Tech Demos (edit)
Snippext is an information extraction tool for extracting opinion aspects, sentiments, and customer experiences from user-generated content (e.g., online hotel reviews). Based on the powerful NLP model BERT, Snippext achieved state-of-the-art accuracies by adopting an array of optimizations including data augmentation, multitask learning, and semi-supervised learning.
Customers seek travel experiences that fulfill their desires. However, e-commerce search engines only support queries involving objective attributes, such as location, price, and cuisine. Whereas, experiential data is relegated to text reviews. Therefore, a database system must model subjective data and process queries in the user’s own words to support experiential queries. At the same time, it must specify predicates involving objective attributes. We developed Opine, a subjective database system, that addresses these challenges as well as its frontend, Voyageur, an experiential travel search engine.