Learning Embeddings from Knowledge Graphs With Numeric Edge Attributes

Authors: Sumit Pai, Luca Costabello

IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments We assess the predictive power of Focus E on the link prediction task with numeric-enriched triples. Experiments show that Focus E outperforms conventional KGE models and its closest direct competitor UKGE [Chen et al., 2019] in discriminating low-valued triples from high-valued ones.
Researcher Affiliation Industry Sumit Pai , Luca Costabello Accenture Labs {sumit.pai, luca.costabello}@accenture.com
Pseudocode No No structured pseudocode or algorithm blocks were found.
Open Source Code Yes Focus E and all baselines are implemented with the Ampli Graph library [Costabello et al., 2019] version 1.4.0, using Tensor Flow 1.15.2 and Python 3.7. Code and experiments are available at https://github.com/Accenture/Ampli Graph.
Open Datasets Yes We experiment with three publicly available benchmark datasets originally proposed by [Chen et al., 2019]. ... CN15K [Chen et al., 2019]. ... NL27K [Chen et al., 2019]. ... PPI5K [Szklarczyk et al., 2016]. ... O*NET20K3. We introduce a subset of O*NET 4... https://www.onetonline.org/
Dataset Splits Yes Table 1: Datasets used in experiments. (...) validation sets only include high-valued triples where w >= 0.8. and "Validation 138 3532 8161 1940" in Table 1.
Hardware Specification Yes All experiments were run under Ubuntu 16.04 on an Intel Xeon Gold 6142, 64 GB, equipped with a Tesla V100 16GB.
Software Dependencies Yes Focus E and all baselines are implemented with the Ampli Graph library [Costabello et al., 2019] version 1.4.0, using Tensor Flow 1.15.2 and Python 3.7.
Experiment Setup Yes For each baseline and for Focus E, we carried out extensive grid search, over the following ranges of hyperparameter values: embedding dimensionality k = [200 600], with a step of 100; baseline losses={negative log-likelihood, multiclass-NLL, self-adversarial}; synthetic negatives ratio "eta" = {5, 10, 20, 30}; learning rate= {1e 3, 5e 3, 1e 4}; epochs= [100 800], step of 100; L3 regularizer, with weight "gamma" = {1e 1, 1e 2, 1e 3}. For Focus E we also tuned the decay "lambda" = [100 800], with increments of 100.