Semi-Supervised Recursive Autoencoders for Predicting Sentiment Distributions

TLDR

The paper proposes a recursive autoencoder framework for predicting sentiment label distributions at the sentence level. The method employs recursive autoencoders to learn vector representations of multi‑word phrases and predicts sentiment label distributions, evaluated on movie reviews and a confessions dataset of user stories annotated with multiple labels. The model outperforms state‑of‑the‑art baselines on movie review datasets and more accurately predicts sentiment label distributions on the confessions dataset.

Abstract

We introduce a novel machine learning framework based on recursive autoencoders for sentence-level prediction of sentiment label distributions. Our method learns vector space representations for multi-word phrases. In sentiment prediction tasks these representations outperform other state-of-the-art approaches on commonly used datasets, such as movie reviews, without using any pre-defined sentiment lexica or polarity shifting rules. We also evaluate the model's ability to predict sentiment distributions on a new dataset based on confessions from the experience project. The dataset consists of personal user stories annotated with multiple labels which, when aggregated, form a multinomial distribution that captures emotional reactions. Our algorithm can more accurately predict distributions over such labels compared to several competitive baselines.