Discrete Choice

We adapt graph learning methods to incorporate network structure into discrete choice models.

We introduce a broad class of context effects in discrete choice and show that these effects are common in choice data, including choices coming from social networks.

We show how confounding can mislead choice models and adapt several causal inference methods to train unbiased models from observational data.

We study the problem of finding the optimal set of alternatives to introduce to a group to minimize their disagreement and other related problems.

Choosing is one of the most common and important actions people take: we choose where to work, how to get there, who to vote for, and what to have for lunch. These scenarios are studied in economics, psychology, and (recently) computer science under the umbrella of “discrete choice” (as opposed to continuous choices, like how much milk to put in your coffee). In a discrete choice setting, we’re presented with a set of options (the choice set) and we make a selection from the available items.