KEML: A Knowledge-Enriched Meta-Learning Framework for Lexical Relation Classification

Authors: Chengyu Wang, Minghui Qiu, Jun Huang, Xiaofeng He13924-13932

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments over multiple datasets show KEML outperforms state-of-the-art methods.In this section, we conduct extensive experiments to evaluate KEML over multiple benchmark datasets, and compare it with state-of-the-arts to make the convincing conclusion.
Researcher Affiliation Collaboration Chengyu Wang,1, 2 Minghui Qiu,2 Jun Huang,2 Xiaofeng He3 1 Zhejiang Lab 2 Alibaba Group 3 School of Computer Science and Technology, East China Normal University
Pseudocode Yes Algorithm 1 Meta-Learning Algorithm for LRC
Open Source Code No The paper states: 'We use the uncased, base version of BERT. See https://github. com/google-research/bert.' but does not provide a link or statement for the open-sourcing of KEML's code.
Open Datasets Yes We use the five public benchmark datasets in English for multi-way classification of lexical relations to evaluate KEML, namely, K&H+N (Necsulescu et al. 2015), BLESS (Baroni and Lenci 2011), ROOT09 (Santus et al. 2016b), EVALution (Santus et al. 2015) and Cog ALex-V Subtask 2 (Santus et al. 2016a).
Dataset Splits Yes K&H+N, BLESS, ROOT09 and EVALution are partitioned into training, validation and testing sets, following the exact same settings as in Shwartz and Dagan (2016). The Cog ALex-V dataset has training and testing sets only, with no validation sets provided (Santus et al. 2016a). Hence, we randomly sample 80% of the training set to train the model, and use the rest for tuning.
Hardware Specification Yes The algorithms are implemented with Tensor Flow and trained with NVIDIA Tesla P100 GPU.
Software Dependencies No The paper states 'The algorithms are implemented with Tensor Flow' but does not provide specific version numbers for TensorFlow or other software dependencies.
Experiment Setup Yes The default hyper-parameter settings of KEML are as follows: N = |R| 1, γ = 1 and α = ϵ = 10 3. We use tanh as the activation function, and Adam as the optimizer to train the neural network. All the model parameters are l2regularized, with the hyper-parameter λ = 10 3. The batch size is set as 256. The dimension of hidden layers is set as the same of d (768 for the base BERT model).