README.md

metadata

license: apache-2.0
task_categories:
  - text-generation
language:
  - en
tags:
  - recommendation
  - cross-domain
  - evaluation
  - kitrec
  - validation-data
size_categories:
  - 10K<n<100K

KitREC Validation Dataset - Set A

Validation dataset for the KitREC (Knowledge-Instruction Transfer for Recommendation) cross-domain recommendation system.

Dataset Description

This validation dataset is designed for evaluating fine-tuned LLMs on cross-domain recommendation tasks during training. It uses the same users as the test set but allows for validation monitoring.

Dataset Summary

Attribute	Value
Candidate Set	Set A (Hybrid (Hard negatives + Random))
Total Samples	12,000
Source Domain	Books
Target Domains	Movies & TV, Music
User Types	4 (2 per target domain)
Rating Range	1.0 - 5.0
Mean Rating	4.171

Set A vs Set B

• Set A (Hybrid): Contains hard negative candidates + random candidates for challenging evaluation
• Set B (Random): Contains only random candidates for fair baseline comparison

Both sets use the same ground truth items but differ in candidate composition.

User Type Distribution

User Type	Count	Percentage
overlapping_books_movies	3,000	25.00%
overlapping_books_music	3,000	25.00%
source_only_movies	3,000	25.00%
source_only_music	3,000	25.00%

User Type Definitions

User Type	Description
overlapping_books_movies	Users with history in both Books and Movies & TV
overlapping_books_music	Users with history in both Books and Music
source_only_movies	Users with ONLY Books history (extreme cold-start for Movies)
source_only_music	Users with ONLY Books history (extreme cold-start for Music)

Dataset Structure

Data Fields

• instruction (string): The recommendation prompt including user history
• input (string): Candidate items for recommendation (100 items per sample)
• gt_item_id (string): Ground truth item ID
• gt_title (string): Ground truth item title
• gt_rating (float): User's actual rating for the ground truth item (1-5 scale)
• user_id (string): Unique user identifier
• user_type (string): User category (4 types)
• candidate_set (string): A or B
• source_domain (string): Books
• target_domain (string): Movies & TV or Music
• candidate_count (int): Number of candidate items (100)

Data Split

Split	Samples	Description
val	12,000	Validation set for training monitoring

Usage

from datasets import load_dataset

# Load validation dataset
dataset = load_dataset("Younggooo/kitrec-val-seta")

# Access validation data
val_data = dataset["val"]
print(f"Validation samples: {len(val_data)}")

# Example: Filter by user type
overlapping_movies = val_data.filter(
    lambda x: x["user_type"] == "overlapping_books_movies"
)
print(f"Overlapping Movies users: {len(overlapping_movies)}")

# Example: Calculate metrics by user type
from collections import defaultdict
user_type_metrics = defaultdict(list)
for sample in val_data:
    user_type_metrics[sample["user_type"]].append(sample["gt_rating"])

Validation vs Test Data

Aspect	Validation (this dataset)	Test
Samples	12,000	30,000
User Types	4 types	10 types
Purpose	Training monitoring	Final evaluation
Usage	During training	After training

Why Separate Validation Set?

• Monitors training progress without data leakage
• Enables early stopping based on validation loss
• Validates model generalization during development

Evaluation Protocol

Metrics

• Hit@K (K=1, 5, 10): Whether GT item is in top-K predictions
• MRR: Mean Reciprocal Rank
• NDCG@10: Normalized Discounted Cumulative Gain

RQ4: Confidence-Rating Alignment

Use gt_rating field to analyze correlation between model's confidence scores and actual user ratings.

Citation

@misc{kitrec2024,
  title={KitREC: Knowledge-Instruction Transfer for Cross-Domain Recommendation},
  author={KitREC Research Team},
  year={2024},
  note={Validation dataset for cross-domain recommendation evaluation}
}

License

This dataset is released under the Apache 2.0 License.