Extracting and classifying psychoacoustic features from online restaurant reviews

Abstract

Recent work has shown that poor acoustic conditions persist in many restaurants. Owners who receive negative reviews of their establishment's soundscape may struggle to interpret the subjective customer responses into actionable corrective measures. Therefore, further work is needed to taxonomize the acoustically relevant keywords and phrases that occur in user reviews. In this study, an open-source database of restaurant, bar, and coffee shop reviews from across the United States has been obtained. Sentiment analysis and keyword count are used to extract positive, negative, and neutral subjective reviews and subjective features related to acoustics. The resulting subjective features are then categorized, weighed, and linked to objective acoustic parameters using machine learning techniques. Results from this study suggest that owners and consultants may be able to utilize online customer reviews to monitor acoustical comfort and ascertain the nature of an acoustical problem, not only in restaurants, but in any business sector where customers submit online user reviews.