Framework for deriving an email response dataset from the Enron email corpus and training models on the dataset. Implements the models in Tensorflow and uses Jupyter notebooks to interact with them. For details, see the full report on this project.
Modern email applications often highlight important or anomalous emails to users. Towards improving these applications, we develop a neural network for the task of predicting whether a user will respond to an email, given the text of the email. We derive an email response dataset from the Enron email corpus. Using this dataset, we train two types of models: a bi-directional LSTM with attention, and a collaborative filtering model that incorporates sender and recipient information. Our best model achieves an F1 score of 77.02, outperforming our baselines and existing machine learning frameworks for the task.
If you have any questions, feel free to email me at [email protected].