Scholarlink

<p>Sentiment Analysis (SA) involves gathering and analyzing text to determine its polarity. Many users express their opinions through Internet-based applications, such as social media websites, comment sections, and blogging platforms. TripAdvisor Inc. is an American company that partners with travel agencies across various domains, including flight booking and the hotel industry. Sentiment analysis has been applied to TripAdvisor.com reviews to gauge user sentiment. For this study, a dataset of hotel reviews with ratings from 1 to 5 is used to train models at different levels. This research analyzes sentiment at three levels: (i) the entire review, (ii) individual sentences, and (iii) specific aspect terms. The BERT model is used for Document-level Sentiment Analysis (DLSA), achieving 84.12% accuracy. A BiLSTM with an attention module analyzes sentences within reviews, obtaining 83.68% accuracy. Both these analyses are five-class classification tasks. Additionally, the BiLSTM with attention performs Aspect-level Sentiment Analysis (ALSA), classifying aspects as positive, negative, or neutral (three-class classification), achieving an average accuracy of 98.69% ± 0.03%. For a real-time demonstration, these architectures have also been used to analyze sentiment in reviews scraped from TripAdvisor.com. The models demonstrate strong inference performance in real-time settings on the evaluated dataset. While ALSA performs well on shorter sentences, longer sentences with more aspect terms may affect performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-level sentiment analysis for the TripAdvisor reviews using deep learning

  • 期刊论文

  • 发表日期：

    2026年6月17日

查看全文
  • Anjum Madan,
  • Devender Kumar

摘要

Sentiment Analysis (SA) involves gathering and analyzing text to determine its polarity. Many users express their opinions through Internet-based applications, such as social media websites, comment sections, and blogging platforms. TripAdvisor Inc. is an American company that partners with travel agencies across various domains, including flight booking and the hotel industry. Sentiment analysis has been applied to TripAdvisor.com reviews to gauge user sentiment. For this study, a dataset of hotel reviews with ratings from 1 to 5 is used to train models at different levels. This research analyzes sentiment at three levels: (i) the entire review, (ii) individual sentences, and (iii) specific aspect terms. The BERT model is used for Document-level Sentiment Analysis (DLSA), achieving 84.12% accuracy. A BiLSTM with an attention module analyzes sentences within reviews, obtaining 83.68% accuracy. Both these analyses are five-class classification tasks. Additionally, the BiLSTM with attention performs Aspect-level Sentiment Analysis (ALSA), classifying aspects as positive, negative, or neutral (three-class classification), achieving an average accuracy of 98.69% ± 0.03%. For a real-time demonstration, these architectures have also been used to analyze sentiment in reviews scraped from TripAdvisor.com. The models demonstrate strong inference performance in real-time settings on the evaluated dataset. While ALSA performs well on shorter sentences, longer sentences with more aspect terms may affect performance.