Add Legal Opinion Sentiment Classification scenario

[yifanmai]

This is a legal opinion sentiment classification task based on the paper Effective Approach to Develop a Sentiment Annotator For Legal Domain in a Low Resource Setting (Ratnayaka et al., 2020).

[yifanmai]

I'm skeptical about the validity of some of the examples.

Example negative example:

They allow a person with an Internet connection to "become a town crier with a voice that resonates farther than it could from any soapbox."

It's not clear to me why this is negative.

[yifanmai]

Hi @ryokawajp and @mtake, please review this pull request.

By approving, you acknowledge that:

• You will be listed as a co-author of the merge commit
• You agree to license your work under Apache 2.0

[ryokawajp]

lgtm.
I realized that classification_* metrics do not appear in the web UI. However, those do exist in stats.json. So, I think that issue has nothing to do with this scenario.

[ryokawajp]

Example negative example:

They allow a person with an Internet connection to "become a town crier with a voice that resonates farther than it could from any soapbox."

It's not clear to me why this is negative.

I either do not know why this sample is negative. Most of the messages are highly domain-specific.