davidpomerenke
commented
6 months ago First solution could be: Just sort the articles by date. Those closest to the protest date are likely to be most relevant.
Later we could use GPT3.5 or so for determining the protests that an article relates to. Compare #47. Especially if we already run a sentiment model over each article anyway, we can as well also ask it which protests the article is about. From experience I would expect GPT3.5 to be okay at this, finetuned smaller models might not be so good at it.
kleinlennart
We want to make a table with articles about a given protest.
The first step for this will be filtering by date (starting from the protest date and ending x days later) and whether the name of the organization occurs in an article.
Yet this will give false positives: Articles about other protests or that do not relate to any protest at all. Or that relate to other protests by the same group but in another country, as mentioned in one user interview.