Researchers at Cornell University have developed a software program that can detect false reviews, with nearly 90 percent accuracy. In many ways, the internet has made shopping easier than ever, by opening people up to hundreds of online stores that carry a variety of products that might not be available in their local stores.
With the internet, consumers don't have to rely on their local market or catalog delivery, they can log on to the internet and access a whole world of products and services.
All of this choice can be overwhelming, however, and many people get stuck in trying to decide where best to invest their money. Many websites include a review feature, where consumers can read other people's experiences with products they are considering buying.
However, some of these reviews are spam, created by people who will profit from you choosing one product over the other. Unfortunately, humans are not that good at telling the difference between a spam and a genuine review.
Thanks to computer science, we can now rely on software to filter out spam reviews. The software uses statistical machine learning algorithms, which detect indicators of spam, such as a high volume of verbs, which usually means spam, as opposed to a high number of nouns, which tend to indicate a genuine product review. The software also detects other differences between fake and real reviews, such as use of keywords, punctuation, and how much the reviewer references him or herself.
Humans tend to fall prey to what Cornell doctoral candidate, Myle Ott, calls a "truth bias." This means that most people tend to assume everything they read is true, until evidence to the contrary is presented. Once people find out what they thought was true, is actually false, they tend to think in the opposite extreme.
They then assume that much of what they read is not true. Neither of these instances is ideal, of course. Ideally, one would be able to decipher what is true from false, but with the endless amount of information on the internet, it is nearly impossible to verify everything one reads. This new software can help people trust what they read.
Reviews are an incredibly valuable resource, but their value is limited to their accuracy. The internet provides a great forum for people all over the world to share experiences with different products and services. This information helps consumers make decisions about how to best invest their money. Because anyone can write these reviews, however, people can write false reviews, over-praising certain products and bashing rival products.
Certain spam reviews are quite easy to spot, but studies show that false reviews can easily slip under our radar. Cornell tested three human judges, to see how accurately they could spot false reviews. They were presented with 400 fraudulent positive reviews and the same number of genuinely positive reviews of 20 Chicago hotels.
The judges scored quite poorly, no better than if they had been guessing which reviews were false. "Ultimately, cutting down on deception helps everyone," said Myle Ott. "Customers need to be able to trust the reviews that read, and sellers need feedback on how best to improve their services."