Can a machine learn to write compelling product reviews?
May 13, 2022
Could you tell if an online review describing the palate-pleasing notes of grapefruit and tangerine in your off-dry Riesling was written by a human, or by an algorithm? Professor Dan Rockmore is part of a team that developed software to produce original expert wine reviews and syntheses of beer reviews using existing datasets and product metadata. Both approaches use transformer-based deep learning neural net architecture. Started by former Neukom Institute fellow Allen Riddell, right, and developed under Rockmore’s guidance by Keith Carlson, Guarini ’21 and postdoctoral research fellow at Tuck, the project became a cross-campus collaboration resulting in a paper in the International Journal of Research in Marketing. Practical applications include tools to assist consumers and professional reviewers, although the paper acknowledges and explores the inherent ethical implications involved. Read more in Dartmouth News and Scientific American, and listen to a short fun segment on BBC Tech Tent (@18:35).