Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,107 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
task_categories:
|
| 3 |
+
- feature-extraction
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
## Model Description
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
As part of the ITANONG project's 10 billion-token Tagalog dataset, we have introduced a collection of pre-trained embedding models. These models were trained using the Formal text dataset from the renowned corpus which has been thoroughly detailed in our paper. Details of the embedding models can be seen below:
|
| 11 |
+
|
| 12 |
+
```{=latex}
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
% Please add the following required packages to your document preamble:
|
| 16 |
+
% \usepackage{multirow}
|
| 17 |
+
\begin{table}[h]
|
| 18 |
+
\centering
|
| 19 |
+
\begin{tabular}{|c|c|c|c|}
|
| 20 |
+
\hline
|
| 21 |
+
Embedding Technique & Variant & Model File Format & Embedding Size \\ \hline
|
| 22 |
+
\multirow{24}{*}{Word2Vec} & \multirow{12}{*}{Skipgram} & \multirow{6}{*}{.bin} & 20 \\ \cline{4-4}
|
| 23 |
+
& & & 30 \\ \cline{4-4}
|
| 24 |
+
& & & 50 \\ \cline{4-4}
|
| 25 |
+
& & & 100 \\ \cline{4-4}
|
| 26 |
+
& & & 200 \\ \cline{4-4}
|
| 27 |
+
& & & 300 \\ \cline{3-4}
|
| 28 |
+
& & \multirow{6}{*}{.txt} & 20 \\ \cline{4-4}
|
| 29 |
+
& & & 30 \\ \cline{4-4}
|
| 30 |
+
& & & 50 \\ \cline{4-4}
|
| 31 |
+
& & & 100 \\ \cline{4-4}
|
| 32 |
+
& & & 200 \\ \cline{4-4}
|
| 33 |
+
& & & 300 \\ \cline{2-4}
|
| 34 |
+
& \multirow{12}{*}{CBOW} & \multirow{6}{*}{.bin} & 20 \\ \cline{4-4}
|
| 35 |
+
& & & 30 \\ \cline{4-4}
|
| 36 |
+
& & & 50 \\ \cline{4-4}
|
| 37 |
+
& & & 100 \\ \cline{4-4}
|
| 38 |
+
& & & 200 \\ \cline{4-4}
|
| 39 |
+
& & & 300 \\ \cline{3-4}
|
| 40 |
+
& & \multirow{6}{*}{.txt} & 20 \\ \cline{4-4}
|
| 41 |
+
& & & 30 \\ \cline{4-4}
|
| 42 |
+
& & & 50 \\ \cline{4-4}
|
| 43 |
+
& & & 100 \\ \cline{4-4}
|
| 44 |
+
& & & 200 \\ \cline{4-4}
|
| 45 |
+
& & & 300 \\ \hline
|
| 46 |
+
\multirow{24}{*}{FastText} & \multirow{12}{*}{Skipgram} & \multirow{6}{*}{.bin} & 20 \\ \cline{4-4}
|
| 47 |
+
& & & 30 \\ \cline{4-4}
|
| 48 |
+
& & & 50 \\ \cline{4-4}
|
| 49 |
+
& & & 100 \\ \cline{4-4}
|
| 50 |
+
& & & 200 \\ \cline{4-4}
|
| 51 |
+
& & & 300 \\ \cline{3-4}
|
| 52 |
+
& & \multirow{6}{*}{.txt} & 20 \\ \cline{4-4}
|
| 53 |
+
& & & 30 \\ \cline{4-4}
|
| 54 |
+
& & & 50 \\ \cline{4-4}
|
| 55 |
+
& & & 100 \\ \cline{4-4}
|
| 56 |
+
& & & 200 \\ \cline{4-4}
|
| 57 |
+
& & & 300 \\ \cline{2-4}
|
| 58 |
+
& \multirow{12}{*}{CBOW} & \multirow{6}{*}{.bin} & 20 \\ \cline{4-4}
|
| 59 |
+
& & & 30 \\ \cline{4-4}
|
| 60 |
+
& & & 50 \\ \cline{4-4}
|
| 61 |
+
& & & 100 \\ \cline{4-4}
|
| 62 |
+
& & & 200 \\ \cline{4-4}
|
| 63 |
+
& & & 300 \\ \cline{3-4}
|
| 64 |
+
& & \multirow{6}{*}{.txt} & 20 \\ \cline{4-4}
|
| 65 |
+
& & & 30 \\ \cline{4-4}
|
| 66 |
+
& & & 50 \\ \cline{4-4}
|
| 67 |
+
& & & 100 \\ \cline{4-4}
|
| 68 |
+
& & & 200 \\ \cline{4-4}
|
| 69 |
+
& & & 300 \\ \hline
|
| 70 |
+
\end{tabular}
|
| 71 |
+
\end{table}
|
| 72 |
+
|
| 73 |
+
```
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
## Training Details
|
| 78 |
+
This model was trained using an Nvidia V100-32GB GPU on DOST-ASTI Computing and Archiving Research Environment (COARE) - https://asti.dost.gov.ph/projects/coare/
|
| 79 |
+
|
| 80 |
+
### Training Data
|
| 81 |
+
The training dataset was compiled from both formal and informal sources, consisting of 194,001 instances from formal channels. More information on pre-processing and training parameters on our paper.
|
| 82 |
+
|
| 83 |
+
## Citation
|
| 84 |
+
Paper : iTANONG-DS : A Collection of Benchmark Datasets for Downstream Natural Language Processing Tasks on Select Philippine Language
|
| 85 |
+
|
| 86 |
+
|
| 87 |
+
Bibtex:
|
| 88 |
+
```
|
| 89 |
+
@inproceedings{visperas-etal-2023-itanong,
|
| 90 |
+
title = "i{TANONG}-{DS} : A Collection of Benchmark Datasets for Downstream Natural Language Processing Tasks on Select {P}hilippine Languages",
|
| 91 |
+
author = "Visperas, Moses L. and
|
| 92 |
+
Borjal, Christalline Joie and
|
| 93 |
+
Adoptante, Aunhel John M and
|
| 94 |
+
Abacial, Danielle Shine R. and
|
| 95 |
+
Decano, Ma. Miciella and
|
| 96 |
+
Peramo, Elmer C",
|
| 97 |
+
editor = "Abbas, Mourad and
|
| 98 |
+
Freihat, Abed Alhakim",
|
| 99 |
+
booktitle = "Proceedings of the 6th International Conference on Natural Language and Speech Processing (ICNLSP 2023)",
|
| 100 |
+
month = dec,
|
| 101 |
+
year = "2023",
|
| 102 |
+
address = "Online",
|
| 103 |
+
publisher = "Association for Computational Linguistics",
|
| 104 |
+
url = "https://aclanthology.org/2023.icnlsp-1.34",
|
| 105 |
+
pages = "316--323",
|
| 106 |
+
}
|
| 107 |
+
```
|