Topic-Guided Variational Autoencoders for Text Generation

TL;DR

This paper introduces TGVAE, a novel text generation model that combines a neural topic-guided prior with a flexible variational autoencoder, enabling the generation of semantically meaningful sentences across diverse topics.

Contribution

The paper presents a new TGVAE model that integrates a neural topic prior with a VAE, using invertible transformations for flexible posterior inference, advancing text generation quality.

Findings

Outperforms existing models in unconditional and conditional text generation

Generates semantically meaningful sentences with diverse topics

Employs invertible Householder transformations for flexible posterior approximation

Abstract

We propose a topic-guided variational autoencoder (TGVAE) model for text generation. Distinct from existing variational autoencoder (VAE) based approaches, which assume a simple Gaussian prior for the latent code, our model specifies the prior as a Gaussian mixture model (GMM) parametrized by a neural topic module. Each mixture component corresponds to a latent topic, which provides guidance to generate sentences under the topic. The neural topic module and the VAE-based neural sequence module in our model are learned jointly. In particular, a sequence of invertible Householder transformations is applied to endow the approximate posterior of the latent code with high flexibility during model inference. Experimental results show that our TGVAE outperforms alternative approaches on both unconditional and conditional text generation, which can generate semantically-meaningful sentences with various topics.