Sbert

Pre-process

Interlude

This is where they need to get embedding with jupyter

Post-wrangle

We average across vectors for what each person said each rep.

Analysis

Convergence within groups

look within a game, across all reps and playerIDs

0 is not same speaker, 1 is same speaker.

We’re not really seeing much, similarity is higher in rounds that are closer together, but if we only look at relatively close reps (within 5 of each other), there’s not much by way of pattern.

Unclear what interactions we should be including.

repDistance: reps that are further apart from each other are less similar (note this is confounded w/ earlier rep)

being said by the same person increases similarity

Divergence between groups

Not seeing anything useful here.

There’s more difference the further apart the reps are, and maybe slight differences between the game types, but we’re not seeing the clear divergence patterns that arise in say tangrams. (Which makes sense if groups are mostly converging on a limited number of patterns we wouldn’t expect difference between them.)