Unifying Knowledge Base Completion with PU Learning to Mitigate the Observation Bias

Authors: Jonas Schouterden, Jessa Bekker, Jesse Davis, Hendrik Blockeel4137-4145

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We aim to empirically answer the following research questions: can we effectively account for observation biases (i.e., obtain more accurate confidence estimates) using the newly proposed propensity-based estimators, (Q1) when the propensities are known, (Q2) when propensities are guessed ( noisy propensities), (Q3) even when the PCA assumption holds?
Researcher Affiliation Academia KU Leuven, Department of Computer Science, B-3000 Leuven, Belgium Leuven.AI KU Leuven Institute for AI, B-3000 Leuven, Belgium {jonas.schouterden, jessa.bekker, jesse.davis, hendrik.blockeel}@kuleuven.be
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Our source code is publicly available.5 5https://github.com/ML-KULeuven/KBC-as-PU-Learning
Open Datasets Yes Our I is the popular KBC benchmark dataset Yago3-10 (Mahdisoltani, Biega, and Suchanek 2015).
Dataset Splits No The paper does not specify explicit training, validation, and test dataset splits with percentages or sample counts to reproduce the data partitioning.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running the experiments.
Software Dependencies No The paper mentions that 'Rules predicting any p P are mined from I with AMIE (Galárraga et al. 2015) with its default settings', but does not provide specific version numbers for software dependencies like programming languages or libraries.
Experiment Setup Yes Rules predicting any p P are mined from I with AMIE (Galárraga et al. 2015) with its default settings and a minimum CWA(R) 0.1. [...] Propensity scores are required to calculate IPW(-PCA)(R). We use correct propensity scores e( ) for the idealized scenarios in Q1 and Q3 and noisy versions ˆe for Q2. More details about the exact setup can be found in Appendix C.