As how helpful product reviews are perceived often depends on the gender of the reviewer, suggests new study
What makes an online customer product review helpful? Certainly numerous factors, but for some products at least, when an author’s username suggests they are a woman, the reviews are perceived as more helpful, according to a paper co-authored by Johannes Wachs.
That’s the case for products in the Movies or Beauty categories. Conversely, reviews of tech products are perceived as more helpful when written by users whose names suggest they are men.
In their study, the scientists analyzed more than 80 million English language customer reviews on the e-commerce platform Amazon across 15 different categories – including Books, Electronics, Clothing, and Movies – to see how users rate the helpfulness of the reviews depending on the gendered signals and behavior of their authors.
Overall, across all categories, the team didn’t find any general trend that signals of the gender identity or gendered behavior of the author – which could be inferred from the username or identified by a machine learning algorithm – had an effect on the success of a review.
Category-specific effects
On the other hand, the study, published in the journal Frontiers in Big Data, identified strong category-specific effects. Reviews by customers whose usernames implied they were women were perceived as more helpful in categories like Toys, Movies and Beauty, while reviews written by customers who indicated they were likely men received more positive scores in categories including Electronics, Kindle, and Computer.
The results also show that gender anonymous reviewers who, according to the machine learning model developed by the team, “performed” as women received better evaluations than users who indicated they were likely women in categories like Electronics or Cellphones. Similar effects were observed for anonymous authors “performing” as men compared with reviewers who implied they were men in categories including Books or Kindle.
On the Internet everybody knows you’re a dog
According to Johannes, the results suggest that in the online world people aren’t so anonymous anymore. He mentions one of the most famous cartoons from the early years of the Internet age that was published in The New Yorker magazine in 1993. The cartoon, by Peter Steiner, depicts two dogs: one, sitting at a computer, says to the other “On the Internet, nobody knows you’re a dog.”
“Almost 30 years later, the idea that you can be whoever you want to be online doesn’t seem to hold true anymore,” says Johannes. “Your words and emojis, and even your chosen username, seem to influence how others react to you – at least on product reviews on Amazon. Even the very weak gender signals we can send seem to make a difference on how people react to a seemingly simple statement as an Amazon review,” he points out.
Linguistic differences
In order to infer the gender of the reviewers, the team first used a name-to-gender prediction tool and was able to identify the gender of 42 percent of users in the sample, the ones who were signaling a likely gender with their username. Next, in order to understand and interpret reviews written by users who hid their gender, consciously or not, the researchers trained and tested a machine learning algorithm designed to identify the linguistic differences between reviews written by a man and a woman, taking into account factors such as style and word choice. The model was able to correctly classify the gender suggested by an author’s username with a success rate of at least 70 percent.
Commenting on the category-specific effects found in the study, Sandipar Sikdar, the main author of the paper, says that “one potential mechanism for this effect is that users judging reviews apply gender stereotypes – for example, that men are more knowledgeable about electronics while women are better informed about beauty products – to rate reviewers when they can infer gender from names. As these ratings influence the ranking and visibility of reviews to shoppers, this can amplify stereotypes.” Sandipar Sikdar is a PostDoc at RWTH Aachen University, in Germany.
The study The effects of gender signals and performance in online product reviews by Sandipan Sikdar, Rachneet Sachdeva, Johannes Wachs, Florian Lemmerich, and CSH External Faculty member Markus Strohmaier appeared in Frontiers in Big Data Vol 4 (2022).