Ensemble Learning for Relational Data

Authors: Hoda Eldardiry, Jennifer Neville, Ryan A. Rossi

JMLR 2020 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the ensemble method on both synthetic and real world data. Furthermore, we demonstrate the effectiveness of the proposed methods for two different network settings: (i) the single graph setting (Section 5.1) and (ii) the multi-graph setting where there are multiple graphs available with different link types (Section 5.2).
Researcher Affiliation Collaboration Hoda Eldardiry EMAIL Department of Computer Science Virginia Tech... Jennifer Neville EMAIL Department of Computer Science and Department of Statistics Purdue University... Ryan A. Rossi EMAIL Adobe Research...
Pseudocode Yes Algorithm 1 Ensemble Learning: EL(Gtr =(Vtr, Etr), m)... Algorithm 2 Relational Subgraph Resampling: RSR(G = (V, E), b)... Algorithm 3 Collective Ensemble Classification (CEC)
Open Source Code No The paper does not contain any explicit statement about providing open-source code for the methodology described, nor does it provide a link to a code repository.
Open Datasets Yes Synthetic data sets are generated with a latent group model Neville and Jensen (2005)... The Facebook dataset used in this work is a sample of Purdue University Facebook network... The second data set is from IMDb (Internet Movie Database)...
Dataset Splits Yes The 10 trials are repeated for 4 training and test pairs... The robustness of the methods to missing labels (in the test set) is evaluated by varying the proportion of labeled test data at 10% through 90%.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or specific computer configurations) used for running its experiments.
Software Dependencies No For the experiments in Section 5, we use relational dependency network (RDN) Neville and Jensen (2007) models as the component collective classification models... No specific software versions are mentioned for RDNs or any other libraries.
Experiment Setup Yes The RSR algorithm uses a subgraph size b = 50 and b = 10 for the synthetic and Facebook experiment, respectively... using 450 500 Gibbs iterations for collective inference... we construct 5 bootstrap pseudosamples and learn the ensemble models (i.e., m = 5).