distilbert_amazon_book_classification
This model is a fine-tuned version of distilbert-base-uncased on an Kaggle Amazon Kindle Books dataset. It achieves the following results on the evaluation set:
- Loss: 1.4475
- Accuracy: 0.5871
- F1 Score: 0.5865
- Precision: 0.5967
- Recall: 0.5871
Model description
This model is a fine-tuned version of distilbert-base-uncased trained directly on structured Amazon book metadata across 31 standardized Kindle categories.
Serving as a clean-data benchmark in comparative analysis, this model evaluates genre classification performance under structured, editorial metadata conditions. It achieves an Accuracy of 58.71% and a Macro F1-Score of 58.65%, demonstrating that high inherent data quality and structured category labels significantly improve the upper-bound performance of transformer architectures.
Intended uses & limitations
-
Direct classification of structured book descriptions into standard Amazon Kindle categories.
-
Comparative benchmarking for domain adaptation and cross-domain transfer learning experiments.
-
May underperform or show sensitivity when applied to highly informal, uncurated, or user-generated text inputs without prior domain adaptation.
Datasets
- Amazon-Dataset: Kaggle Amazon Kindle Books Dataset)
Training and evaluation data
-
Source: Official Amazon Kindle book product listings and metadata.
-
Target Taxonomy: 31 standardized Amazon Kindle categories (e.g., Literature & Fiction, Sci-Fi & Fantasy, Romance, Business & Money).
-
Input Features: Concatenated book title and author.
-
Metadata Quality: High quality, structured, and editorially curated product descriptions with minimal noise compared to community-driven tags.
-
Splits: Partitioned into stratified (80/20) train and test sets across all 31 target classes.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 Score | Precision | Recall |
|---|---|---|---|---|---|---|---|
| 1.6436 | 0.9999 | 9679 | 1.4688 | 0.5680 | 0.5624 | 0.5822 | 0.5680 |
| 1.0845 | 1.9998 | 19358 | 1.4475 | 0.5871 | 0.5865 | 0.5967 | 0.5871 |
Framework versions
- Transformers 4.45.2
- Pytorch 2.5.1
- Datasets 4.1.1
- Tokenizers 0.20.1
Academic Context & Citation / Akademischer Kontext
This repository and model were developed as part of a Bachelor's thesis in 2026.
- Title: Classification of Goodreads genres: A methodological comparison of Doc2Vec and DistilBERT
- License: CC BY-NC 4.0 (Free for research, education, and personal use; commercial use prohibited)
Dieses Repository und Modell wurden im Rahmen einer Bachelorarbeit im Jahr 2026 entwickelt.
- Titel: Klassifikation von Goodreads-Genres: Ein methodischer Vergleich von Doc2Vec und DistilBERT
- Lizenz: CC BY-NC 4.0 (Frei für Forschung, Lehre und private Nutzung; kommerzielle Nutzung untersagt)