Shopping Online - Are Reviews Trustworthy?

Chris B. -  What I'm working on at the moment is how we can apply the concept of trust in modern e-commerce settings.  So a lot of systems, a lot of interactions, these days take place online.  More people are buying things online, people are booking hotels online, they're finding partners online, and the concept of trust becomes more important because we might never meet these people we're buying things from.  How do we know anything about them?  How do we know if we can trust them?

Chris -  Well obviously, a lot of people do because they're spending a lot of money and a lot of time on the internet.

Chris B. -  That's true.  So really, because more people are spending time on the internet, there is a large resource of people who already have experiences.  They're writing reviews, they're giving recommendations, and one of the big questions in my field that I'm interested in is how can we tap into those large quantities of reviews or ratings, and help people discover services that are right for them?Chris -  And that are reliable.

Chris B. -  That's right, yes.

Chris -  So how are you doing that?

Chris B. -  At the moment, we're using statistical methods, but really the problem is, in a large base of experiences, we need a lot of people to write reviews and give their opinions.  If you don't have that, it becomes very difficult.  In new systems, if a new website starts up and lots of people begin to use it, at the beginning there might not be a lot of ratings or previous opinions.  In those kind of experiences we're really looking at semantic web technologies, letting people describe their experiences in a very rich way and write reviews that a computer can understand.

Chris -  I see, so rather than writing a review that means something to me but which has no parameters a computer can extract any useful information from, you're saying you guide people in writing some kind of review or feedback on a resource that a computer can extract value from and then return value to other people when they're asking, "Well what's the use of this?" or whatever...

Chris B. -  Yes, exactly.  That's one part of it, helping people to do it, but the other thing is we need to develop languages that computers understand for people to describe their experiences.  We need to, as you said, give people an easy way to describe them in a computer readable format.  Then there's the other layer of technologies which we need to help people manage these large numbers of experiences, because we don't want to have to read thousands of reviews to understand how good a product is.  I want to see a rating saying, "This is good for you because I, the system, know something about you, the user, and I can say that this is the right product for you."

Chris -  And how are you doing that?

Chris B. -  That's an excellent question.  One of the techniques is semantic matchmaking.  So we know something about what makes a good product in your opinion.  As a system, I know what constitutes a good product for you.  I can then look at the large number of ratings that already exist and try to maybe filter out ones that wouldn't apply to you, maybe if people have different criteria than you, I can say those reviews are not appropriate for you and so therefore, that's one technique: to limit the number of reviews that people are exposed to.

Chris -  So it comes down really to writing some clever computer software that can extract the right information from what's already there.

Chris B. -  Yes, in one way, but there are other sources of information.  Now we have social networks, people have connections, people are using Facebook to say that they like this product and their friends might hold their opinions in higher regard, so another question is how can we use social information to add weight to certain reviews and opinions, and reduce the weight to others.

Chris -  Because what works for me might not be appropriate for you.

Chris B. -  That's exactly right.  That's the filtering question.  There is still the problem that you mention of how do we actually compute, or how we'd arrive a decision and say, "This is the one you should go for.  This is the one you can trust. This person is deceptive.  This person is lying."  Those are questions that I don't think we have an answer to yet, but really need to solve them because more things are happening online and I think it's only going to become more important in the future.