File size: 2,802 Bytes
4be2513
 
2c0661b
 
 
 
 
 
 
 
 
 
 
 
 
 
4be2513
2c0661b
 
 
 
fefeab0
2c0661b
4da2863
 
2c0661b
 
 
 
 
 
 
 
 
 
 
 
 
fefeab0
 
2c0661b
36f7396
2c0661b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4da2863
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
---
license: cc-by-4.0
task_categories:
- text-classification
- zero-shot-classification
language:
- en
- tl
tags:
- code-switching
- sentiment analysis
- low-resource languages
- taglish
- Filipino
size_categories:
- 10K<n<100K
---
# Dataset Card for Filipino-English Reviews with Code-Switching (FiReCS)

### Dataset Summary

We introduce FiReCS, the first sentiment-annotated corpus of product and service reviews involving Filipino-English code-switching. The data set is composed of 10,487 reviews with a fairly balanced number per sentiment class. Inter-annotator agreement is high with a Kripendorffs’s α for ordinal metric of 0.83. Three human annotators were tasked to manually label reviews according to three polarity classes: Positive, Neutral, and Negative. 

[UPDATE 2024/05/09] An updated dataset containing four polarity classes can be found at [SentiTaglish: Products and Services](https://huggingface.co/datasets/ccosme/SentiTaglishProductsAndServices).

### Supported Tasks and Leaderboards

Sentiment analysis of bilingual text with code-switching / code-mixing.

### Languages

- Filipino
- English

## Dataset Structure

### Data Fields

* `review`: a string containing the body of the review
* `label`: an integer containing the label encoding of the gold-truth label provided by the human annotators

#### Label encoding
* 2 - Positive
* 1 - Neutral
* 0 - Negative

### Data Splits

| Data set split | Positive | Neutral | Negative | Total |
| -------------- | -------- | ------- | -------- | ----- |
| Train set      | 2,410    | 2,549   | 2,381    | 7,340 |
| Test set       | 1,033    | 1,087   | 1,027    | 3,147 |

### Dataset Creation and Annotation

The data set was created using publicly available online service and product reviews from Google Maps Reviews and Shopee Philippines. Only the rating and review fields were collected and stored.

Three annotators, all native speakers of Filipino and fluent in English, were tasked to manually label the data set. The first two annotators labeled the same full set of reviews. Any disagreements were sent to a third annotator.

### Personal and Sensitive Information

No personal information were collected and stored.

### Licensing Information

The FiReCS data set version 1.0 is released under the [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/) license.

### Citation Information

Cosme, C.J., De Leon, M.M. (2024). Sentiment Analysis of Code-Switched Filipino-English Product and Service Reviews Using Transformers-Based Large Language Models. In: Iglesias, A., Shin, J., Patel, B., Joshi, A. (eds) Proceedings of World Conference on Information Systems for Business Management. ISBM 2023. Lecture Notes in Networks and Systems, vol 834. Springer, Singapore. https://doi.org/10.1007/978-981-99-8349-0_11