identify the predictors (across topics) that lead to transition of topic

[jeremymanning]

between boundaries for each event -- ask whether there are predictors in the prior event(s)

one way of studying this: temporal correlation matrix

could also build explicit predictive models (HMM, Kalman filter, regression, deep learning, etc.)

@paxtonfitzpatrick has a script for getting start and end of each event

[futingzou]

we may start with:

• [ ] calculating the distances between event boundaries and temporal/spatial relationships words

• [ ] finding words appearing in the nearest sentence from each event boundary most frequently