akoksal commited on
Commit
ecfc692
1 Parent(s): 07c88b3

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +52 -0
README.md ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: "tr"
3
+ tags:
4
+ - sentiment
5
+ - twitter
6
+ - turkish
7
+ ---
8
+
9
+ This Turkish Sentiment Analysis model is a fine-tuned checkpoint of pretrained [BERTurk model 128k uncased](https://huggingface.co/dbmdz/bert-base-turkish-128k-uncased) with [BounTi dataset](https://ieeexplore.ieee.org/document/9477814).
10
+ ## Usage in Hugging Face Pipeline
11
+ ```
12
+ from transformers import pipeline
13
+ bounti = pipeline("sentiment-analysis",model="akoksal/bounti")
14
+ print(bounti("Bu yemeği pek sevmedim"))
15
+ >> [{'label': 'negative', 'score': 0.8012508153915405}]
16
+ ```
17
+
18
+ ## Results
19
+ The scores of the finetuned model with BERTurk:
20
+
21
+ ||Accuracy|Precision|Recall|F1|
22
+ |-------------|:---------:|:---------:|:------:|:-----:|
23
+ |Validation|0.745|0.706|0.730|0.715|
24
+ |Test|0.723|0.692|0.729|0.701|
25
+
26
+
27
+
28
+ ## Dataset
29
+ You can find the dataset in [our Github repo](https://github.com/boun-tabi/BounTi-Turkish-Sentiment-Analysis) with the training, validation, and test splits.
30
+
31
+ Due to Twitter copyright, we cannot release the full text of the tweets. We share the tweet IDs, and the full text can be downloaded through official Twitter API.
32
+
33
+ | | Training | Validation | Test |
34
+ |----------|:--------:|:----------:|:----:|
35
+ | Positive | 1691 | 188 | 469 |
36
+ | Neutral | 3034 | 338 | 843 |
37
+ | Negative | 1008 | 113 | 280 |
38
+ | Total | 5733 | 639 | 1592 |
39
+
40
+ ## Citation
41
+ You can cite the following paper if you use our work:
42
+ ```
43
+ @INPROCEEDINGS{BounTi,
44
+ author={Köksal, Abdullatif and Özgür, Arzucan},
45
+ booktitle={2021 29th Signal Processing and Communications Applications Conference (SIU)},
46
+ title={Twitter Dataset and Evaluation of Transformers for Turkish Sentiment Analysis},
47
+ year={2021},
48
+ volume={},
49
+ number={}
50
+ }
51
+ ```
52
+ ---