avramandrei/RoIt-XMASA

---
language:
- ro
- it
license: cc-by-4.0
task_categories:
- text-classification
task_ids:
- sentiment-classification
pretty_name: RoIt-XMASA
size_categories:
- 100K<n<1M
tags:
- sentiment-analysis
- cross-lingual
- cross-domain
- romanian
- italian
- reviews
---

# RoIt-XMASA

**Romanian–Italian Cross-domain Multi-domain Sentiment Analysis**

RoIt-XMASA is a multilingual, cross-domain sentiment analysis dataset containing user reviews in Romanian (RO) and Italian (IT) across three domains: Books, Movies, and Music. Reviews are annotated with 1–5 star ratings (excluding 3).

## Dataset Summary

| Split | Rows | Labeled |
|------------|---------|---------|
| train | 12,000 | yes |
| validation | 12,000 | yes |
| test | 12,000 | yes |
| unlabeled | 202,141 | no |

### Splits breakdown (labeled)

Each labeled split is perfectly balanced:
- **Domains**: 4,000 reviews per domain (Books / Movies / Music)
- **Languages**: 6,000 reviews per language (RO / IT)

## Data Fields

| Field | Type | Description |
|------------|--------|---------------------------------------------------|
| `id` | int64 | Unique review identifier |
| `title` | string | Review title (may be empty) |
| `text` | string | Review body |
| `domain` | string | Domain: `Books`, `Movies`, or `Music` |
| `language` | string | Language code: `RO` (Romanian) or `IT` (Italian) |
| `rating` | int64 | Star rating: 1, 2, 4, or 5 (null for unlabeled) |

## Usage

```python
from datasets import load_dataset

ds = load_dataset("avramandrei/RoIt-XMASA")

# Labeled splits
train = ds["train"]
val = ds["validation"]
test = ds["test"]

# Filter by language or domain
ro_books = train.filter(lambda x: x["language"] == "RO" and x["domain"] == "Books")
```