Neural Variational Document Model

Tensorflow implementation of Neural Variational Inference for Text Processing.

model_demo

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

Usage

To train a model with Penn Tree Bank dataset:

$ python main.py --dataset ptb

To test an existing model:

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

Results

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

scalar

histogram

Author

Taehoon Kim / @carpedm20