Enabling Abductive Learning to Exploit Knowledge Graph

Authors: Yu-Xuan Huang, Zequn Sun, Guangyao Li, Xiaobin Tian, Wang-Zhou Dai, Wei Hu, Yuan Jiang, Zhi-Hua Zhou

IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on four different tasks show that ABL-KG can automatically extract useful rules from large-scale and noisy knowledge graphs, and significantly improve the performance of machine learning with only a handful of labeled data.
Researcher Affiliation Academia Yu-Xuan Huang , Zequn Sun , Guangyao Li , Xiaobin Tian , Wang-Zhou Dai , Wei Hu , Yuan Jiang and Zhi-Hua Zhou National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China {huangyx, daiwz, jiangy, zhouzh}@lamda.nju.edu.cn, {zqsun, gyli, xbtian}.nju@gmail.com, whu@nju.edu.cn
Pseudocode Yes Algorithm 1 Abductive Learning with Knowledge Graph
Open Source Code Yes The code is available for download1. 1https://github.com/AbductiveLearning/ABL-KG
Open Datasets Yes The zoo animal classification dataset [Dua and Graff, 2017] contains animals attributes (e.g., backbone, legs) and their categories (e.g., bird, fish), along with the names of each attribute and class (task specification). We consider the widely-used cross-lingual dataset DBP-EN-FR, which was proposed in the Open EA benchmark study [Sun et al., 2020]. We consider the benchmark dataset FB15K-237 [Toutanova and Chen, 2015] and choose the most popular model, Trans E [Bordes et al., 2013], as the basic learning model in the experiment. The task s input is images from ADE20K [Zhou et al., 2017].
Dataset Splits Yes We follow the training/validation/test data splits of FB15K-237, and report the average test results of five runs in Table 4.
Hardware Specification No No specific hardware details (GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments were found.
Software Dependencies No The paper mentions software like GloVe, Random Forest, Align E, Trans E, ResNet-50, but does not provide specific version numbers for any of them.
Experiment Setup Yes Rules are mined using our algorithm in Section 4.3, where d is set to 2, and we set the confidence of Is A relation to be 1.0 and others 0.9. The entity alignment model learns embeddings for the two KGs to measure entity similarity. In our experiment, we consider the widely-used cross-lingual dataset DBP-EN-FR, which was proposed in the Open EA benchmark study [Sun et al., 2020]. Align E+ employs self-training [Yarowsky, 1995] and selects the predicted entity alignment pairs whose embedding similarity is greater than 0.9 to augment training data. In our task, only 5% or 10% of all 20k images are labeled.