[
https://issues.apache.org/jira/browse/OPENNLP-840?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15192461#comment-15192461
]
Shivam Bharuka commented on OPENNLP-840:
----------------------------------------
What are the 7 classes in which we classify tweets? While binary classification
is straight-forward but multi class will require working with appropriate
features. Some of the things for the algorithm could be removing stopwords and
finding keywords. SVM kernels can be used for emotion classification.
> Sentiment Analysis
> ------------------
>
> Key: OPENNLP-840
> URL: https://issues.apache.org/jira/browse/OPENNLP-840
> Project: OpenNLP
> Issue Type: New Feature
> Reporter: Mondher Bouazizi
> Labels: gsoc, gsoc2016, nlp
>
> The objective of the "Sentiment Analysis" component is to determine the
> sentiment of the author towards the object of his text.
> Different techniques are proposed in the academic literature, and some state
> of the art approaches present very high accuracy.
> Sentiment analysis can have different granularity levels:
> - Binary classification: in this case, the text is to be classified into two
> classes which are "positive" and "negative".
> - Ternary classification: in addition to the two classes present in the
> binary classification, a third class is added which is "neutral".
> - Multi-class sentiment analysis: the two classes "positive" and "negative"
> are further divided into sub-classes (e.g., "love" happiness", etc. for the
> positive class; and "hate", "anger", etc. for the negative class). Therefore
> the classification objective is to determine the sentiment sub-class instead
> of the main polarity
> In this component, we will implement some of the state of the art approaches,
> in particular the one presented here[1]. approaches use machine-learning
> techniques to learn a classifier from labeled training sets.
> -----------------------------------------------
> [1] http://www.ieice.org/ken/paper/20160129DbfF/eng/
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
