Create a HF model for labeling urls as relevant or irrelevant

[EvilDrPurple]

Context

Part of #12 Now that we've created an experimental model for record type labeling, we want to create another model that simply labels urls based on relevancy.

Requirements

• Model should be able to predict whether a link is relevant or irrelevant with high enough accuracy

[josh-chamberlain]

as another step toward #12 we could use the subcategories from here: https://docs.pdap.io/activities/data-dictionaries/record-types-taxonomy

i.e. Police & public interactions, Info about officers, etc.

We're going to start grouping them this way, so the record_type can have less fidelity.

[maxachis]

@EvilDrPurple when we are talking about relevancy, what counts as relevant and irrelevant, and what are representative samples of each?

[EvilDrPurple]

@EvilDrPurple when we are talking about relevancy, what counts as relevant and irrelevant, and what are representative samples of each?

@maxachis I just uploaded a new dataset to the PDAP Hugging Face page Here. Essentially, a URL is relevant if it has relevant data pertaining to criminal justice, otherwise it is irrelevant. You can look at the record types taxonomy link Josh posted above to see the categories for what types of webpages are considered "relevant"

Hope this helps!

[josh-chamberlain]

yes, "relevant" ≠ "useful". we're just checking whether it contains information about a police, jail, or court agency in the United States.