ginkgogo/setfit-absa-bge-small-en-v1.5-restaurants-aspect

SetFit Aspect Model with sentence-transformers/all-MiniLM-L6-v2

This is a SetFit model that can be used for Aspect Based Sentiment Analysis (ABSA). This SetFit model uses sentence-transformers/all-MiniLM-L6-v2 as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification. In particular, this model is in charge of filtering aspect span candidates.

The model has been trained using an efficient few-shot learning technique that involves:

1. Fine-tuning a Sentence Transformer 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 this SetFit model to filter these possible aspect span candidates.
3. Use a SetFit model to classify the filtered aspect span candidates.

Model Details

Model Description

• Model Type: SetFit
• Sentence Transformer body: sentence-transformers/all-MiniLM-L6-v2
• Classification head: a LogisticRegression instance
• spaCy Model: en_core_web_sm
• SetFitABSA Aspect Model: ginkgogo/setfit-absa-bge-small-en-v1.5-restaurants-aspect
• SetFitABSA Polarity Model: ginkgogo/setfit-absa-bge-small-en-v1.5-restaurants-polarity
• Maximum Sequence Length: 256 tokens
• Number of Classes: 2 classes

Model Sources

• Repository: SetFit on GitHub
• Paper: Efficient Few-Shot Learning Without Prompts
• Blogpost: SetFit: Efficient Few-Shot Learning Without Prompts

Model Labels

Label	Examples
aspect	'food:They made it into more American food, added burgers and ribs and got rid of the tequila selection. We were so bummed. Used to be one of our favorite places to go for good Mexican food. The owner said the new direction was to appeal to more tourists.' "seating:Such a cute little spot for desserts! I'm so glad we had time on our short visit to Santa Barbara to grab a slice of cake from here. My husband and I each got our own to slice to share of course. He said we didn't come all this way just to get one so we chose a slice of the berry cake and chocolate decadence. The berry cake was nice and fluffy without being too sweet. The acidity from the fruits balanced the sweetest of the cake wonderfully. If you're up for something rich then the chocolate decadence will not disappoint. Service was great and seating was comfortable. Order your sweet treats at the counter then a number will be given to you. Pick a table and get ready to enjoy because your sweets will be brought out to your table when ready." 'food:One brisk Saturday morning after asking workers during a stop for tylenol from the Hotel California Boutique the best breakfast place, they recommended Goat Tree. We crossed the busy street and greeted the hostess. The very kind young lady walked us to our table on the sunny patio. We skimmed the menu and decided on the chicken and waffle and a chocolate croissant. The wait was quite short and we spent it discussing the beautiful surrounding area. Soon, our food was delivered, and let me tell you, it was beautiful. On top of that, it was scrumptious. The fried chicken was perfect and tender. The waffle had the perfect balance of crunch and fluff. And how dare I forget the exquisite honey. Now this honey was the best I have ever tasted. It was topped with chia and pumpkin seeds. My daughter asked for her croissant warmed, and once again it was marvelous. After paying, I told our waitress how amazing the honey was. Next thing we knew, she brought out two large to go cups full of it! \n\nAbsolutely loved this place and everything about it. 100% recommend! I strongly award them 5 stars!'
no aspect	'burgers:They made it into more American food, added burgers and ribs and got rid of the tequila selection. We were so bummed. Used to be one of our favorite places to go for good Mexican food. The owner said the new direction was to appeal to more tourists.' 'ribs:They made it into more American food, added burgers and ribs and got rid of the tequila selection. We were so bummed. Used to be one of our favorite places to go for good Mexican food. The owner said the new direction was to appeal to more tourists.' 'tequila selection:They made it into more American food, added burgers and ribs and got rid of the tequila selection. We were so bummed. Used to be one of our favorite places to go for good Mexican food. The owner said the new direction was to appeal to more tourists.'

Evaluation

Metrics

Label	Accuracy
all	0.9570

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

from setfit import AbsaModel

# Download from the 🤗 Hub
model = AbsaModel.from_pretrained(
    "ginkgogo/setfit-absa-bge-small-en-v1.5-restaurants-aspect",
    "ginkgogo/setfit-absa-bge-small-en-v1.5-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	21	152.7030	268

Label	Training Sample Count
no aspect	383
aspect	21

Training Hyperparameters

• batch_size: (50, 50)
• num_epochs: (5, 5)
• 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
0.0003	1	0.2856	-
0.0169	50	0.2755	0.3092
0.0339	100	0.2895	0.2962
0.0508	150	0.2845	0.2876
0.0678	200	0.2471	0.2826
0.0847	250	0.2124	0.2691
0.1017	300	0.1357	0.184
0.1186	350	0.0362	0.0871
0.1355	400	0.07	0.0848
0.1525	450	0.0184	0.092
0.1694	500	0.0179	0.096
0.1864	550	0.0033	0.097
0.2033	600	0.0037	0.0978
0.2203	650	0.04	0.1046

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• SetFit: 1.0.3
• Sentence Transformers: 2.6.0
• spaCy: 3.7.4
• Transformers: 4.38.2
• PyTorch: 2.2.1+cu121
• Datasets: 2.18.0
• Tokenizers: 0.15.2

Citation

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}
}