Anonymized Product Data

This dataset contains anonymized train and test data used for reproducing results on product score prediction based on both product attributes and images. It is specifically used for evaluating the performance of the model, which predicts the selling potential or conversion of new listings. The dataset is designed for benchmarking and validation purposes in e-commerce ranking and recommendation systems. The product attributes have been anonymized by replacing actual values with generic identifiers (e.g., fabric → fabric_1, fabric_2; brand → brand_1, brand_2), ensuring privacy while retaining the statistical properties necessary for analysis.

File Details

• Format: Pickle (.pkl)
• Contents:
• • Product attributes: Structured metadata (e.g., type, price, fabric, etc.).
  • Image embeddings: Precomputed feature vectors extracted from product images using deep learning models.
  • Product score (target variable): Incorporates logged transformations of sales units, product page views (PPV), and impressions with a weighting parameter (α).