Improving Hypernymy Prediction via Taxonomy Enhanced Adversarial Learning
Authors: Chengyu Wang, Xiaofeng He, Aoying Zhou7128-7135
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments to confirm the effectiveness of TEAL over three standard NLP tasks: unsupervised hypernymy classification, supervised hypernymy detection and graded lexical entailment. We also show that TEAL can be applied to non-English languages and can detect missing hypernymy relations in taxonomies. |
| Researcher Affiliation | Academia | Chengyu Wang,1 Xiaofeng He,1 Aoying Zhou2 1School of Computer Science and Software Engineering, East China Normal University 2School of Data Science and Engineering, East China Normal University chywang2013@gmail.com, xfhe@sei.ecnu.edu.cn, ayzhou@dase.ecnu.edu.cn |
| Pseudocode | Yes | Algorithm 1 Semantic Filtering Algorithm for AS-TEAL and Algorithm 2 Adversarial Model for Hypernymy Prediction |
| Open Source Code | No | The paper does not provide any specific links to source code or explicit statements about releasing the code for the methodology described. |
| Open Datasets | Yes | The taxonomy we use is Microsoft Concept Graph4, a large public dataset generated from Probase (Wu et al. 2012). It contains 33,377,320 hypermymy relations... To show TEAL does not require the training of task-specific word embeddings, we employ the Glovec model trained over the Wikipedia and Gigaword corpus (Pennington, Socher, and Manning 2014). |
| Dataset Splits | Yes | We learn a threshold τ ( 1, 1) over s(x, y) to distinguish the two type of relations using the same experimental settings as in the previous study (Nguyen et al. 2017) where there is a 98%:2% split between validation and test sets. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions 'Word2Vec', 'Glove', and 'Adam optimization algorithm' as techniques used, but does not provide specific version numbers for any software libraries or dependencies, nor for the programming language used. |
| Experiment Setup | Yes | For simplicity, we employ only one fully-connected 100 dimensional layer as the hidden layer, with hyperbolic tangent (tanh) as the activation function. The model parameters are learned using the Adam optimization algorithm (Kingma and Ba 2014) in 500 epochs. The batch size is set to 64. |