Abstract:
This thesis presents a framework for fake news detection from the social media
platform.The data is extracted from Twitter because it is one of the largest social
platforms that promote news rapidly. Mostly the news promoters in the field of media
use Twitter for spreading of news. There is a lot of work that has been done by
different researches in the field of fake news detection in which they used different
approaches, however, most of such approaches are validated on English. We are
presenting a method that uses a hybrid approach for the detection of fake news in
Urdu.
The first method in our framework is content based approach in which embedding
feature is used that converts text data into vector. Word2vec technique is applied for
this purpose. The second approach that has been used for this research is Social Base
approach, in which we used engagement feature as our data set comprises of extra
information such as re-tweet count, likes, Favourite Count. The third approach is
Network Based Approach, in which network feature is used considering Friends count
and follower count.
The presented work uses Deep Learning Models for better performance. Two types of
deep learning models such as CNN and LSTM are implemented. The results from
LSTM Model are more accurate as 72% accuracy obtained from this model when
Content Base approach used in this model. When we use hybrid approach in LSTM
the accuracy increased to 73%. In CNN model when we used Content Base Approach
the 67% accuracy obtained, but in hybrid approach we got 64% accuracy. The LSTM
results are higher than CNN in fake news detection.
