This repository contains the dataset and code to address the classification task working with the SciCite dataset.
Aim: Given an input citation sentence (“context”), classify its intent as one among {background, method, comparison}.
For details on the approach, refer to our report that is included in the repository.
The following directories contains the code to the respective experiments:
- attention: This directory contains the code comparing uniLSTM with and without self-attention mechanism.
- BERT_models: This directory contains (1) code for the baseline bert model and with sectionName as a feature (2) code for the SciBERT model (3) code for the SciBERT embeddings + LSTM model
- traditional_model: This directory contains the code to all the traditional models we experimented with.
- Word2Vec_LSTM: This directory contains the code to feature selection with LSTM, and comparison with biLSTM.