Semi-Supervised Recursive Autoencoders for Predicting Sentiment Distributions

摘要: 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 multi-word phrases. In tasks these outperform other state-of-the-art approaches commonly used datasets, such as movie reviews, without using any pre-defined lexica or polarity shifting rules. also evaluate the model's ability to predict distributions new dataset confessions from experience project. The consists personal user stories annotated with multiple labels which, when aggregated, form multinomial distribution that captures emotional reactions. algorithm can more accurately over compared several competitive baselines.