Title:
|
A SUBJECTIVITY DETECTION METHOD FOR OPINION MINING BASED ON LEXICAL RESOURCES |
Author(s):
|
Eduardo Bezerra, Bianca Firmino, Rafael Castaneda, Jorge Soares, Eduardo Ogasawara, Ronaldo Goldschmidt |
ISBN:
|
978-989-8533-01-2 |
Editors:
|
Bebo White, Pedro Isaías and Flávia Maria Santoro |
Year:
|
2011 |
Edition:
|
Single |
Keywords:
|
Document classification, opinion mining, subjectivity detection. |
Type:
|
Full Paper |
First Page:
|
317 |
Last Page:
|
324 |
Language:
|
English |
Cover:
|
|
Full Contents:
|
click to dowload
|
Paper Abstract:
|
Extraction and monitoring of opinions about a product or service is much valuable to users of social media and to companies as a feedback mechanism and as a source of information to define marketing campaigns. Opinion Mining is a subfield of Data Mining which aims at automatically classify opinions with respect to their polarity, i.e., to infer if the author of an opinion has either a positive or negative sentiment about some subject (e.g., a movie, a car model, an appliance, etc). A common approach in the opinion mining task is to preprocess the collection of opinion texts used as training set in order to remove the sentences of an opinion that do not contain subjective content. These so called subjective extracts, one for each opinionated text, are then used in a second step to train a polarity classifier, that is used to predict the orientation of the original opinion (positive or negative). In this paper, we propose a new method for the subjectivity detection step of the opinion mining task. Our method is based on Part-of-Speech (POS) tagging each sentence of an opinionated text, and on the use of lexical resources to better generate the corresponding subjective extract. We take advantage of WordNet and SentiWordNet, two publically available lexical resources, to calculate the association degrees between sentences of an opinionated document in the subjectivity detection step. We use well-known movie review datasets (from the Internet Movie Database) to provide comparative experiments and we show a statistically significant increase in classification accuracy of the resulting opinion mining system that can be up to 9%. |
|
|
|
|