5roop commited on
Commit
df57f85
1 Parent(s): 87f0240

Add README. Correct config with human readable labels.

Browse files
Files changed (2) hide show
  1. README.md +59 -0
  2. config.json +6 -6
README.md ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: "hr"
3
+
4
+ tags:
5
+ - text-classification
6
+ - sentiment-analysis
7
+
8
+ widget:
9
+ - text: "Potpredsjednik Vlade i ministar branitelja Tomo Medved komentirao je Vladine planove za zakonsku zabranu pozdrava 'za dom spremni'."
10
+ ---
11
+ # bcms-bertic-parlasent-bcs-ter
12
+
13
+ Text classification model based on [`classla/bcms-bertic`](https://huggingface.co/classla/bcms-bertic) and fine-tuned on the BCS Political Sentiment dataset.
14
+
15
+
16
+
17
+
18
+ ## Fine-tuning hyperparameters
19
+
20
+ Fine-tuning was performed with `simpletransformers`. Beforehand a brief sweep for the optimal number of epochs was performed and the presumed best value was 9.
21
+
22
+ ```python
23
+
24
+ model_args = {
25
+ "num_train_epochs": 9
26
+ }
27
+ ```
28
+
29
+ ## Performance
30
+
31
+ The same pipeline was run with two other transformer models and `fasttext` for comparison. Macro F1 scores were recorded for each of the 6 fine-tuning sessions and post festum analyzed.
32
+
33
+ | model | average macro F1 |
34
+ |---------------------------------|-------------------|
35
+ | bcms-bertic-parlasent-bcs-ter | 0.7941 ±0.0101 ** |
36
+ | EMBEDDIA/crosloengual-bert | 0.7709 ± 0.0113 |
37
+ | xlm-roberta-base | 0.7184 ± 0.0139 |
38
+ | fasttext + CLARIN.si embeddings | 0.6312 ± 0.0043 |
39
+
40
+ Two best performing models have been compared with the Mann-Whitney U test. (** denotes $p<0.01$).
41
+
42
+
43
+ ## Citation
44
+
45
+ If you use the model, please cite the following paper on which the original model is based:
46
+ ```
47
+ @inproceedings{ljubesic-lauc-2021-bertic,
48
+ title = "{BERT}i{\'c} - The Transformer Language Model for {B}osnian, {C}roatian, {M}ontenegrin and {S}erbian",
49
+ author = "Ljube{\v{s}}i{\'c}, Nikola and Lauc, Davor",
50
+ booktitle = "Proceedings of the 8th Workshop on Balto-Slavic Natural Language Processing",
51
+ month = apr,
52
+ year = "2021",
53
+ address = "Kiyv, Ukraine",
54
+ publisher = "Association for Computational Linguistics",
55
+ url = "https://www.aclweb.org/anthology/2021.bsnlp-1.5",
56
+ pages = "37--42",
57
+ }
58
+ ```
59
+
config.json CHANGED
@@ -10,16 +10,16 @@
10
  "hidden_dropout_prob": 0.1,
11
  "hidden_size": 768,
12
  "id2label": {
13
- "0": "LABEL_0",
14
- "1": "LABEL_1",
15
- "2": "LABEL_2"
16
  },
17
  "initializer_range": 0.02,
18
  "intermediate_size": 3072,
19
  "label2id": {
20
- "LABEL_0": 0,
21
- "LABEL_1": 1,
22
- "LABEL_2": 2
23
  },
24
  "layer_norm_eps": 1e-12,
25
  "max_position_embeddings": 512,
 
10
  "hidden_dropout_prob": 0.1,
11
  "hidden_size": 768,
12
  "id2label": {
13
+ "0": "Negative",
14
+ "1": "Neutral",
15
+ "2": "Positive"
16
  },
17
  "initializer_range": 0.02,
18
  "intermediate_size": 3072,
19
  "label2id": {
20
+ "Negative": 0,
21
+ "Neutral": 1,
22
+ "Positive": 2
23
  },
24
  "layer_norm_eps": 1e-12,
25
  "max_position_embeddings": 512,