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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
NeuSymEA: Neuro-symbolic Entity Alignment via Variational Inference
Authors: Shengyuan Chen, Zheng Yuan, Qinggang Zhang, Wen Hua, Jiannong Cao, Xiao Huang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments 5.1 Experimental settings Datasets. Main experiments use the DBP15K dataset, comprising three cross-lingual KG pairs: JA-EN, FR-EN, and ZH-EN. ... Baselines and metrics. Baseline models include seven neural models GCNAlign [19], Align E, Boot EA [16], RREA [17], Dual-AMN [18], Light EA [20], PEEA [31], one symbolic models PARIS [12], and two neuro-symbolic models PRASE [14], EMEA [24]. We use Hit@1, Hit@10, and MRR as the evaluation metrics. |
| Researcher Affiliation | Academia | Shengyuan Chen Department of Computing The Hong Kong Polytechnic University Hung Hom, Hong Kong SAR EMAIL Zheng Yuan Department of Computing The Hong Kong Polytechnic University Hung Hom, Hong Kong SAR EMAIL Qinggang Zhang Department of Computing The Hong Kong Polytechnic University Hung Hom, Hong Kong SAR EMAIL Wen Hua Department of Computing The Hong Kong Polytechnic University Hung Hom, Hong Kong SAR EMAIL Jiannong Cao Department of Computing The Hong Kong Polytechnic University Hung Hom, Hong Kong SAR EMAIL Xiao Huang Department of Computing The Hong Kong Polytechnic University Hung Hom, Hong Kong SAR EMAIL |
| Pseudocode | Yes | A.3 Pseudo-code of Explainer Below is the pseudo-code of how the explainer generates supporting rules as interpretations for the query pair. It consists of two stages: searching reachable anchor pairs, and parsing rule paths as well as calculating rule confidences. Algorithm 1 Generating Interpretations for the Queried Entity Pair with Weighted Rules |
| Open Source Code | Yes | Codes are released at https://github.com/chensyCN/NeuSymEA-NeurIPS25. |
| Open Datasets | Yes | Datasets. Main experiments use the DBP15K dataset, comprising three cross-lingual KG pairs: JA-EN, FR-EN, and ZH-EN. ... For additional experiments on large KGs, we employ Open EA [30] and DBP1M. ... Dataset statistics are provided in Appendix B.1. |
| Dataset Splits | Yes | Two different dataset split strategies are used in the EA literature: ❶a 3:7 train/test split, and ❷a 5-fold cross-validation scheme with a 2:1:7 ratio for training, validation, and test sets, as used in Open EA [30]. We adopt the latter for all algorithms to ensure fair comparison. |
| Hardware Specification | Yes | Table 8: Machine configuration. Component Specification GPU NVIDIA GeForce RTX 3090 CPU Intel(R) Xeon(R) Silver 4214R CPU @ 2.40GHz |
| Software Dependencies | No | The paper states that codes are provided in supplementary material with a readme file for guidance, but the main text or appendices do not explicitly list specific software dependencies with version numbers. |
| Experiment Setup | Yes | Hyperparameters. Neu Sym EA involves two main hyperparameters: the number of EM iterations and the symbolic model s threshold δ for selecting positive pairs. We tune them on the validation set, searching δ in 0.6, 0.7, 0.8, 0.9, 0.95, 0.98, 0.99 and the number of iterations from 1 to 9. |