Using Emotions to Help Detect Fake Tourism Reviews

Almansour, Mansour and Piao, Scott and Rayson, Paul (2026) Using Emotions to Help Detect Fake Tourism Reviews. PhD thesis, Lancaster University.

Text (2026mansourphd)
2026mansourphd.pdf - Published Version
Download (4MB)

Abstract

Detecting fake reviews has emerged as a urgent critical challenge in NLP, with significant implications for practical applications across various industries. In the tourism sector, fake reviews pose a serious threat by misleading potential customers, damaging the credibility of genuine businesses, and eroding trust in online platforms. The ability to effectively identify and handle deceptive content is essential for safeguarding tourism consumers and maintaining the integrity of related review platforms. While the NLP community has made notable advancements in fake tourism reviews detection, there remains limited understanding of how emotion information can be leveraged to enhance the detection of tourism fake reviews. This research fills this gap by exploring the integration of emotional information into LLM-based approaches for tourism fake review detection. Specifically, the study evaluates the performance of three prominent large language models (LLMs), including BERT, DistilBERT, and RoBERTa on a tourism-specific dataset, named GeFaRe, both with and without the inclusion of emotion-based information. The dataset used in this research was annotated with emotional labels derived from Plutchik’s eight primary emotions, allowing for the incorporation of rich emotional information into the fake reviews classification process. Various experimental configurations were designed to assess the impact of emotional information on the performance of each of the models. My study provides compelling evidence of the value of integrating emotion information into LLM-based fake tourism review detection models. Across all of my configurations, the inclusion of emotional information led to an improvement of most models’ performance, with DistilBERT achieving the highest levels of accuracy up to 0.956. This underscores the potential of emotional information as key information in the context of fake tourism review detection. Furthermore, my findings highlight the importance fine-tuning the LLMs for domain-specific datasets for emotion and fake tourism reviews detection.