Milton E. Terry's research while affiliated with Connecticut Agricultural Experiment Station and other places

Citations

... Since deep learning is more suitable for continuous features, we convert ordinal and categorical features into continuous features. For the ordinal feature like the ranking position p, we transform it into a continuous one with a Bradly-Terry model [5,28,40,41] as , = + , where , is the probability that the ranking position is lower than the position . is the transformed continuous score, where a larger score represents a higher rank. ...
... comparison. The most fundamental distribution over pairwise comparisons is the BTL model [BT52, Luc59] where the probability that the learner observes = when the items , are compared is ( )/( ( ) + ( )). The learner's goal is mainly to estimate the underlying distribution using a small number of pairwise comparisons. ...