Diverse Weighted Bipartite b-Matching

Authors: Faez Ahmed, John P. Dickerson, Mark Fuge

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we demonstrate the efficacy of our methods on three real-world datasets, and show that the price of diversity is not bad in practice. Our code is publicly accessible for further research.
Researcher Affiliation Academia Faez Ahmed Dept. of Mechanical Engg. University of Maryland faez00@umd.edu John P. Dickerson Dept. of Computer Sci. University of Maryland john@cs.umd.edu Mark Fuge Dept. of Mechanical Engg. University of Maryland fuge@umd.edu
Pseudocode Yes Algorithm 1: GD-WBM Greedy Diverse Matching
Open Source Code Yes Our code is publicly accessible for further research.1 1https://github.com/faezahmed/diverse matching
Open Datasets Yes We use a subset of the Movie Lens 1M dataset [Harper and Konstan, 2016]... We use the multi-aspect review assignment evaluation dataset [Karimzadehgan and Zhai, 2009]... We use the Scholarly Paper Recommendation dataset provided by Sugiyama et al. [Sugiyama and Kan, 2010]
Dataset Splits No The paper uses various datasets for evaluation (MovieLens, UIUC, Sugiyama) but does not specify explicit training, validation, or test splits by percentages or sample counts.
Hardware Specification No The paper does not provide specific details about the hardware used to run experiments, such as CPU or GPU models, or memory specifications.
Software Dependencies Yes We solve it using two different approaches, first using Gurobi s Mixed Integer Quadratic Programming (MIQP) Solver [Gurobi Optimization, 2016]
Experiment Setup Yes We set the constraints such that each paper matches with at least 3 reviewers and every reviewer is allocated at least 1 paper, while no reviewer is allocated more than 10 papers. We select all papers from KDD 2000 to KDD 2010 from this dataset (a total of 1184 papers) and find three reviewers for each of them from the given set of 50 reviewers. We calculate edge weights between papers and reviewers as the cosine distance between the tf-idf vector of the query paper and reviewer s latest paper. No limit of maximum number of papers that a reviewer can review is imposed but each reviewer must be allocated at least one paper. In all our simulations we terminate D-WBM after 1 hour and take the best solution, while WBM converges to true optima.