Amazon Commerce Reviews for Multi-Class Text Classification

2000 product reviews from Amazon form a dataset for multi-class classification tasks. The data is structured with 100 columns and 10 target classes, using stratified sampling. It originates from the OpenML platform, though the specific author and collection date are unknown.

Use Cases

• Predict the review's target class (1 of 10 categories) using features from the 100 available columns.
• Perform sentiment analysis by modeling the relationship between review text features and a potential rating or sentiment column.
• Analyze feature importance among the 100 columns to identify key terms or metadata influencing review classification.
• Train a stratified multi-class classifier (e.g., logistic regression, BERT) on the 10 balanced classes to prevent model bias.

Strengths

• Dataset contains 2000 rows, providing a solid sample for model training.
• Stratified sampling ensures balanced representation across the 10 target classes.
• Includes 100 columns, suggesting a potentially detailed feature set for analysis.

Limitations

• Sample size of 2000 rows may be insufficient for training very deep neural networks without augmentation.
• Unknown column definitions and data types complicate direct feature engineering and interpretation.
• Lack of provenance details (author, time range) limits understanding of data context and potential biases.

Provenance

Source

OpenML platform

Collection Method

Presumably extracted or sampled from a larger corpus of Amazon product reviews.