This dataset contains millions of Amazon customer reviews paired with star ratings. It serves as a powerful resource for training fastText models for sentiment analysis.

• Scale: The dataset is substantial, reflecting real business data on a reasonable scale, making it suitable for practical applications.
• Efficiency: Despite its size, the dataset can be trained in minutes on a modest laptop, making it ideal for fastText applications.
• Realistic: The reviews represent genuine customer sentiments, providing a valuable insight into real-world opinions.

The dataset adheres to the format required for the fastText supervised learning tutorial:

__label__<X> __label__<Y> ... <Text>

• X and Y represent class names (e.g., __label__1, __label__2).
• Reviews are categorized into classes with no quotes, all on one line.
• For this dataset, __label__1 corresponds to 1- and 2-star reviews, and __label__2 corresponds to 4- and 5-star reviews.
• Reviews with neutral sentiment (3-star) were not included in the original dataset.
• Review titles, followed by ':' and a space, are prepended to the text.
• While most reviews are in English, there are some in other languages, such as Spanish.

The data was sourced from Xiang Zhang's Google Drive directory. Note that the original data was in .csv format, which was reformatted for compatibility with fastText.

Follow the instructions in the fastText supervised learning tutorial to set up the directory.

Use the following command to train the model:

./fasttext supervised -input train.ft.txt -output model_amzn

This process typically takes just a few minutes.

Verify the accuracy of the trained model with the following test command:

./fasttext test model_amzn.bin test.ft.txt

If everything is in order, you should expect a precision and recall of 0.916.

Additionally, you can perform training and testing using Python; refer to the provided Kernel for details.

Happy analyzing! 🚀