3 ms·
Details for the Science4Cast Competition
- webmaven 5y agoThis looks like an interesting competition, aimed at predicting the connections between future research papers. The dataset is only about 37MB, so barriers to participation should be (relatively) low. The biggest difficulty IMO is that the dataset isn't actually representative of the stated task, which is "[T]he prediction of which scientific pairs of concepts will be investigated by scientists over three years." Rather, the dataset excludes all the dead ends, the things that were thought to be worth investigating, but didn't pan out with results worth writing up for publication. So, in reality, the problem is a lot harder: predicting the pairs of concepts that are going to yield results worth writing up (admittedly not a super high bar, but the selection bias still skews the data).