Large-scale category-aware sentiment analysis for sport tourism destinations using tourist reviews

Online reviews offer extensive, continuously updated insights into visitor experiences at sport tourism destinations. However, analysing these data at scale presents three challenges: the sentiment polarity of many expressions varies with the sport activity under review, ratings are heavily biased towards positive feedback, and the large volume of text calls for models with computational costs comparable to those of a standard encoder. To tackle these challenges, this study introduces STC-Sent, a sport tourism category-aware sentiment network designed for three-class sentiment classification. STC-Sent uses the Robustly Optimized BERT Pretraining Approach (RoBERTa) to encode each review. A trainable embedding of the normalized sport category acts as the query in a category-aware attention (CAA) module, which assigns weights to contextual token representations to generate a category-aware review representation. Four sentiment scores obtained from the Valence Aware Dictionary and sEntiment Reasoner (VADER) are projected into the same representation space and integrated with the review representation through a feature-wise lexicon-informed gated fusion (LGF) module. The entire network is trained end to end using cross-entropy loss. BERTopic is then applied to reviews predicted as positive or negative to identify the principal sources of satisfaction and dissatisfaction. Experiments used 186,742 reviews covering 4,913 sport-related businesses across six categories in the Yelp Open Dataset, with a business-level 70:15:15 split and five random seeds. STC-Sent attained a mean accuracy of 0.913 and a mean macro-F1 of 0.845, exceeding the strongest baseline, RoBERTa, by 1.8 and 3.0% points, respectively, and achieved the highest mean macro-F1 in all six categories. These gains required about 0.8 million additional parameters (less than 1% of RoBERTa), 4.9% more training time, and 0.4 s of extra inference time per 1,000 reviews. Topic analysis reveals that positive reviews primarily focus on scenery, routes, and staff service, while negative reviews predominantly address crowding, price, and maintenance.

Authors

Institutions

Publication Details

Journal
Journal Of Big Data
Published
2026-11-20
DOI
https://doi.org/10.1186/s40537-026-01623-1
Primary Topic
Sentiment Analysis and Opinion Mining
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Large-scale category-aware sentiment analysis for sport tourism destinations using tourist reviews

Mohammad Jafar Mokarram, Ping Liu, Mengmeng Lei, Ali Fayazi et al.
3 links
Journal Of Big Data
Sentiment Analysis and Opinion Mining
article

Large-scale category-aware sentiment analysis for sport tourism destinations using tourist reviews

Mohammad Jafar Mokarram, Ping Liu, Mengmeng Lei, Ali Fayazi, Hossein Ghayoumi Zadeh
article en

Abstract

Online reviews offer extensive, continuously updated insights into visitor experiences at sport tourism destinations. However, analysing these data at scale presents three challenges: the sentiment polarity of many expressions varies with the sport activity under review, ratings are heavily biased towards positive feedback, and the large volume of text calls for models with computational costs comparable to those of a standard encoder. To tackle these challenges, this study introduces STC-Sent, a sport tourism category-aware sentiment network designed for three-class sentiment classification. STC-Sent uses the Robustly Optimized BERT Pretraining Approach (RoBERTa) to encode each review. A trainable embedding of the normalized sport category acts as the query in a category-aware attention (CAA) module, which assigns weights to contextual token representations to generate a category-aware review representation. Four sentiment scores obtained from the Valence Aware Dictionary and sEntiment Reasoner (VADER) are projected into the same representation space and integrated with the review representation through a feature-wise lexicon-informed gated fusion (LGF) module. The entire network is trained end to end using cross-entropy loss. BERTopic is then applied to reviews predicted as positive or negative to identify the principal sources of satisfaction and dissatisfaction. Experiments used 186,742 reviews covering 4,913 sport-related businesses across six categories in the Yelp Open Dataset, with a business-level 70:15:15 split and five random seeds. STC-Sent attained a mean accuracy of 0.913 and a mean macro-F1 of 0.845, exceeding the strongest baseline, RoBERTa, by 1.8 and 3.0% points, respectively, and achieved the highest mean macro-F1 in all six categories. These gains required about 0.8 million additional parameters (less than 1% of RoBERTa), 4.9% more training time, and 0.4 s of extra inference time per 1,000 reviews. Topic analysis reveals that positive reviews primarily focus on scenery, routes, and staff service, while negative reviews predominantly address crowding, price, and maintenance.

Journal Of Big Data
Vali Asr University of Rafsanjan (IR), Anhui Xinhua University (CN), Hebei Finance University (CN)
Sentiment Analysis and Opinion Mining
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon