co.uk/2009/01/27/user_generated_reviews/

User-generated reviews - blessing or bull?

Crowd wisdom v mob stupidity

By Rik Myslewski

Posted in Applications, 27th January 2009 05:00 GMT

Analysis Last week, connectivity-hardware maker Belkin admitted that one of its employees had been using Amazon's Mechanical Turk hiring service to pay for positive - and false - reviews of Belkin products.

While we congratulate Belkin for quickly admitting its employee's unethical behavior, and while we can only assume that said employee - identified by Engadget as bizdev rep Michael Bayard - has either joined the growing ranks of the unemployed or is now on a very short leash, the entire affair raises a larger question: How reliable are user-generated reviews, anyway?

More specifically, how can an online reviews service ensure that the product evaluations its users contribute aren't tainted by self-serving bias?

After speaking with spokesfolks from a selection of online reviews sites, we've concluded that there's no way user-generated reviews can be guaranteed 100 per cent accurate.

But we've also concluded that this is OK. Sorta.

Wisdom of crowds or stupidity of mobs?

User-generated content (UGC) is the bedrock of Web 2.0. Videos on YouTube, photos on Flickr, recipes on Recipezaar ("Choose from 341,000 member-contributed recipes") - the intertubes are stuffed with the creativity, considered commentary, and crackpot craziness of literally millions of UGCers.

Such participation is all well and good. But when you want an authoritative product recommendation before making a purchase, which would you prefer as your guide - expert analysis from an experienced reviewer or a consensus opinion derived from "The Wisdom of Crowds"?

Or both?

Some product evaluation websites, such as Yelp and Epinions, rely exclusively on UGC reviews. Others, most prominently Amazon, mix UCG with "Editorial Reviews". Still others - Digital Photography Review, Macworld, Consumer Reports, and many, many more - select and vet experts before publishing their often comprehensive analyses.

One clear advantage to the site choosing to host UGC reviews is cost. The Us Ging the C offer their opinions for free, while expert reviewers are paid for their time and trouble.

But there are three other less mercenary advantages to UGC reviews: breadth of coverage, the averaging of bias, and community-based expertise.

Take Yelp, for example, which publishes UGC reviews of over 20 categories of businesses and services in over 160 US cities. A single city can have thousands of reviews - San Francisco, for example, currently has 26,409 UGC reviews of everything from fondue restaurants to dog walkers. No expert review site could come close to that reach.

Such an enormous collection of reviews also provides a large UGC site with some protection against spurious reviews by their sheer volume. On Amazon, for example, JK Rowling's Harry Potter And The Goblet Of Fire has 5,180 UGC reviews. Even if you agree with the 75 reviewers who gave that book one star, it's still instructive to know that 4,402 others gave it five stars.

Then there's the intimate, streetwise smarts that come from living in a community whose offerings are being reviewed, such as those 26K-plus San Francisco reviewers. According to Stephanie Ichinose of Yelp, "Our posters take a certain amount of pride in the reviews they post, and in their community - New York, Chicago, San Francisco, and the rest."

But there's an even more important aspect of community: the desire for a UGC community to police itself to keep bad apples out of its beloved barrel. As Ichinose says, "The community self-polices themselves. If someone spots a review they think is inappropriate, they submit a note to our customer-service team."

Patty Smith of Amazon agrees, saying "Over time, the marketplace will out false reviews as 'unhelpful'," referring to how Amazon allows UGC reviewers to review not only products, but also each other.

The value of expertise

But even a thousand reviews of a complex piece of kit submitted by inexperienced UGC noobs can't match a comprehensive review by a single expert or team of experts who thoroughly test a product and compare its capabilities to similar products they've also used.

Take, for example, the review of the Nikon D90 on Digital Photography Review. This 40,000-word, 37-page review is profusely illustrated and includes a broad range of test photographs and comparisons with competitive cameras. No UGC review could ever come close.

But buying a thousand-dollar camera is clearly a bigger commitment than deciding where to go for fondue.

To advise on such major purchases, some expert-review sites also maintain their own product-testing labs - think Consumer Reports, for example, or computer-review sites such as Macworld or PC World.

Labs enable review teams to not only run a product through a suite of standardized tests, but also compare its performance with their database of results from testing comparable products. UGC reviewers use more of a "trust me on this" model.

Smith told us that Amazon provides both UGC and expert reviews "because there are certain people who are experts... Sometimes you want people who have more experience."

Picking the right reviewer is key. As Jason Snell of Macworld told us, "We try to find reviewers who are experts in that product category, and ideally with that product. We also try to use people who have experience reviewing products, so they know what to look for when evaluating any product.

"It's easy to slag a product because some feature of it doesn't mesh with some personal quirk of yours; it's also easy to praise a product uncritically because it's new and cool and the PR person who told you about it is nice.

"Neither of those, however, will give users a real idea about a product's strengths and weaknesses, and that's the hallmark of a good reviewer."

In other words, an editor - the chooser of the reviewer - is as important as the reviewer.

Editor-less editorial

This lack of up-front human interaction with a reviewer is where UGC reviews fall short.

Both Ichinose and Smith admitted to The Reg that neither of their sites employ humans to vet reviews for accuracy before they're posted. Amazon and Yelp do, however, both have software that reviews the reviews before posting.

As Ichinose put it: "The opinions posted to the site go up immediately with no screening, but we have a software program that runs in the background that identifies patterns of abuse."

Neither Smith nor Ichinose would give any details about the way their sites' software flags false reviews - a reasonable precaution, since publishing those details would make it easier for miscreants to circumvent them.

Both sites also have humans who investigate possibly shady reviews, but only after receiving a complaint from a user. As Smith says, "When we hear complaints about a review we take a look and remove it if it's outside of our guidelines".

Even with these precautions, according to Smith: "From time to time people may try to game the system, but over time, it's a sound system. We believe that customers use it as it's intended."

After all, she argues, "It does us no good if we sell a product that's faulty, because that product will [be returned to Amazon]. It works to our benefit if [customers] have a good experience - they'll stay a loyal customer."

Perhaps, but when asked if there were any way to prevent a determined sleazeball from gaming their systems, both Smith and Ichinose allowed that, yes, it could be done. Still, both were confident that their user communities would discover and unmask such behavior. Eventually.

Despite this vulnerability, even expert-review purveyor Snell, when asked why his site doesn't host UGC reviews, said: "The short answer is, because we haven't gotten around to it."

Snell has no philosophical aversion to UGC reviews. His reasoning is similar to that of Smith and Ichinose. "For user-review systems to work, you need to find ways for the community to police itself and to lift up the very best of the reviewers, so that the community can help stop those who will almost certainly try to game the system."

The fact that Snell's lab-based, expert-driven reviews site is contemplating UGC reviews proves that such reviews have gained credibility since they first appeared well over a decade ago.

As Amazon's Smith said, "We've had customer reviews since we launched the site in 1995... people thought we were nuts." ®