Online reviews shape consumer choices, platform governance and corporate reputation. Fake reviews corrupt this channel by injecting deceptive evidence into rating systems, recommendation pipelines and public trust mechanisms. The rise of large language models (LLM) changes the problem in two ways: LLMs can produce fluent, context‑aware deceptive reviews, while pre‑trained language models (PLM) and LLMs provide stronger semantic representations for detection.
This survey reviews fake review detection from an information‑fusion perspective, covering 211 studies published between 2018 and early 2026. We organize the work by evidence source and fusion level, including review text, sentiment, rating behavior, temporal metadata, user‑product graphs, multimodal content, external knowledge and LLM‑generated signals.
The development trajectory moves from traditional machine learning and deep learning to PLM‑based and LLM‑based methods, examining how textual, behavioral, structural and multimodal evidence are combined. We also analyze performance trends on widely used benchmarks such as Amazon, Yelp and OpSpam, noting limitations caused by varying label construction, data splits and evaluation protocols.
Finally, we identify open challenges: adversarial generation defenses, cross‑domain transfer robustness, uncertainty‑aware fusion, missing‑source robustness, interpretability improvements and trustworthy evaluation of AI‑generated deceptive content.
Review