There are two kinds of people when it comes to online shopping–those that read customer reviews in depth and those that just look at the star rating. Unfortunately, I tend to fall into both categories, depending on what I’m buying. However, most people would fall under one of them. But no matter which category you fall into, there is one common information we’re all looking for: is it worth buying whatever it is that we’re trying to buy? Of course, we all have our own metrics of gauging a product’s worth–independent of what others have to say–based on our own use case. Even if that’s true, there are certain features and traits of any given product that hold true universally.
It always seemed to me that those universal truths contributed to the bulk of our purchasing decision rather than the marginal use cases of our own. But how do you know what those universally true things are? Do you go by the manufacturer’s description of the item? The TV commercials and banner ads? Certainly not. Then do you go by customer reviews? Not quite. Customer reviews have a lot of noise and the star rating you see is never a true measure of its utility and worth. Then how do you separate the noise and extract the features that are universally true? This was the question that I wanted to answer with Fooreviews.
Fooreviews is not a product review site. It’s not a meta review site like ReviewMeta either. Instead, it’s an attempt to draw a consensus from thousands of users on the features that characterize a product. These are what I would call universally true facts of a product.
One of the main challenges of Fooreviews is lack of data. The category of products where we can extract useful information reliably, after accounting for noise, is few. Without enough data, it’s very easy to misinterpret and misrepresent the findings on a product. In addition, labeling the data for machine learning, one of the main techniques I use, is a time-consuming process.
The initial idea behind Fooreviews was to automate the process all the way from product identification to feature extraction and data visualization. But achieving the complete process requires a lot of resources and dedication. Unfortunately, I had to abandon Fooreviews after the initial prototype. It was right around the same time I started grad school so the incentives to continue working on it went down even further.
Recently I came across one site that does something between Fooreviews and ReviewMeta. The Reddit commenters seemed to love it. And I agree with them. It’s a feature that’s sorely lacking in our online shopping experience. In fact, this is so painfully obvious that even Amazon has launched a keyword explorer for customer reviews. Even though Amazon would never do a full implementation of what Fooreviews or these other sites set out to do, there is an undeniable need to reform how we consume customer reviews.