In this repository, we present our approach and codes for the BigMM Grand Challenge 2020. The contest focuses on building Multimodal classification model on tweets relevant to #MeToo Movement. The tweets comprise of multiple annotated linguistic aspects, like Stance, Relevance, Presence of Hate Speech, Sarcasm, Justification aspects etc.
Our best performing approach ranked first (among 38 submissions) in the leader-board.
- 01-Preproc_baseline.ipynb -> Code for Data Preprocessing and EDA
- 02-Preproc_Tree.ipynb -> Code for Tree and Ensemble methods
- 03-Embedding_BiLSTM.ipynb -> Code for Glove Embeddings and BiLSTM
- 04-BERT_Classification.ipynb -> Code for BERT and Minority Class Oversampling
- 05-ULMFiT_Stance.ipynb -> Code for our ULMFiT Model, load vocabulary from
model/and language model from here.
- Challenge Link : https://sites.google.com/iiitd.ac.in/bigmm2020datachallenge/home
- Competition Platform and Leaderboard : https://www.kaggle.com/c/ieee-bigmm-data-challenge/overviewv
- System Description paper describing the methodology to appear at IEEE Multimedia Big Data