By Fangzhou Liu
Polite language that appeals to authority and cultural norms can boost product sales, Stanford scholars found.
Second-year computer science PhD candidate Reid Pryzant, linguistics professor Dan Jurafsky and Young-joo Chung, a lead scientist at Japanese e-commerce platform Rakuten and a former visiting scholar at Stanford, applied machine learning methods to over 90,000 food and health-related product descriptions on Rakuten. They found that the greater the number of words that demonstrated respect for the customer or suggested authority, the greater the volume of product sales.
“Product descriptions are fundamentally a kind of social discourse, one whose linguistic contents have real control over consumer purchasing behavior,” the researchers wrote in an article presented at the SIGIR Workshop on eCommerce in Tokyo, Japan. “Business owners employ narratives to portray their products, and consumers react accordingly.”
The online retail industry has long sought to understand why the same product sells with varying degrees of success across different listings and websites. Previous research has explored consumers’ responses to product reviews and word-of-mouth recommendations, but understanding the effect of advertising language on sales has been tricker, according to the researchers.
For instance, brand names like “Nike” or phrases like “free shipping” might boost sales — but these words only indicate facts about the product and the sales strategy rather than the linguistic devices that advertisers employ.
“We’re more interested in framing,” Jurafsky told Stanford News. “How do advertisers frame the text to appeal to people independent of those other obvious sales factors?”
To isolate the impact of advertising language, Pryzant suggested adversarial machine learning, a cutting-edge statistical technique that plies predictive models against each other.
The method allowed researchers to identify words that correlated with high sales independently of a predetermined pricing strategy or factual information about the product. But Pryzant and Jurafsky did not expect the approach to be as successful as it turned out to be.
“Adversarial learning is a really hot topic right now,” Jurafsky said. “But it’s been challenging to get it to work for language. So this is really exciting technically and suggests other potential applications.”
Previously, adversarial learning was mostly applied to image analysis rather than language, forcing the team to adapt the method to the new task.
“The idea came quickly, but fitting the technique to our needs was hard and took time,” Pryzant said. “But the model was good at predicting sales on the first try, which was a gratifying result.”
The method allowed researchers to identify sales-boosting words and make generalizations about the kind of advertising language that appeals to Japanese consumers. In addition to linking product success to words and suffixes that indicated deference and respect, the researchers found that higher-selling items tended to give more specifics about product features.
The team also found that tradition was a major framing device among successful products. Words and phrases that suggest authority and cultural institutions, such as “long-standing shop,” “Christmas” or “year-end gift” helped products to sell better.
Jurafsky noted that the results built on his findings in his 2014 book “The Language of Food: A Linguist Reads the Menu,” which analyzed the language of food advertising and menus.
“Using words that appeal to tradition — we also saw that on American menus and even on the back of potato chip bags,” Jurafsky said. “Talking about authenticity and tradition is just a really useful framing device.”
According to Jurafsky and Pryzant, the next step is to expand the scope of their study to other languages, such as Chinese and English. Jurafsky commented that he was interested in parallels and contrasts between advertising words in different languages, since different cultures might have different norms for interacting with consumers and invoking tradition.
More generally, the researchers were also concerned with the moral implications of studies on framing devices, which might be used to manipulate the public both in business and in politics.
“From my perspective as a linguist, I think the more we know about how people are using language to influence us, the better,” said Jurafsky. “If we as consumers know that people are using certain kinds of framings, that has to help us spot when we’re being manipulated.”