Εντοπισμός και ανάλυση ψευδών ειδησεογραφικών άρθρων με χρήση βαθέων τεχνητών νευρωνικών δικτύων και μοντελοποίησης θεμάτων
Fake news detection and analysis using artificial neural networks and latent topic analysis
KeywordsΕντοπισμός ψευδών ειδήσεων ; Μοντελοποίηση θεμάτων ; Βαθιά μάθηση ; LDA ; LSTM ; Tensorflow
In the internet era, the spread of fake news has grown and might even lead to legal actions per occasion. The dissemination of information without a prior credibility check on their reliability, combined with their popularity and the targeted audience, may involve risks especially on important issues, such as public health and safety. The increased volume and diffusion speed of the information makes their timely analysis almost impossible except when new technologies are utilized. This dissertation approaches the problem of fake news detection using Lοng Short-Term Memory deep neural networks. Additional semantic analysis was performed to identify and compare patterns in the topics of fake and real news respectively, using natural language processing and topic modeling techniques. The dataset of the experiment is a combination of four separate datasets constituting a unified set of 60,000 fake and real news articles. Two more datasets were combined to construct the unseen dataset for the finalized models. The theoretical background is discussed covering aspects of machine learning and data science. The representation of the methodologies and the implementation of the models is followed by the research results. In conclusion observations and future improvements are discussed.