Simultaneous Clustering and Ranking from Pairwise Comparisons

Authors: Jiyi Li, Yukino Baba, Hisashi Kashima

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments using several real datasets collected using crowdsourcing. These datasets include the examples of the idea and design collections that we aim to make decisions on them. The experimental results illustrate that our approach can generate better neighbor and preference estimation results than the approaches that only focus on a single type of pairwise comparisons, by only increasing a small number of cost on collecting additional labels.
Researcher Affiliation Academia Jiyi Li1,4, Yukino Baba2,4, Hisashi Kashima3,4 1 University of Yamanashi, Japan 2 University of Tsukuba, Japan 3 Kyoto University, Japan 4 RIKEN Center for AIP, Japan
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the methodology described.
Open Datasets Yes We select six dog breeds and five images for each breed (30 images in total) from the Stanford Dogs Dataset [Khosla et al., 2011].
Dataset Splits Yes In one experimental trial, we randomly select a subset of all object pairs with sampling rate r = 0.1. For both two types of pairwise comparison, we only use five labels in the ten labels of each object pair. We denote the subset of labels as Pk and Sk, where k is the index of the subset. We evaluate the average performance of ten trials." and "We use the label subsets Pk and Sk which are the same ones used by our approach to learning the embeddings and preference criterion vector as the surrogate ground truth. We tune the parameters by the measures of the pairwise preference accuracy on the held out subset of Pk and the pairwise similarity accuracy on the held out subset of Sk.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or detailed computer specifications used for running the experiments.
Software Dependencies No The paper does not provide specific software dependency details, such as library names with version numbers, needed to replicate the experiment.
Experiment Setup Yes The dimension of embeddings d is set to 10. The regularization terms are set to η = 0.1 and γ = 0.1. ... The value of α is chosen from {0.25, 0.5, 1, 2, 4}; the value of β is chosen from {0.25, 0.5, 1, 2, 4}.