Amazon Product Reviews for Multi-Class Classification

2000 Amazon product reviews are structured for a 10-class classification task with 100 features. This dataset was created for machine learning benchmarking, likely sourced from the OpenML platform. The specific sampling method used stratification to maintain class distribution.

Use Cases

• Predict product review sentiment or rating using the 100 engineered text features.
• Perform multi-class classification across 10 distinct product categories or sentiment labels.
• Analyze feature importance among the 100 columns for understanding review determinants.
• Benchmark model performance on a stratified, balanced sample of 2000 review records.

Strengths

• Dataset contains 2000 rows, providing a moderate sample for model training.
• Stratified sampling ensures balanced representation across 10 target classes.
• Includes 100 features per row, offering a high-dimensional feature space for analysis.

Limitations

• Sample size of 2000 rows may be insufficient for training complex deep learning models.
• The origin and preprocessing of the 100 features are unknown, potentially obscuring data meaning.
• Lacks metadata such as original review text, timestamps, and product identifiers.

Provenance

Source

OpenML platform, likely derived from original Amazon review data.

Collection Method

Sampled and preprocessed for a classification task, with stratification applied.