How Consumer Feedback Can Predict Product Sales
As the demand for methods continues to outpace what theory can keep up with, researchers like Matthew Schneider, assistant professor of business analytics in LeBow’s Decision Sciences and MIS Department, have turned to application and empirical research.
In one of his latest publications, “Forecasting Sales of New and Existing Products Using Consumer Reviews: A Random Projections Approach,” Schneider takes a different path to understanding the relationship between online reviews and product sales, one that he argues can help companies easily change their strategies regardless of product category and the size of the dataset.
“Intuitively, it made sense that people’s words should have an effect on future purchases of a product,” says Schneider. “The idea of the paper was figuring out how to combine the textual content of consumer reviews with typical product data - such as brand, price, product features, and star rating - to forecast the future weekly sales of that product.”
To save on pre-processing and data collection time, Schneider and his former colleague and co-author, Sachin Gupta, professor of marketing at Cornell University, took a step away from many of the current models and combined two methods: random projections and the bag-of-words approach, a move they argue is far more efficient and performed significantly better than those that ignore the content of consumer reviews.
But instead of combining the 10,000 commonly used words or 10 million two-word phrases with the sales ranks, price and consumer reviews for the roughly 2,000 tablet computers selected in data from an article in Marketing Science Journal – a task that proved nearly impossible – they reduced the computational complexity of the dataset by using random projections to improve the predictive power of the model. With the ability to make one-week-ahead sales forecasts on a rolling basis, the model works well for short-term forecasting, though Schneider states the model can predict further into the future if needed.
“This paper could help companies determine the value of a new product feature compared to the value of a positive (or negative) review. Thus, they’d be able to decide whether they should continue in product development or focus on improving their reviews with customer service,” Schneider says.
The paper is the first to use random projections in business forecasting and since publication has been cited by 18 articles thus far. “But that can only grow,” says Schneider. “A shorter term forecast of popularity is in the number of prospective PhD students that contact Sachin Gupta and I about the paper.”
But Schneider’s research doesn’t stop there. With a background in military and business, Schneider has studied under several advisors in a variety of fields that he credits with helping him find his research niche, “My first academic publication was in the International Journal of Forecasting and I attended my first conference in Nice, France in 2008 with my master’s advisor, Wil Gorr from Carnegie Mellon University,” he said. “It was one of the main reasons I became an academic.”
Prior to joining Drexel, Schneider was an assistant professor of marketing at Northwestern University. His work there in addition to his research while working as a visiting scholar at the Johnson Graduate School of Management at Cornell University and director of research at Fort Rock Asset Management helped Schneider carve a niche for himself in data privacy, an area that he continues to explore. “I started this topic with my PhD advisor, John Abowd, but it did not become popular in business until all the data breaches hit the news,” said Schneider. Schneider’s current research focus is on data privacy in consumer markets.
