This is a course project for Deep Learning for NLP course (CMPUT-651) from 2019 [ https://lili-mou.github.io/teaching/651_2019/651_2019.html ] This project attempts to detect propaganda spans from News Articles which is the first Prblem of SemEval 2020 Task-11 [ https://alt.qcri.org/semeval2020/ ]

Implements a BiLSTM for learning representations and uses a CRF for span generation. Models beat the performance from the original paper by Martino et. al. [ https://aclanthology.org/2020.semeval-1.186/ ]

Details of the Dataset The following table gives the summary stats of the dataset:

Sentence Based Models The first approach is to follow the footsteps of (Martino et al., 2019) and do sentence-level classification. Second approach involves treating entire news articles as data samples for the models. The Results are as follows:

Article Based Models The motivation behind the article-based approach is to utilize the context information of the articles which is lost in sentence-level tagging. The Results are as follows: