Ανίχνευση Ψευδών Ειδήσεων στο Twitter μέσω Μηχανικής Μάθησης
Date Issued
February 27, 2023
Type
Πτυχιακή Εργασία
Abstract
Internet use has prompted new ways of communicating among people. Social media has become integrated into the daily lives of hundreds of millions of people covering a wide range of their needs, including the need for information; in recent years, significant dimensions have taken on the problem of spreading fake news on twitter due to a number of advantages that this offers (low cost and convenience). One way to tackle it is to mobilize machine learning methods and techniques. This work tested supervised machine learning algorithms offered by the Weka environment that offers user-friendliness, a wide variety of algorithms, and ease of educational data processing. The original data on which the work was based relates to the dissemination of fake news on Twitter as it took place at the Hong Kong protests in 2019. After selecting the characteristics to achieve dimension reduction, two training sets with 8 and 4 characteristics were produced, respectively. All supervised machine learning algorithms achieved accuracy and recall performance of over 90 % with the best being the multi-layer Perceptron neural network and the decision tree C4.5 (J-48) with a mean accuracy, recall and F1 performance of 93%. This means that effective detection of fake news is possible if more representative training data are created enhanced with information on senders.
Subjects