A REVIEW ON DEEP SEQUENTIAL MODELS FOR FAKE NEWS IDENTIFICATION USING WORD EMBEDDING TECHNIQUES

Abstract:-Fake news has emerged as a critical challenge in the digital ecosystem, accelerating misinformation across social media, news portals, and online communication networks. Existing machine learning approaches often fail to capture long-range contextual semantics, resulting in limited accuracy and poor real-world generalization. This review paper provides a comprehensive analysis of deep sequential models LSTM, BiLSTM, and GRU for fake news detection, focusing on how these architectures improve contextual understanding and classification performance. The study examines key preprocessing methods such as tokenization, padding, and word embedding techniques including TF-IDF, Word2Vec, Fast Text, and embedding layers. It also synthesizes findings from recent literature, highlighting that many traditional and lightweight NLP models struggle to achieve accuracy beyond 80% due to dataset imbalance, lexical ambiguity, and domain-shift challenges. The review compares model performance, evaluates their strengths and limitations, and identifies critical gaps related to dataset diversity, multimodal fusion, and transformer integration.

---