Modeling Interactive Information Retrieval as a Stochastic Process
Norbert Fuhr · 2017
Stochastic models have a long history in IR, starting from probabilistic models for (non-interactive) IR, and later for evaluation measures like e.g. average precision or discounted cumulated gain. The probabilistic ranking principle defines the order in which a user should inspect the documents retrieved by the system, for maximizing the retrieval quality experienced by the user. In a similar way, for interactive IR, the interactive probabilistic ranking principle (IPRP) defines a framework for the optimum ordering of suggestions / possible actions to be considered by the user.