Commit
•
3fdd67d
1
Parent(s):
eb31495
Update README.md
Browse files
README.md
CHANGED
@@ -3,8 +3,23 @@ license: mit
|
|
3 |
tags:
|
4 |
- feature-extraction
|
5 |
library_name: generic
|
|
|
|
|
|
|
|
|
6 |
---
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
Usage
|
9 |
```
|
10 |
import fasttext.util
|
|
|
3 |
tags:
|
4 |
- feature-extraction
|
5 |
library_name: generic
|
6 |
+
datasets:
|
7 |
+
- ubertext2.0
|
8 |
+
widget:
|
9 |
+
- text: "доброго вечора ми з україни"
|
10 |
---
|
11 |
|
12 |
+
_name_ is pre-trained word vectors for the Ukrainian language, trained with fastText on (yet unreleased) UberText2.0 dataset, released by the [lang-uk](https://lang.org.ua/en/). This model was trained using skipgram in dimension 300, with character n-grams range of 2-5, and 15 negative samples.
|
13 |
+
|
14 |
+
Our model increases Accuracy by 6.3% compared to the [Facebook Ukrainian word vectors](https://fasttext.cc/docs/en/crawl-vectors.html) on the word analogy task. The dataset for Ukrainian word analogy is available [here](https://github.com/lang-uk/vecs/).
|
15 |
+
|
16 |
+
Extrinsic evaluations were performed on two sequence labeling tasks: NER and POS tagging. NER-UK dataset was released by the lang-uk, and Ukrainian (UD) corpus was developed by a non-profit organization Institute for Ukrainian.
|
17 |
+
|
18 |
+
Results:
|
19 |
+
1) spaCy NER F-score 0.818
|
20 |
+
2) POS Flair Accuracy 0.824
|
21 |
+
3) POS spaCy Accuracy 0.911
|
22 |
+
|
23 |
Usage
|
24 |
```
|
25 |
import fasttext.util
|