metadata
language: de
tags:
- bert
license: apache-2.0
widget:
- text: Das Frühstück ist sehr gut, es gibt auch Laktosefreie Produkte.
example_title: Example 1
- text: Das Personal ist sehr kompetent und sehr freundlich.
example_title: Example 2
- text: >-
Die Zimmer sind wie beschrieben sehr klein, vergleichbar mit einer
Kreuzfahrtschiffkabine.
example_title: Example 3
- text: Scheinwerfer vor dem Zimmer ganze Nacht an und zu hell
example_title: Example 4
German Hotel Review Sentiment Classification
A model trained on German Hotel Reviews from Switzerland. The base model is the bert-base-german-cased. The last hidden layer of the base model was extracted and a classification layer was added. The entire model was then trained for 5 epochs on our dataset.
Model Performance
| Classes | Precision | Recall | F1 Score |
|---|---|---|---|
| Room | 84.62% | 88.00% | 86.27% |
| Food | 79.17% | 82.61% | 80.85% |
| Staff | 63.64% | 70.00% | 66.67% |
| Location | 83.33% | 62.50% | 71.43% |
| GeneralUtilitys | 76.92% | 76.92% | 76.92% |
| HotelOrganisation | 26.67% | 30.77% | 28.57% |
| Unknown | 25.00% | 16.67% | 20.00% |
| ReasonForStay | 100.00% | 50.00% | 66.67% |
| Accuracy | 69.00% | ||
| Macro Average | 67.42% | 59.68% | 62.17% |
| Weighted Average | 69.36% | 69.00% | 68.79% |
Confusion Matrix
