Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Modeling Heterogeneous Hierarchies with Relation-specific Hyperbolic Cones

Authors: Yushi Bai, Zhitao Ying, Hongyu Ren, Jure Leskovec

NeurIPS 2021 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on standard knowledge graph benchmarks show that Con E obtains state-of-the-art performance on hierarchical reasoning tasks as well as knowledge graph completion task on hierarchical graphs. (Abstract) and Given a KG containing many hierarchical and non-hierarchical relations, our experiments evaluate: (A) Performance of Con E on hierarchical reasoning task of predicting if entity h1 is an ancestor of entity h2. (B) Performance of Con E on generic KG completion tasks. (Section 4)
Researcher Affiliation Academia Yushi Bai EMAIL Tsinghua University; Rex Ying EMAIL Stanford University; Hongyu Ren EMAIL Stanford University; Jure Leskovec EMAIL Stanford University
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code of our paper is available at http://snap.stanford.edu/cone. (Section 4)
Open Datasets Yes We use four knowledge graph benchmarks (Table 1): Word Net lexical knowledge graph (WN18RR [5, 17]), drug knowledge graph (DDB14 [18]), and a KG capturing common knowledge (FB15k-237 [31]). Furthermore, we also curated a new biomedical knowledge graph GO21, which models genes and the hierarchy of biological processes they participate in. (Section 4) and Table 1 includes a 'training' column for dataset statistics.
Dataset Splits Yes Table 1: Datasets statistics. #training #validation #test (Table 1 clearly provides a 'validation' column with sample counts for each dataset).
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper mentions using 'Adam [32] as the optimizer' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes During training, we use Adam [32] as the optimizer and search hyperparameters including batch size, embedding dimension, learning rate, angle loss weight and dimension of subspace for each hierarchical relation. (Section 4)