Category-Aware Model Improves Sentiment Analysis for Sport Tourism
A study in the Journal of Big Data introduces STC-Sent, a category-aware sentiment network designed for sport tourism destinations using online tourist reviews. The model targets three practical issues: expressions that change sentiment by sport type, a skew of user ratings toward positive feedback, and the need to keep computational cost close to standard encoders. STC-Sent uses RoBERTa to encode review text, then feeds a trainable embedding of the normalized sport category as a query into a category-aware attention module. It also projects four VADER sentiment scores into the same feature space and combines them with a gated fusion module. Tested on 186,742 Yelp Open Dataset reviews across 4,913 businesses and six categories, it reached mean accuracy of 0.913 and mean macro-F1 of 0.845, outperforming baseline RoBERTa by 1.8 and 3.0 percentage points, respectively, with under 1% extra parameters.







