Multi-task-Neural-Networks-for-Hierarchical-Text-Classification-Machine-Learning-Research-Project-

Datasets

• Amazon Product Reviews https://www.kaggle.com/kashnitsky/hierarchical-text-classification
• DBPedia https://www.kaggle.com/danofer/dbpedia-classes
• Coronavirus Tweets https://www.kaggle.com/smid80/coronavirus-covid19-tweets

Experiment Design

• Accuracy is used for evaluation metric
• Randomly train test split with percentages: 80%, 20%
• NLTK is used for text preprocessing and tokenization
• Word2Vec for word embedding

Models tried

• Fully Connected Network as Baseline
• Shallow/Deep CNN
• Shallow/Deep LSTM
• With/Without Multitask network structure

Results

• All the Multitask networks outperform the vanilla networks
• Shallow LSTM achieve the highest accuracy, but didn't benefit from stacking more layers
• CNN benefit from stacking more layers with residual connections

---

Evaluating-Student-Writing-kaggle (Experiments with Transformer based model)

https://www.kaggle.com/c/feedback-prize-2021