The plates have been cleared, the check has been paid, and you’re basking in the afterglow of a delicious restaurant meal. But instead of reaching for that after-dinner mint, you immediately grab your smartphone to tell the world about your experience.
Sound familiar? Over the past decade, online reviews have become an increasingly pervasive part of the experience economy.
For customers, leaving a review is an opportunity to sing the praises of a restaurant or product—or provide a few choice words about a poor experience.
In turn, businesses get immediate feedback from customers. But online reviews can be tricky to analyze. Star ratings never tell the whole story, and text reviews are rife with unstructured data that’s difficult to quantify.
That’s where Ishita Chakraborty comes in. An assistant professor of marketing with a background in computer engineering, Chakraborty harnesses artificial intelligence and algorithms to help businesses make sense of what customers are trying to say—or not say.
“My research team realized that we didn’t need to limit ourselves to just looking at the numerical rating, but we could actually start analyzing in real time the text people were writing,” says Chakraborty. “That really opened up a new Pandora’s box of questions.”
It’s research that Chakraborty began as a PhD student and has continued since joining the Wisconsin School of Business faculty as the Thomas and Charlene Landsberg Smith Faculty Fellow in 2021. One of her first major studies used a text mining framework and specially designed algorithm to help quantify tens of thousands of open-ended restaurant reviews.
“There’s really a lot to learn from consumer silence.”Ishita Chakraborty
The study’s model broke down reviews into individual phrases while identifying relevant attributes in each phrase (such as food or price) and their corresponding sentiments (such as “super tasty” or “my wallet hurts”). From there, the algorithm assigned sentiment scores from one to five, based on how positive or negative the comment was, and aggregated the total scores.
However, user reviews are often missing key information about attributes that would interest a restaurant owner. For instance, a review might mention a restaurant’s amazing food or atmosphere, but say nothing about service or price.
“There’s really a lot to learn from consumer silence,” Chakraborty says. “We want to correctly interpret how a customer actually felt about any missing attributes to help obtain the proper rating. To do that, it helps to understand why reviewers are leaving reviews in the first place.”
While it’s true that reviews can be motivated by many factors, Chakraborty narrows reviewer motivation down to three major segments: status seeking, a need to vent or praise, and good old-fashioned altruism.
“When you have a model that factors in these incentives and then puts the machine learning outputs into the model, you’re able to better evaluate those missing attributes and have more accurate scores,” she says.
It’s a model that could be especially beneficial for chain restaurants. Chakraborty’s research shows that on a five-star scale, chains with strong brand recognition are typically rated almost a full star lower than comparable independent restaurants.
“Reviews of established brands tend to use a lot of words regarding expectations being met or not, whereas when we looked at the reviews of independent restaurants, we generally saw words related to novelty or being surprised,” Chakraborty says.
Meanwhile, as online reviews get more complex, new avenues for research are opening up. Algorithms will need to be adjusted as many reviewers now augment their text reviews with photos. Others have taken to video-based social media platforms, such as YouTube and TikTok, to rant or rave about their experiences.
Reviews are also increasingly becoming a two-way street, with businesses now being able to publicly respond to reviews. Chakraborty is interested in learning what strategies are most effective when responding and how a manager’s response to a review may impact future reviews.
“We always end up finding new questions and that’s what’s exciting about this research,” Chakraborty says. “The nature of questions has been evolving so there’s always something to look forward to.”