🧑‍🔬 Eric Huang

Eric Huang works on neural representations of language, specifically how continuous vectors for words and longer stretches of text can be learned and then used for tasks such as sentiment prediction. Their papers here, from 2011 and 2012, share a concern with going beyond simple word counts and narrow context windows. They capture composition, word sense and topical context through learned vector spaces, ranking-style language models and clustering of word occurrences.

The most cited of these works is Semi-Supervised Recursive Autoencoders for Predicting Sentiment Distributions (2011). It builds sentence representations by repeatedly merging neighbouring word vectors into phrase vectors, choosing at each step the pair that is best reconstructed. This yields a tree structure and a fixed-length sentence vector without relying on a parser or hand-built sentiment lexica, and a classifier on the nodes predicts distributions over sentiment. The second, Improving Word Representations via Global Context and Multiple Word Prototypes (2012), adds a document-level scoring term to a local window model, so that embeddings reflect topical cues as well as word order. It then clusters the contexts in which each word appears and learns a separate vector for each cluster, giving words several prototypes for different senses. The paper also introduces an evaluation in which word pairs are judged within whole sentences, with WordNet used to sample distinct senses.

The 2012 paper is the newest work on the site. It moves the earlier focus on composing sentence meaning toward the question of how individual word vectors should handle polysemy and wider document context.

The bylines place Huang at Stanford University between 2011 and 2015. Both papers here were written with Richard Socher, Christopher D. Manning and Andrew Y. Ng. The sentiment work falls under the topic Sentiment & emotion.

Eric Huang has 2 papers in our library, in 2011–2012, most of them on Sentiment & emotion; often with Andrew Y. Ng, Christopher D. Manning, Richard Socher.