Lisa Whelan has quit Yelping.

Well, it’s not exactly breaking news, but she’s got some great insights on the problems that exist with review sites. Lisa left a glowing review for her dentist. It was one of only two reviews for them. Then one day, she got a call from her dentist asking her why she removed her review and if they’d done something wrong. No, they hadn’t. She went on a quest to find out why her review had been removed. After a few back and forths with Yelp, they still couldn’t tell her, nor could they put it back. Apparently, her review had been spotted by some spam filter for some undisclosed violation and they just took it out.

And therein likes the problem. How do review sites serve up trusted relevant reviews without allowing people to game the system? There was a recent hubub Brideshead Revisited the movie (also at Yelp) when they booted an undisclosed number of users for being in an alleged gaming ring providing glowing reviews to each other within a womens’ business network. It led to a class action lawsuit.

So how do you get around the problem of insincere reviews? I see three potential solutions. The first is increasing relevance by providing binary recommendations from people you already know and trust. It’s not very “real world” to give something a one to five star review. Think about the last time you called a friend for a recommendation on where to go to eat and they told you to go to Golden Bowl Chinese restaurant and gave it a three star? Never. If you give a restaurant a three star review, would you tell your friends to go there or not? Recommendations from friends are usually binary. Either go here, or don’t go here. Review sites need to simplify this process.

The second involves finding people you have something in common with in order to be able to evaluate their recommendations. Most review sites lump everyone together into a big pool of reviews. They try to back into relevancy based on the reviews you make. In effect, their hypothesis is that people who like the same things may be similar to each other and like their other reviews. I think they’ve got it backwards. My hypothesis is that people will like the same things if they already know that they have something in common.

Imagine if you could join a group of people based on affinity. Maybe you’re all parents who need kid friendly restaurants. Maybe you’re all executives who work in Palo Alto who take power lunches. Maybe you’re all neighbors who frequent the same local places. In any case, if you can find people who you have some key thing in common with and then get opinions from them, your relevancy will increase. And because you’ve filtered these things based on whatever that affinity is before anyone does any reviewing, you’ll have a greater degree of trust.

The third potential solution is context. People typically don’t pick a service provider or store or restaurant in a vaccuum. They have a purpose such as “a nice place to take out of town guests,” or a “fence contractor who specializes in wrought iron,” or “a vet who is good with little dogs.” Users need to be able to specify a context for their request in order to get the best recommendations back. By providing the context of their query, they’ll be able to get back more specific, and therefore more relevant, results. The best vet in town might only work on Great Danes so he might not be the best choice for your chihuahua.

There will always be people who try to game the system. By addressing a few fundamental things about how recommendations work however, we can provide much more relevant results and reduce the impact of gamed reviews. Eventually, someone will get this right. And when they do, I think it will create huge opportunities for businesses to connect with their best customers in a much more meaningful way.