bs-thesis

Sentiment Analysis and Factuality classification of medical conversations on based on eDisease dataset. Full English report here.

Dataset

Polarity classes:

• POSITIVE
• NEGATIVE
• NEUTRAL

Factuality classes:

• EXPERIENCE
• OPINION
• FACT

Classifiers

Machine learning

Feature extraction was done using tf-idf algorithm. Machine learning models used:

• Gaussian Naive Bayes
• Random Forest
• Passive Aggressive

Deep learning

Word vector representation

Word vectors were trained based on medical text and also forums conversations text using gensim python module

Deep learning architectures

• 1 Dimensional CNN Model based on Convolutional Neural Networks for Sentence Classification, Yoon Kim
• Deep Bidirectional RNN with LSTM cells
• 1 Layer of 1 Dimensional CNN + Deep RNN Layer

Poster