Princeton Prosody Archive: Item Classification and Related Work

Classification

Code for item classification is found in classification/prosody_classifier.py.

Results for different classifier types:

• Single-layer neural network (NN): 63% accuracy

• Naive Bayes classifier: 87% accuracy

• Support Vector Machine (SVM): 89% accuracy

Run prosody_classifier.py for SVM results. The trained NN is in models/prosody_classifier.pt.

Topic Modelling

Code for topic modeling is found under topic_modeling. Word clouds are zipped under topic_modeling/results.

ID Extraction

Code for extracting IDs from text is found in id_extraction/id_extraction.py.

Authors and Acknowledgment

This project is the work of Eve Fleisig and was created for the Princeton Prosody Archive in 2019-2020.