Kinoux/french-customer-review-sentiment-free-2k

French Customer Review Sentiment Free 2K

French Customer Review Sentiment Free 2K is a free 2,000-record sample extracted from the full commercial dataset French Customer Review Sentiment (100k synthetic reviews) provided by Kinoux.

Each entry is a synthetic French customer review labeled with a 3-class sentiment:

• positive
• neutral
• negative

The data is 100% synthetic (no personal data, no real platform exports) and was generated and curated specifically for training and evaluating French-native sentiment models.

Official product page for the full dataset and model: https://kinoux.com/datasets-models/french-customer-review-sentiment/

---

Dataset Description

Summary

This dataset contains 2,000 French customer reviews, each annotated with a single sentiment label among positive, neutral, and negative.

It is designed as a free preview of the full commercial dataset (100,000 records) used to fine-tune a CamemBERT v2-based sentiment classifier (3 classes) for French customer feedback.

Supported Tasks and Leaderboards

• Task: Sentiment analysis / text classification
• Input: French review text (text)
• Output: Sentiment label (sentiment) in {positive, neutral, negative}

This sample can be used to:

• Train and evaluate French sentiment classifiers
• Benchmark CamemBERT / Transformers pipelines on French sentiment
• Build internal benchmarks and dashboards for Positive / Neutral / Negative KPIs
• Explore edge cases like soft negatives, polite complaints, and French idioms

Languages

• Input language: French (fr)
• Labels: English strings (positive, neutral, negative)

---

Dataset Structure

Data Instances

Each line is a JSON object like:

{"text": "Le support a résolu mon problème rapidement, je suis très satisfaite.", "sentiment": "positive"}

A few more examples:

{"text": "J'ai vraiment eu l'impression d'avoir été piégée avec le taux d'intérêt. Pas du tout clair et ça fait mal au porte-monnaie !", "sentiment": "negative"}
{"text": "L'accès au WiFi a été tout à fait satisfaisant. La connexion était rapide et stable, ce qui m'a permis de rester en contact avec mes proches sans aucune difficulté.", "sentiment": "positive"}
{"text": "L'appli est géniale ! Parfait pour composer des tenues sans stress, le site mobile répond super bien.", "sentiment": "positive"}

Data Fields

• text (string): Synthetic French customer review text (single review per line).

• sentiment (string): Sentiment label with three possible values:

  • positive
  • neutral
  • negative

Data Splits

This free 2K sample is provided as a single split.

Users are encouraged to create their own train/validation splits (e.g. 85/15 stratified by label).

The full 100k dataset (commercial) includes a recommended 85/15 stratified split and additional evaluation utilities.

---

Data Generation and Quality

• All records are synthetically generated with French-native patterns.

• No real platform exports, no customer IDs, and no personal data are included.

• Texts cover a variety of tones and scenarios:

  • App reviews
  • E-commerce feedback
  • Service and support interactions
  • Product and UX comments

The goal is to provide a clean, homogeneous, and controllable corpus for sentiment modeling, without the noise and licensing ambiguity commonly found in open data scraped from the web.

---

Usage

Example: Loading with datasets

from datasets import load_dataset

# Replace `kinoux` with your actual HF organization/username if different

ds = load_dataset(
    "kinoux/french-customer-review-sentiment-free-2k",
    split="train"
)

print(ds[0])
# {'text': '...', 'sentiment': 'positive'}

Example: Training a CamemBERT classifier (sketch)

from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer

# Load and split the dataset
dataset = load_dataset("kinoux/french-customer-review-sentiment-free-2k", split="train")
dataset = dataset.train_test_split(test_size=0.15, stratify_by_column="sentiment")

# Tokenization
tokenizer = AutoTokenizer.from_pretrained("camembert-base")

def tokenize_batch(batch):
    return tokenizer(batch["text"], truncation=True, padding="max_length")

tokenized = dataset.map(tokenize_batch, batched=True)

# Label encoding
label2id = {"negative": 0, "neutral": 1, "positive": 2}
id2label = {v: k for k, v in label2id.items()}

def encode_labels(batch):
    batch["labels"] = [label2id[s] for s in batch["sentiment"]]
    return batch

tokenized = tokenized.map(encode_labels, batched=True)

# Model
model = AutoModelForSequenceClassification.from_pretrained(
    "camembert-base",
    num_labels=3,
    id2label=id2label,
    label2id=label2id,
)

# Training arguments
args = TrainingArguments(
    output_dir="./sentiment_fr_model",
    evaluation_strategy="epoch",
    learning_rate=2e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
)

trainer = Trainer(
    model=model,
    args=args,
    train_dataset=tokenized["train"],
    eval_dataset=tokenized["test"],
)

trainer.train()

---

Limitations

• This is a small sample (2k): suitable for experimentation, teaching, and sanity checks, but not sufficient alone for production-grade models in demanding domains.
• Domain coverage is broad but synthetic; real-world performance should always be validated on in-domain data.
• Labels are single-label sentiment and do not capture aspects (e.g. "delivery vs product vs support").

For serious production work, consider using the full 100k dataset and/or augmenting with your own in-domain examples.

---

Full Dataset & Model (Commercial)

The full French Customer Review Sentiment product from Kinoux includes:

• 100,000 synthetic French reviews (JSONL)
• 3-class sentiment labels (negative, neutral, positive)
• Recommended 85/15 stratified split and evaluation template
• A CamemBERT v2-based fine-tuned sentiment model
• ONNX export guidance for low-latency deployment
• Clear commercial license for on-prem / private cloud use

More details and licensing options on our website.

---

Licensing Information

This dataset is distributed under the Kinoux Dataset License (Version 2025-06-01).

• You may use this sample for internal research, prototyping, model training and evaluation.
• Redistribution of the dataset, in whole or in part, is prohibited, including uploading copies to other platforms or public repos.
• For detailed terms, please refer to the LICENSE.md file in this repository.

To obtain commercial access to all data and the model, please visit our website.

---

Citation

If you use this dataset in a publication, product, or internal report, please cite:

Kinoux – French Customer Review Sentiment Free 2K (synthetic dataset). Available as a sample on Hugging Face. Complete product and documentation on our website.