Model Details

• The SENTIMENTAL classifier trained to predict the likelihood that a comment will be perceived as positive or negative.
• BERT based Text Classification.

Intended Use

• Intended to be used for a wide range of use cases such as supporting human moderation and extracting polarity of review comments.
• Not intended for fully automated moderation.
• Not intended to make judgments about specific individuals.

Factors

• Identity terms referencing frequently positive and negative emotions.

Metrics

• Accuracy, which measures the percentage of True Positive and True Negative.

Ethical Considerations

• TODO

Quantitative Analyses

• TODO

Training Data

• TODO

Evaluation Data

• TODO

Caveats and Recommendations

• TODO