Online reviews are now an important aspect to consider for many businesses. It is becoming increasingly common for customers to check online reviews for a product or business before buying, particularly in sectors such as retail and hospitality. This trend has only increased during the pandemic as more customers buy online.
As we now depend on online reviews for our purchasing decisions, it is important to consider who is making and policing these reviews. Fakes reviews, to either drive up or destroy a business's reputation, are growing. Silencing genuine reviews is also a major concern as businesses seek to apply pressure on websites hosting these reviews.
Reviews can make or break a business, so what can we do?
A Legitimate Case For Removing Reviews?
A transparency report by the Danish company Trustpilot showed that the business removed over 2.2 million fake reviews on its website in 2020. Over 38 million reviews were submitted on the consumer review website in the past year in total. Almost three-quarters of those removed fake reviews were automatically removed by Trustpilot’s bespoke software, with another 600,000 fake reviews being removed by Trustpilot’s Content Integrity Team.
For a website that is designed to show accurate reviews of other businesses, it is clearly important for them to remove fake reviews. But sometimes reviews are removed, even if they are legitimate.
23% of reviews flagged by businesses themselves were removed for being harmful or illegal. A further 12.5% were removed as they were supposedly advertising or promotional. Personal information in the reviews was the cause for 4% of removals.
Many of these removals would be justified, but who decides this? Trustpilot themselves. If a review for a takeaway said, “I found a rat in my food, 1 star, I will be going to the takeaway across the road next time”, would that be a legitimate review, or be classed as advertising? This is an important question considering 82% of reviews flagged by businesses were one-star reviews. To what extent should a business be able to remove poor reviews?
Protecting Businesses When They Do Wrong
At the end of January, there was a huge wave in the amateur investment world. The now infamous GameStop stock was rising in price rapidly, going from below $40 per share to over $400 per share. It was largely amateur investors profiting off this increase thanks to trading apps such as Robinhood.
Robinhood eventually decided to limit purchases of the stock, while continuing to allow traders to sell. This move was widely criticised and is believed to be one of the contributing factors to why the share price started to fall again. There was so much backlash that the CEO of Robinhood was called to testify before the US Congress as part of the GameStop investigation.
Robinhood also received backlash on Google Play, where thousands of angry users decided to leave one-star reviews. The bombardment of bad reviews saw the app go from four stars to just above one star in a matter of hours.
Google eventually stepped in and removed over 100,000 negative reviews of the app to bring its rating back up to four stars. At the time Google claimed it was doing this to remove fake reviews, but most of the anger at Robinhood was justified.
The negative reviews continued to come in, and Robinhood plunged back down to nearly one star. This time Google decided to keep the negative reviews there, recognising that most of the reviews were genuine.
The big question for Google is how can they determine whether a review is legitimate? In the first instance they removed the bad reviews, but when the reviews came back, they decided to keep them. It is difficult to balance these things when you need to protect businesses but also protect customers.
How Do Businesses Moderate Reviews?
Different online businesses moderate reviews in different ways, but the use of machine learning is becoming more common for detecting fake or harmful reviews.
One company that uses machine learning algorithms to detect fake reviews is Amazon. While the inner-workings of the large multinational company are not completely known, insights can be gained through the open-source datasets that they publish. In the past, Amazon has released review data so that researchers can find methods of detecting fake reviews. These datasets have become very common for use in learning machine learning methods too. So, what have we learned?
There are many factors that can be used to analyse whether a review is genuine or fake. Word and sentiment analysis can be used to determine whether a review is fake, as certain words are more likely to appear in fake reviews.
It is also important to look at reviews together as opposed to individually. Several reviews containing almost identical wording can indicate that an army of bots are trying to change the review score of a product – bots use similar code, and so write similar words. Multiple reviews appearing in a similar timeframe on a product that is rarely reviewed can also indicate fake reviews.
Artificial intelligence has improved at detecting fake reviews, but sometimes sweeps up a few genuine reviews in the process. What happens after machines remove reviews? Often, nothing. For many businesses, human moderators only look at a small percentage of content that is removed by computers. This means that we are placing a lot of trust in machine algorithms, and it is sometimes the case that we don’t know how these algorithms are even making decisions. This opens the door to potential biases and pitfalls.
So Where Do We Stand?
Online reviews are clearly important to us. They allow customers to see whether a product is worth buying and can be a great way for businesses to prove that they have value. Many customers prefer a world with online reviews.
However, there are two main issues. Firstly, there are many fake reviews out there – often posted by automated scripts and bots. It can be damaging to consumer confidence if they see many fake reviews and could have negative consequences for businesses.
The second problem is caused by a reaction to the first problem. Consumer review websites need to remove fake reviews, but the quantity is so large that they also rely on automated scripts for this process. This can lead to real reviews being taken out in the process. The alternative is for human moderators to check each review. This would require a huge, costly workforce – and would then be susceptible to human bias, such as protecting businesses that we see as valuable.
There is no easy answer to the problems caused by online reviews and is something that we may have to live with as we transfer to a digital world.
Comentários