140 likes | 249 Vues
This presentation by Jun-Yi Wu, supported by Xue Bai from National Yunlin University of Science and Technology, discusses a novel approach for predicting consumer sentiments from online texts. The focus is on a heuristic search-enhanced Markov blanket model that captures word dependencies while providing an effective vocabulary for sentiment extraction. Using Bayesian networks and Tabu search techniques, the method is validated against existing models, showing comparable and often superior predictive performance in sentiment analysis.
E N D
Predicting consumer sentiments from online text Presenter: Jun-Yi Wu Authors: XueBai 國立雲林科技大學 National Yunlin University of Science and Technology 2011 DSS
Outline • Motivation • Objective • Methodology • Experiments • Conclusion • Comments
Motivation • In recent years, due to the sheer volume of online reviews and news corpora available in digital form. • An accurate method not only predicting consumer sentiments but also can be used to reduce the risk.
Objective • To propose a heuristic search-enhanced Markov blanket model that is able to capture the dependencies among words and provide a vocabulary that is adequate for the purpose of extracting sentiments. News
Methodology • Bayesian network and Markov blanket • Tabu search • The Markov blanket for a sentiment variable
Methodology • Bayesian network and Markov blanket
Methodology • Tabu search • Tabu search is a meta-heuristic search method that is able to guide traditional local search methods to escape local optima with the assistance of adaptive memory. • Tabu search starts with a feasible solution and chooses the best move according to an evaluation function, while taking steps to ensure that the method does not revisit a solution previously generated.
Methodology • The Markov blanket for a sentiment variable • The algorithm for learning the Markov Blanket for a sentiment variable is called a Markov Blanket Classifier.
Conclusion • 1 12
Comments • Advantage • This method yields predictive performance comparable and in many cases superior to those of other state-of-the-art classification methods. • Application • Sentiment analysis 13