Online Program Home
  My Program

Abstract Details

Activity Number: 529 - SPEED: Machine Learning
Type: Contributed
Date/Time: Wednesday, August 2, 2017 : 10:30 AM to 11:15 AM
Sponsor: Section on Statistical Learning and Data Science
Abstract #325321
Title: Joint Sentiment Topic Modeling of Text Data
Author(s): Sahba Akhavan Niaki* and George Michailidis
Companies: University of Florida and University of Florida
Keywords: Topic Modeling ; Sentiment Analysis ; Latent Dirichlet Allocation
Abstract:

Objective texts like reviews, critics, and comments are mostly written with known sentiments toward each topic in mind. In this work, we propose a novel extension of Latent Dirichlet Allocation model for joint sentiment-topic modeling of text data that extracts latent topic proportions and sentiment proportions toward each of the topics. Incorporating prior knowledge on sentiment words, we propose a new iterative approach based on Gibbs Sampling for approximate estimation of hyperparameters of the introduced three level hierarchical model. We used the new model for (topic) sentiment classification of different synthetic datasets and observed significant improvement over some existing models. Analyzing a real data set also shows the applicability of the proposed model.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2017 program

 
 
Copyright © American Statistical Association