Analyzing soundscape descriptors in online user reviews of restaurants in the United States

Restaurants exhibit complex soundscapes that may vary from venue to venue. While previous works have taxonomized the typical sources in restaurant soundscapes, the extent to which soundscape attributes are described in online user reviews has not been fully explored. This study analyzed an open-source dataset of over 10 million online user-submitted reviews of restaurants across the United States. Potential soundscape-related reviews were flagged using a lexicon of keywords curated by the authors, and false matches were filtered out with a natural language processing pipeline, including a sentence transformer trained on an annotated sub-sample of reviews. Reviews were further analyzed to extract and count soundscape descriptors, source types, and sentiments. Results from this study suggest that only a small proportion of online restaurant reviews are related to soundscapes, but valuable features can still be extracted. Where soundscape descriptors appear in online user-submitted reviews, an informal vocabulary emerges that may differ from professional reviews considered in other works. Results from this project support future efforts to correlate sentiments and descriptors in online reviews with in-situ room acoustic measurements.