Modeling documents with Generative Adversarial Networks

TL;DR

This paper introduces a novel approach using Generative Adversarial Networks with a Denoising Autoencoder discriminator to learn distributed representations of natural language documents, evaluated through various methods.

Contribution

It presents a new GAN-based model for document representation that leverages a Denoising Autoencoder as the discriminator, enhancing the quality of learned embeddings.

Findings

Effective document embeddings learned

Quantitative and qualitative evaluation conducted

Demonstrates potential for NLP applications

Abstract

This paper describes a method for using Generative Adversarial Networks to learn distributed representations of natural language documents. We propose a model that is based on the recently proposed Energy-Based GAN, but instead uses a Denoising Autoencoder as the discriminator network. Document representations are extracted from the hidden layer of the discriminator and evaluated both quantitatively and qualitatively.