Gabriele Kukauskaite Undergraduate Dissertation 2016/17

Wikipedia Truth-teller

Supervised by A.Vlachos

Abstract

Fact checking is a process of identifying the correctness of a statement. However, carrying out this task manually can become a daunting and time-consuming experience. Therefore, an increasing amount of research has been centered on developing an automated fact checking system. However, despite the efforts no such system exists yet.

Thus, the aim of this project is to build a small scale automated fact checking system that would use the information found on the web. When given a statement, the system would recognize its meaning, decide the correctness of said statement and output True or False.

Distant supervision proved to be a method enabling the generation of a lot training data easily at the expense of its accuracy. It was deemed that the multinomial logistic classifier achieved the best results, reaching an F1 of 95%. However, the system struggles to recognize user entered sentences that are grammatically incorrect.