Table 3 Summary of existing research studies employing DL models for clickbait detection.
From: Clickbait detection in news headlines using RoBERTa-Large language model and deep embeddings
Ref | Models | Dataset | Features | Results | Limitation |
|---|---|---|---|---|---|
LSTM | News Headlines | POST | 97 | Limited to linguistic and POS features. | |
CNN | News headlines | Glove | 95 | Assessed only on a specific dataset, reducing generalizability. | |
LSTM | English news | Word embeddings | 94 | Combined DL models but lacked interpretability and scalability evaluation. | |
BERT | Webis-Clickbait-17 | Word2vec | 85 | Computationally expensive and resource heavy. | |
ALBERT | Telugu clickbait headlines | Word2vec, Glove, fast text | 93 | Limited to Telugu dataset. | |
XGB | Twitter and Instagram post | FastText | 88 | Stacking classifiers effective but complex and less interpretable. | |
C-LSTM | News Headlines of US publishers | Word Embeddings | 98 | Proposed extension limited to malicious. | |
BERT | Indonesian news headlines | Sentence embedding | 91 | Lacks cross-domain testing. | |
SBERT | Webs Clickbait Corpus 2017 | Sentence embedding | 84 | dataset specific. | |
BiLSTM | Romanian Clickbait Corpus | Word2Vec | 89 | Ensemble methods effective but dataset (Romanian) small and limited. | |
BiLSTM | News headlines clickbait | Word2Vec | 93 | Bi-LSTM effective but dataset size limited and domain-specific. | |
BERT | WELFake, Fake Newsnet | Fast Text | 97 | Lacks explanation of decision process. | |
DNN | Corpus fake news | Glove | 98 | Limited evaluation scope. | |
BERT | Fake news detection | Word2Vec embedding | 84 | Computationally heavy and dataset specific. | |
CNN | Arabic dataset. | Word2Vec embedding | 77 | Results are tied to specific optimizers and dataset. | |
XLNet | Tweets on 2020 U.S. election | Transformer-based embeddings | 87 | Domain-specific limited to text-only, scalability and real-world deployment not fully addressed |
