Communications of the ACM
Stanford Algorithm Analyzes Sentence Sentiment, Advances Machine Learning
The new Neural Analysis of Sentiment software features a recursive deep-learning algorithm that advances machine learning by enabling computers to understand the meaning of words in context.
Credit: Coursera.org
Stanford University computer scientists have developed Neural Analysis of Sentiment (NaSent), software that analyzes sentences from movie reviews and rates the sentiments they express on a five-point scale.
NaSent is a recursive deep-learning algorithm that advances machine learning by enabling computers to understand the meaning of words in context, similarly to humans. Using NaSent, computers can extract information from language without referring constantly to dictionaries or rules created by humans.
"Here the system is being trained to move beyond words to phrases and sentences, and to capture the sentiments of these word combinations," says University of Montreal professor and LISA Machine Learning Laboratory head Yoshua Bengio.
The scientists began with a dataset of approximately 12,000 movie review sentences, and used automated techniques to parse groups of words into grammatical units, consisting of 214,000 phrases and sentences. The grammatical units were each read by three humans and rated on intensity of like or dislike. Without further intervention, NaSent then computed its own framework for predicting the sentiments that these words, phrases, and sentences conveyed.
The researchers say NaSent currently has about an 85-percent accuracy rate, compared with 80 percent for previous word-based evaluation systems.
The team is soliciting public input on NaSent, and hopes to refine the system through crowdsourcing.
From Stanford University
View Full Article
Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA
No entries found