Deep Learning Techniques in Social Web Text Classification
Date Issued
February 5, 2021
Type
Μεταπτυχιακή Διπλωματική Εργασία
Abstract
Due to the rapid development of social networks and distribution of vast volumes of user generated content and online information in large amounts, fake news and deception dissemination by malicious accounts is a concerning phenomenon that has emerged in the past decade. In the present Master’s thesis, a complete framework for fake news and deception detection from unstructured social web text is proposed, based on machine learning, text mining and NLP techniques and methods. Novel data sets, consisting of tweets containing fake news and real news regarding the Hong Kong protest movement (summer, 2019), are created. Feature sets with both linguistic-oriented and network-oriented features are examined. Binary classification with various machine learning models and different inputs achieves high overall performance compared to recent related work.
Subjects