Sentence Polarity Annotation

[dafajon]

• I realized while annotating POSITIVE, NEGATIVE labels for product_review corpus, there are distinct sentences that carry positive intent and other with negative intent and some with neutral.
• I think we should be able or provide the ability to build sentence polarity annotator (continous or categorical target value) to build a less noisy sentiment/polarity dataset in any domain and train a model with it that scores sentences.
• This would be a first step to increase granularity in sentiment/polarity detection of a document. Document sentiment score would be calculated from sum or average of scores of fine granular entities like in Google Cloud NLP services.
• We can start off with something basic such as pigeon to prepare a sentence polarity corpus using existing datasets.

[inancdokurel]

How can I be of assistance here? From what I understand, the requested feature is similar to a modified version of the Sadedegel Annotator. Is the issue, in core, about creating an annotation engine for corpora, or embedding this functionality inside sadedegel library itself?

[dafajon]

A separate annotator as a modified version of the previous one.

• To begin with only we will use it.
• In the future I envision it as a part of the annotator in SadedeGel-as-a-service.