Analysis and Evaluation of Feature Selectors

J. Isabella / Indian Journal of Computer Science and Engineering (IJCSE)

Analysis and evaluation of Feature selectors
in opinion mining
Research Scholar, Sathyabama university,

Jerusalem.Engineering College,
Computational performance is improved by use of basic feature selection in most of the research works.
Sentiment analysis identifies whether opinion expressed on a topic in a document is positive or negative.
But many potential sentiment analysis applications are not feasible because of the voluminous amount of
features present in the corpus. This paper evaluates a range of feature selectors systematically with
respect to their efficiency in improving the performance of the classifiers for sentiment analysis. Movie
reviews are used for sentiment analysis in this study.
KEYWORDS: Sentiment analysis, Movie reviews, Feature Selection, Correlation based feature selector
(CFS), Information Gain, Support Vector Machine (SVM), Principal component analysis (PCA)
Machine learning studies algorithms which automatically improves performance with experience. Prediction is
the main outcome of performance. An algorithm is said to have learned when it improves its ability to predict
key elements of a task when presented with proper data. Machine learning algorithms are characterized by
languages which represent knowledge. Research reveals that no single learning approach is superior and usually
different learning algorithms produce the same results [1]. A factor which has much impact on learning
algorithm’s success is the nature of data that characterizes the task to be learned. Learning fails in machine
learning algorithms when data does not exhibit statistical regularity. While new data can be constructed from old
to exhibit statistical regularity and facilitate learning, it is a complex task that a fully automatic method is