Fake news detection on social media is a critical issue due to the rapid spread of misinformation. This survey provides a comprehensive review of fake news detection, including its characteristics, existing algorithms, evaluation metrics, and datasets. Fake news is intentionally created to mislead readers, making it challenging to detect based solely on content. Auxiliary information, such as user social engagements, is necessary for effective detection. However, this data is often large, incomplete, and noisy, posing additional challenges. The survey discusses the psychological and social theories behind fake news, existing detection methods, and open research problems. It also highlights the importance of social context in fake news detection, including malicious accounts, echo chambers, and network-based features. The survey presents a formal definition of fake news detection as a binary classification problem and outlines the features extracted from news content and social context. Existing methods include knowledge-based, style-based, and social context models. The survey also discusses available datasets and evaluation metrics, noting their limitations. Finally, it proposes a new dataset, FakeNewsNet, to address the shortcomings of existing datasets. The survey concludes that fake news detection remains a challenging and important research area with many open problems requiring further investigation.Fake news detection on social media is a critical issue due to the rapid spread of misinformation. This survey provides a comprehensive review of fake news detection, including its characteristics, existing algorithms, evaluation metrics, and datasets. Fake news is intentionally created to mislead readers, making it challenging to detect based solely on content. Auxiliary information, such as user social engagements, is necessary for effective detection. However, this data is often large, incomplete, and noisy, posing additional challenges. The survey discusses the psychological and social theories behind fake news, existing detection methods, and open research problems. It also highlights the importance of social context in fake news detection, including malicious accounts, echo chambers, and network-based features. The survey presents a formal definition of fake news detection as a binary classification problem and outlines the features extracted from news content and social context. Existing methods include knowledge-based, style-based, and social context models. The survey also discusses available datasets and evaluation metrics, noting their limitations. Finally, it proposes a new dataset, FakeNewsNet, to address the shortcomings of existing datasets. The survey concludes that fake news detection remains a challenging and important research area with many open problems requiring further investigation.