Sentiment analysis refers to the classical use case of classification, where text documents (documents, paragraphs or sentences) are mapped onto two categories: positive or negative.

The approach to sentiment classification supported by PredictiveWorks. implements an algorithm proposed by Vivek Narayanan. It is based on a combination of methods like negation handling, word n-grams and feature selection by mutual information results in a significant improvement in accuracy.

Corpus

The training corpus consists of a set of sentiment tokens assigned to a sentiment label. Token set and assigned label have to be provided line by line as shown below:

positive -> amazing voice acting
negative -> horrible acting
negative -> very bad
positive -> very fantastic

Plugins

SentimentBuilder

@Plugin(type = SparkSink.PLUGIN_TYPE)
@Name("SentimentBuilder")
@Description("A building stage for a Sentiment Analysis model based on the sentiment algorithm "
	+ "introduced by Vivek Narayanan. The training corpus comprises a labeled set of sentiment "
	+ "tokens.")
public class SentimentBuilder extends TextSink {

    ...

}

Parameters

Sentiment

@Plugin(type = SparkCompute.PLUGIN_TYPE)
@Name("Sentiment")
@Description("A transformation stage that predicts sentiment labels (positive or negative) for "
	+ "text documents, leveraging a trained Sentiment Analysis model.")
public class Sentiment extends TextCompute {

    ...

}

Parameters