---
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
dataset_info:
  features:
  - name: text
    dtype: string
  - name: label
    dtype: int64
  - name: language
    dtype: string
  - name: dataset
    dtype: string
  splits:
  - name: train
    num_bytes: 147033765
    num_examples: 670450
  - name: validation
    num_bytes: 772660
    num_examples: 3386
  - name: test
    num_bytes: 734021
    num_examples: 3386
  download_size: 81782267
  dataset_size: 148540446
---
# Dataset Card for "sentiment"

## Dataset Description

- **Point of Contact:** [Oliver Kinch](mailto:oliver.kinch@alexandra.dk)
- **Size of dataset:** 139 MB

### Dataset Summary

The dataset includes review comments from sources like TrustPilot and comments from social media platforms.

It's a binary sentiment dataset where a label of 0 represents "negative" text and a label of 1 indicates "positive" text.

### Supported Tasks

Sentiment classification is the intended task for this dataset.

### Languages

The dataset predominantly contains Danish (da) text but also includes Norwegian (no), Swedish (sv), and English (en) text.

## Dataset Structure

An example from the dataset looks as follows.

```
{
  "text": "some text",
  "label": 0,
  "language": "en",
  "dataset": "da.csv"
}
```

### Data Fields

- `text`: a `string` feature.
- `label`: an `int64` feature.
- `language`: a `string` feature.
- `dataset`: a `string` feature.

The `dataset` feature specifies the sample's source. If it's `danish-some-sentiment.tsv`, it's a social media comment; otherwise, it's a review comment. See `Source Data` for more information.

#### Number of samples

- Train: 670,450
- Val: 3,386 
- Test: 3,386

#### Language distribution

There are 10,000 samples for each language except for Danish, for which there are significantly more samples.

![image/png](https://cdn-uploads.huggingface.co/production/uploads/61e0713ac50610f535ed2c88/F9_AMuQVlwPC_scMJDagk.png)

#### Label distribution

- Class 0: 195,242
- Class 1: 481,980

![image/png](https://cdn-uploads.huggingface.co/production/uploads/61e0713ac50610f535ed2c88/kX-y_BCjopcmyVXQuSyc8.png)

#### Text lengths (Number of characters)

- Minimum: 1 (for example just an emoji)
- Maximum: 22,689
- Mean: 180.0
- Median: 76.0

![image/png](https://cdn-uploads.huggingface.co/production/uploads/61e0713ac50610f535ed2c88/iDTzqDuqRWmE2RvVs-V0y.png)


## Dataset Creation

### Curation Rationale

Larger sentiment datasets in the Scandinavian languages are needed to improve sentiment model training for these languages.

### Source Data

The dataset is a compilation of multiple datasets.

*Review datasets*:

- `da-reviews.tsv`
- `da.csv`
- `en.csv`
- `no.csv`
- `sv.csv`

*Social media comment datasets*:

- `danish-some-sentiment.tsv`

For specific dataset inquiries, please contact [Oliver Kinch](https://huggingface.co/oliverkinch)


## Additional Information

### Dataset Curators

[Oliver Kinch](https://huggingface.co/oliverkinch) from the [The Alexandra
Institute](https://alexandra.dk/)

### Licensing Information

The dataset is licensed under the [CC0
license](https://creativecommons.org/share-your-work/public-domain/cc0/).