Abstract: ‍ ‍The spread of fake news has a great negative impact on personal development， social stability and national security. Therefore， the detection of fake news has always been a hot issue. Fake news detection is typically a classification problem aiming at verifying the authenticity of news which is composed of multimedia contents such as texts， images and videos. This paper conducted a systematic study on fake news detection and its current mainstream methods. We pointed out an intrinsic characteristic of fake news， that is， fake news usually has specific propagation intentions， such as misleading the public， which is different from truth. Based on this characteristic， we first generally divided the propagation intentions of fake news into three categories， and analyzed the current research methods according to the corresponding relevant characteristics， aiming to enable readers to better understand this field from a new perspective. This paper also introduced the problem definition， basic paradigm， common datasets， evaluation metrics and state-of-the-art performances of fake news detection， and outlined some potential directions for future research.