JacoDuToit11/CESMClassification

Hierarchical classification of South African research outputs # Hierarchical text categorization of South African research outputs # This objective of this project is to build a hierachical text classifier for South African research outputs. We consider different feature extraction methods to obtain a semantic representation of the text document which comprises the title and abstract of a research output. We evaluate several classifier architectures which include Convolutional Neural Networks, Recurrent Neural networks and attention mechanisms. ## Install dependencies ## $pip install -r requirements.txt ## Dataset construction and feature extraction ## ### Data.py ### Builds the dataset, removes invalid entries and balances dataset: $python Data.py ### LDA.py ### Obtains embeddings from LDA model: $python LDA.py ### FeatureExtraction.py ### Obtain embeddings from SciBERT model and performs fusion strategies: $python FeatureExtraction.py ## Model training and evaluation ## ### ClassifierTuning.py ### Trains, tunes and evaluates four of the classifier architectures for the fusion strategies. $python ClassifierTuning.py ### AttentionEmbedding.py ### Trains, tunes and evaluates label embedding CNN for the fusion strategies. $python ClassifierTuning.py ### TokenizerClassifier.py ### Trains, tunes and evaluates four of the classifier architectures for baseline feature extraction method. $python TokenizerClassifier.py ### TokenizerEmbeddingAttention.py ### Trains, tunes and evaluates label embedding CNN for baseline feature extraction method. $python TokenizerEmbeddingAttention.py