[Tests] Create testset

[JasonLo]

Problem:

• Some demo output is misbehaving?

To-dos:

• Create testset from hackathon files
• Populate ideal answers
• Decide metric for testset: top-5 accuracy of hitting the relevant document object?

[JasonLo]

Asking for ideal answer source if available.

[JasonLo]

Metrics

Consideration: Does the QA interface reduce the manual browsing time for a search? It's assumed that browsing time decreases when the target results appear higher in the rank.

Target metric: Hit rate on the manually labeled xDD object(s) in top1/top5/top10 results.

[iross]

Before we go too far down the road of evaluation, I definitely need to re-visit the database population process and make sure that all the relevant documents have been embedded and stored in weaviate. I'm concerned that the snapshot I pushed in doesn't include everything, and the most likely things to be missing are the single documents we manually brought into the system (as opposed to the bulk acquisition that forms the vast majority of the corpus).