Embedding Semantic Relations into Word Representations
Authors: Danushka Bollegala, Takanori Maehara, Ken-ichi Kawarabayashi
IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our proposed method statistically significantly outperforms the current state-of-the-art word representations on three benchmark datasets for proportional analogy detection, demonstrating its ability to accurately capture the semantic relations among words. |
| Researcher Affiliation | Academia | Danushka Bollegala Takanori Maehara Ken-ichi Kawarabayashi University of Liverpool Shizuoka University National Institute of Informatics JST, ERATO, Kawarabayashi Large Graph Project. |
| Pseudocode | Yes | Algorithm 1 Learning word representations. ... The pseudo code for the proposed method is shown in Algorithm 1. |
| Open Source Code | No | We use the publicly available implementations2,3 by the original authors for training the word representations using the recommended parameter values. 2https://code.google.com/p/word2vec/ 3http://nlp.stanford.edu/projects/glove/. These links are for comparative methods, not the authors' proposed method. |
| Open Datasets | Yes | We use the uk Wa C corpus1 to extract relationally similar (positive) and dissimilar (negative) pairs of patterns (pi, pj) to train the proposed method. The uk Wa C is a 2 billion word corpus constructed from the Web limiting the crawl to the .uk domain. 1http://wacky.sslmit.unibo.it |
| Dataset Splits | No | The paper mentions 'The total number of training instances we select is N = 50,000 + 50,000 = 100,000.' and uses 'test' datasets, but it does not explicitly define a separate 'validation' split or its size/proportion. |
| Hardware Specification | No | The paper does not provide specific details regarding the hardware used for running the experiments (e.g., GPU/CPU models, memory specifications). |
| Software Dependencies | No | The paper mentions using 'publicly available implementations' for baselines, but does not provide specific version numbers for any software dependencies required to replicate their own work or the baselines. |
| Experiment Setup | Yes | All methods compared in Table 1 are trained on the same uk Wa C corpus of 2B tokens to produce 300 dimensional word vectors. ... In all of our experiments, the proposed method converged with less than 5 iterations. |