--- library_name: setfit tags: - setfit - absa - sentence-transformers - text-classification - generated_from_setfit_trainer metrics: - accuracy widget: - text: gulungan biasa menjadi gulungan luar dalam,:dibutuhkan biaya tambahan $2 untuk mengubah gulungan biasa menjadi gulungan luar dalam, tetapi gulungan tersebut berukuran lebih dari tiga kali lipat, dan itu bukan ha dari nasi. - text: -a-bagel (baik di:ess-a-bagel (baik di sty-town atau midtown) sejauh ini merupakan bagel terbaik di ny. - text: mahal wadah ini pengelola:ketika kami sedang duduk makan makanan di bawah standar, manajer mulai mencaci-maki beberapa karyawan karena meletakkan wadah bumbu yang salah dan menjelaskan kepada mereka betapa mahal wadah ini pengelola - text: staf sangat akomodatif.:staf sangat akomodatif. - text: layanan luar biasa melayani:makanan india yang enak dan layanan luar biasa melayani pipeline_tag: text-classification inference: false base_model: BAAI/bge-m3 model-index: - name: SetFit Polarity Model with BAAI/bge-m3 results: - task: type: text-classification name: Text Classification dataset: name: Unknown type: unknown split: test metrics: - type: accuracy value: 0.7898320472083522 name: Accuracy --- # SetFit Polarity Model with BAAI/bge-m3 This is a [SetFit](https://github.com/huggingface/setfit) model that can be used for Aspect Based Sentiment Analysis (ABSA). This SetFit model uses [BAAI/bge-m3](https://huggingface.co/BAAI/bge-m3) as the Sentence Transformer embedding model. A [LogisticRegression](https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html) instance is used for classification. In particular, this model is in charge of classifying aspect polarities. The model has been trained using an efficient few-shot learning technique that involves: 1. Fine-tuning a [Sentence Transformer](https://www.sbert.net) with contrastive learning. 2. Training a classification head with features from the fine-tuned Sentence Transformer. This model was trained within the context of a larger system for ABSA, which looks like so: 1. Use a spaCy model to select possible aspect span candidates. 2. Use a SetFit model to filter these possible aspect span candidates. 3. **Use this SetFit model to classify the filtered aspect span candidates.** ## Model Details ### Model Description - **Model Type:** SetFit - **Sentence Transformer body:** [BAAI/bge-m3](https://huggingface.co/BAAI/bge-m3) - **Classification head:** a [LogisticRegression](https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html) instance - **spaCy Model:** id_core_news_trf - **SetFitABSA Aspect Model:** [firqaaa/indo-setfit-absa-bert-base-restaurants-aspect](https://huggingface.co/firqaaa/indo-setfit-absa-bert-base-restaurants-aspect) - **SetFitABSA Polarity Model:** [firqaaa/indo-setfit-absa-bert-base-restaurants-polarity](https://huggingface.co/firqaaa/indo-setfit-absa-bert-base-restaurants-polarity) - **Maximum Sequence Length:** 8192 tokens - **Number of Classes:** 4 classes ### Model Sources - **Repository:** [SetFit on GitHub](https://github.com/huggingface/setfit) - **Paper:** [Efficient Few-Shot Learning Without Prompts](https://arxiv.org/abs/2209.11055) - **Blogpost:** [SetFit: Efficient Few-Shot Learning Without Prompts](https://huggingface.co/blog/setfit) ### Model Labels | Label | Examples | |:--------|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | netral | | | negatif | | | positif | | | konflik | | ## Evaluation ### Metrics | Label | Accuracy | |:--------|:---------| | **all** | 0.7898 | ## Uses ### Direct Use for Inference First install the SetFit library: ```bash pip install setfit ``` Then you can load this model and run inference. ```python from setfit import AbsaModel # Download from the 🤗 Hub model = AbsaModel.from_pretrained( "firqaaa/setfit-indo-absa-restaurants-aspect", "firqaaa/setfit-indo-absa-restaurants-polarity", ) # Run inference preds = model("The food was great, but the venue is just way too busy.") ``` ## Training Details ### Training Set Metrics | Training set | Min | Median | Max | |:-------------|:----|:--------|:----| | Word count | 3 | 20.6594 | 62 | | Label | Training Sample Count | |:--------|:----------------------| | konflik | 34 | | negatif | 323 | | netral | 258 | | positif | 853 | ### Training Hyperparameters - batch_size: (16, 16) - num_epochs: (1, 1) - max_steps: -1 - sampling_strategy: oversampling - body_learning_rate: (2e-05, 1e-05) - head_learning_rate: 0.01 - loss: CosineSimilarityLoss - distance_metric: cosine_distance - margin: 0.25 - end_to_end: False - use_amp: True - warmup_proportion: 0.1 - seed: 42 - eval_max_steps: -1 - load_best_model_at_end: True ### Training Results | Epoch | Step | Training Loss | Validation Loss | |:----------:|:--------:|:-------------:|:---------------:| | 0.0000 | 1 | 0.2345 | - | | 0.0006 | 50 | 0.2337 | - | | 0.0013 | 100 | 0.267 | - | | 0.0019 | 150 | 0.2335 | - | | 0.0025 | 200 | 0.2368 | - | | 0.0032 | 250 | 0.2199 | - | | 0.0038 | 300 | 0.2325 | - | | 0.0045 | 350 | 0.2071 | - | | 0.0051 | 400 | 0.2229 | - | | 0.0057 | 450 | 0.1153 | - | | 0.0064 | 500 | 0.1771 | 0.1846 | | 0.0070 | 550 | 0.1612 | - | | 0.0076 | 600 | 0.1487 | - | | 0.0083 | 650 | 0.147 | - | | 0.0089 | 700 | 0.1982 | - | | 0.0096 | 750 | 0.1579 | - | | 0.0102 | 800 | 0.1148 | - | | 0.0108 | 850 | 0.1008 | - | | 0.0115 | 900 | 0.2035 | - | | 0.0121 | 950 | 0.1348 | - | | **0.0127** | **1000** | **0.0974** | **0.182** | | 0.0134 | 1050 | 0.121 | - | | 0.0140 | 1100 | 0.1949 | - | | 0.0147 | 1150 | 0.2424 | - | | 0.0153 | 1200 | 0.0601 | - | | 0.0159 | 1250 | 0.0968 | - | | 0.0166 | 1300 | 0.0137 | - | | 0.0172 | 1350 | 0.034 | - | | 0.0178 | 1400 | 0.1217 | - | | 0.0185 | 1450 | 0.0454 | - | | 0.0191 | 1500 | 0.0397 | 0.2216 | | 0.0198 | 1550 | 0.0226 | - | | 0.0204 | 1600 | 0.0939 | - | | 0.0210 | 1650 | 0.0537 | - | | 0.0217 | 1700 | 0.0566 | - | | 0.0223 | 1750 | 0.162 | - | | 0.0229 | 1800 | 0.0347 | - | | 0.0236 | 1850 | 0.103 | - | | 0.0242 | 1900 | 0.0615 | - | | 0.0249 | 1950 | 0.0589 | - | | 0.0255 | 2000 | 0.1668 | 0.2132 | | 0.0261 | 2050 | 0.1809 | - | | 0.0268 | 2100 | 0.0579 | - | | 0.0274 | 2150 | 0.088 | - | | 0.0280 | 2200 | 0.1047 | - | | 0.0287 | 2250 | 0.1255 | - | | 0.0293 | 2300 | 0.0312 | - | | 0.0300 | 2350 | 0.0097 | - | | 0.0306 | 2400 | 0.0973 | - | | 0.0312 | 2450 | 0.0066 | - | | 0.0319 | 2500 | 0.0589 | 0.2591 | | 0.0325 | 2550 | 0.0529 | - | | 0.0331 | 2600 | 0.0169 | - | | 0.0338 | 2650 | 0.0455 | - | | 0.0344 | 2700 | 0.0609 | - | | 0.0350 | 2750 | 0.1151 | - | | 0.0357 | 2800 | 0.0031 | - | | 0.0363 | 2850 | 0.0546 | - | | 0.0370 | 2900 | 0.0051 | - | | 0.0376 | 2950 | 0.0679 | - | | 0.0382 | 3000 | 0.0046 | 0.2646 | | 0.0389 | 3050 | 0.011 | - | | 0.0395 | 3100 | 0.0701 | - | | 0.0401 | 3150 | 0.0011 | - | | 0.0408 | 3200 | 0.011 | - | | 0.0414 | 3250 | 0.0026 | - | | 0.0421 | 3300 | 0.0027 | - | | 0.0427 | 3350 | 0.0012 | - | | 0.0433 | 3400 | 0.0454 | - | | 0.0440 | 3450 | 0.0011 | - | | 0.0446 | 3500 | 0.0012 | 0.2602 | * The bold row denotes the saved checkpoint. ### Framework Versions - Python: 3.10.13 - SetFit: 1.0.3 - Sentence Transformers: 2.2.2 - spaCy: 3.7.4 - Transformers: 4.36.2 - PyTorch: 2.1.2+cu121 - Datasets: 2.16.1 - Tokenizers: 0.15.0 ## Citation ### BibTeX ```bibtex @article{https://doi.org/10.48550/arxiv.2209.11055, doi = {10.48550/ARXIV.2209.11055}, url = {https://arxiv.org/abs/2209.11055}, author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren}, keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences}, title = {Efficient Few-Shot Learning Without Prompts}, publisher = {arXiv}, year = {2022}, copyright = {Creative Commons Attribution 4.0 International} } ```