Detection of propaganda techniques is a research and development project aimed at developing
a high-quality machine learning model to detect propaganda content in online Social Networks.
Limited work has been done in the field. Active research work related to text classification of
news content began after 2016.
Traditional methods of natural language processing have been
outperformed by deep learning methods in the modern era. This research will also revolve
around state-of-the-art deep learning techniques. The content post in the training set is first processed
to form a vector representation of the whole content post which is used to identify the spans containing
propaganda while each propaganda span is then processed to identify the underlying
propaganda technique. Initially, a random statistical model shall be used to establish a baseline
while various variants of recurrent neural networks (such as LSTM and Bi-LSTM) shall be
used for both of these tasks.
Finally, a pre-trained BERT model shall be finely tuned on the
training set to identify the span containing propaganda and the type of propaganda technique
used in the span. The best-performing model is then selected to be used in the second half of
the research. Due to the growing importance of this research work, SemEval has announced
this as one of the tasks for this year’s international workshop. SemEval is responsible for
hosting the datasets and final evaluations for this project.