Learning to Generate Reviews and Discovering Sentiment

TL;DR

This paper demonstrates that byte-level recurrent language models can learn disentangled high-level features like sentiment in an unsupervised manner, achieving state-of-the-art results with minimal labeled data and enabling sentiment-controlled text generation.

Contribution

It reveals that a single unit in unsupervised byte-level models can perform sentiment analysis and influence generation, advancing understanding of learned representations.

Findings

A single unit encodes sentiment in the model.

State-of-the-art sentiment classification with minimal labeled data.

Sentiment control in generated samples by fixing the sentiment unit.

Abstract

We explore the properties of byte-level recurrent language models. When given sufficient amounts of capacity, training data, and compute time, the representations learned by these models include disentangled features corresponding to high-level concepts. Specifically, we find a single unit which performs sentiment analysis. These representations, learned in an unsupervised manner, achieve state of the art on the binary subset of the Stanford Sentiment Treebank. They are also very data efficient. When using only a handful of labeled examples, our approach matches the performance of strong baselines trained on full datasets. We also demonstrate the sentiment unit has a direct influence on the generative process of the model. Simply fixing its value to be positive or negative generates samples with the corresponding positive or negative sentiment.