Using neutral examples for learning polarity

DMCA

Using neutral examples for learning polarity (2005)

Cached

Download Links

Other Repositories/Bibliography

DBLP

by Jonathan Schler

Venue:	In Proc. of IJCAI (poster
Citations:	5 - 0 self

BibTeX

@INPROCEEDINGS{Schler05usingneutral,
    author = {Jonathan Schler},
    title = {Using neutral examples for learning polarity},
    booktitle = {In Proc. of IJCAI (poster},
    year = {2005}
}

OpenURL

Abstract

Sentiment analysis is an example of polarity learning. Most research on learning to identify sentiment ignores “neutral ” examples and instead performs training and testing using only examples of significant polarity. We show that it is crucial to use neutral examples in learning polarity for a variety of reasons and show how neutral examples help us obtain superior classification results in two sentiment analysis test-beds. Many machine-learning problems involve predicting an example’s polarity: is it (significantly) greater than or less than some standard. One canonical example of learning polarity is sentiment analysis, the determination of whether a particular text expresses positive or negative sentiment regarding some issue. The problem of how to exploit a labeled corpus to learn models for sentiment analysis has attracted a good deal of interest in recent years [Dave et al 2003, Pang et al 2002, Shanahan et al 2005]. One common characteristic of almost all this work has been the tendency to define the task as a two-category problem: positive versus negative. In almost all actual polarity problems, including sentiment analysis, there are, however, three categories that must be distinguished: positive, negative and neutral. Not every comment on a product or experience expresses purely positive or negative sentiment. Some – in many cases, most – comments might report objective facts without expressing any sentiment, while others might express mixed or conflicting sentiment. Researchers are aware, of course, of the existence of neutral documents. The rationale for ignoring them has been a reliance on two tacit assumptions: • Solving the binary positive vs. negative problem automatically solves the three-category problem since neutral documents will simply lie near the boundary of the binary model • There is less to learn from neutral documents than from documents with clearly defined sentiment

Keyphrases

neutral example sentiment analysis neutral document negative sentiment sentiment analysis test-beds binary positive v negative problem experience express particular text significant polarity binary model sentiment ignores neutral example positive versus many case common characteristic good deal example polarity canonical example polarity learning superior classification result labeled corpus three-category problem two-category problem objective fact tacit assumption many machine-learning problem recent year actual polarity problem conflicting sentiment

DMCA