Inducing Relational Knowledge from BERT

Authors: Zied Bouraoui, Jose Camacho-Collados, Steven Schockaert7456-7463

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we experimentally analyze the performance of our method. Our main question of interest is whether the proposed method allows us to model relations in a better way than is possible with pre-trained word vectors.
Researcher Affiliation Academia Zied Bouraoui CRIL CNRS & Univ Artois, France zied.bouraoui@cril.fr Jose Camacho-Collados Cardiff University, UK camachocolladosj@cardiff.ac.uk Steven Schockaert Cardiff University, UK schockaerts1@cardiff.ac.uk
Pseudocode No The paper describes its methodology in prose but does not include any pseudocode or algorithm blocks.
Open Source Code No The paper states, "We used the BERT implementation available at https://github. com/huggingface/transformers" (Footnote 5). This refers to a third-party library used, not the authors' own source code for their proposed method.
Open Datasets Yes We consider relations taken from the following three standard benchmark datasets: the Google analogy Test Set (Google)... the Bigger Analogy Test Set (BATS)... the Diff Vec Test Set (DV)...
Dataset Splits No To this end, for a given relation, we first split the set of available examples in two sets: a training set that contains 90% of words pairs and a test set that contains the remaining 10%. The paper does not mention a separate validation set split.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU models, CPU types, memory) used to run the experiments.
Software Dependencies No The paper mentions using a "BERT implementation" from Hugging Face but does not provide specific version numbers for this or any other software dependencies (e.g., Python version, PyTorch version).
Experiment Setup No While the paper describes the general experimental process (e.g., filtering templates, fine-tuning BERT, negative sampling), it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs) or detailed system-level training settings.