Achieving Global Justice through Business

Kamishima, Yuko

doi:10.5840/wcp232018511085

Cited by 1 publication

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Comparing Tables 1 and 2, we find that the repeated selection and repeated elimination models generally perform quite comparably for MNL-based and CDM-based models, though RE models tend to perform slightly worse than RS models, except on the sushi dataset. Noting that the sushi data has previously been distributed and analyzed in prior work in the "wrong" order (Kamishima, 2018), it is interesting that the data is in fact slightly more predictable (under RS M N L , which is Plackett-Luce) in the wrong/reversed order than in the correct order. The performance difference between RS P CM C and RE P CM C is much more noticeable, with RE P CM C performing worse than all other RE models.…”

Section: Log-likelihood For Ranking Datamentioning

confidence: 98%

“…For example, the widely-studied sushi dataset that consists of preference rankings over types of sushi was originally reported as having low ranks for items with low priority, but was later corrected so that low ranks represent high priority. Rankings built from these scores prior to the correction were thus "backwards" (see the warning by Kamishima (2018)). And as we've seen in this section, the difference between learning a ranking distribution from forwards vs. backwards ranking data amounts to entirely different families of ranking distributions.…”

Section: Appendixmentioning

confidence: 99%

See 1 more Smart Citation

Choosing to Rank

Ragain¹,

Ugander²

2018

Preprint

View full text Add to dashboard Cite

Ranking data arises in a wide variety of application areas but remains difficult to model, learn from, and predict. Datasets often exhibit multimodality, intransitivity, or incomplete rankingsparticularly when generated by humans-yet popular probabilistic models are often too rigid to capture such complexities. In this work we leverage recent progress on similar challenges in discrete choice modeling to form flexible and tractable choice-based models for ranking data. We study choice representations, maps from rankings (complete or top-k) to collections of choices, as a way of forming ranking models from choice models. We focus on the repeated selection (RS) choice representation, first used to form the Plackett-Luce ranking model from the conditional multinomial logit choice model. We fully characterize, for a prime number of alternatives, the choice representations that admit ranking distributions with unit normalization, a desirably property that greatly simplifies maximum likelihood estimation. We further show that only specific minor variations on repeated selection exhibit this property. Our choice-based ranking models provide higher out-of-sample likelihood when compared to Plackett-Luce and Mallows models on a broad collection of ranking tasks including food preferences, ranked-choice elections, car racing, and search engine relevance tasks.

show abstract

Section: Log-likelihood For Ranking Datamentioning

confidence: 98%

Section: Appendixmentioning

confidence: 99%