what features, what predictors?

[ajstent]

Daniel and I met today and discussed what we are trying to predict, what features to use and how to sample the data accordingly: what are we trying to predict? edits alt what are we trying to predict? authors/other reviewers' ratings of reviews (but how many of these do we have?) ok we also have the response, so that might be useful for prediction

what features? argumentation; specificity; politeness? also check out Daniels previous work on social media analysis

do you want to sample only papers with author responses? only papers with author responses and discussions?

[nnkennard]

Some ideas of things to measure:

• Best reviewers (need to link with identities)
• Edits to manuscripts, to ratings, and to reviews
• Author response length
• Richness of discourse; polite v/s violent
• Aspects coverage e.g. Summary, Clarity, Methodology...
• A year when someone rated? Can we rate them (would be a proxy)
• We need to predict something about the author response -- a proxy for the author's perception of the review

Features

• Lightweight features e.g. negative and positive reviews (LIWC)
• Number of lists, number of references...

[ajstent]

look at metareviewers' decisions and author responses for predictors