Neural Variational Document Model

Tensorflow implementation of Neural Variational Inference for Text Processing.

This implementation contains:

1. Neural Variational Document Model
   • Variational inference framework for generative model of text
   • Combines a stochastic document representation with a bag-of-words generative model
2. Neural Answer Selection Model (in progress)
   • Variational inference framework for conditional generative model of text
   • Combines a LSTM embeddings with an attention mechanism to extract the semantics between question and answer

Prerequisites

• Python 2.7 or Python 3.3+
• NLTK
• TensorFlow

Usage

To train a model with Penn Tree Bank dataset:

normal$ python main.py --dataset ptb
normal

To test an existing model:

normal$ python main.py --dataset ptb --forward_only True
normal

Results

Training details of NVDM. The best result can be achieved by onehost updates, not alternative updates.

Author

Taehoon Kim / @carpedm20