isolation-forest/setfit-absa-polarity

SetFit Polarity Model with cointegrated/rubert-tiny2

This is a SetFit model that can be used for Aspect Based Sentiment Analysis (ABSA). This SetFit model uses cointegrated/rubert-tiny2 as the Sentence Transformer embedding model. A LogisticRegression 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 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: cointegrated/rubert-tiny2
• Classification head: a LogisticRegression instance
• spaCy Model: ru_core_news_lg
• SetFitABSA Aspect Model: isolation-forest/setfit-absa-aspect
• SetFitABSA Polarity Model: isolation-forest/setfit-absa-polarity
• Maximum Sequence Length: 2048 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
Positive	'И порции " достойные ":И порции " достойные " .' 'Салаты вообще оказались вкуснейшими:Салаты вообще оказались вкуснейшими .' 'порадовала , большая пивная тарелка , действительно оказалась:Кухня порадовала , большая пивная тарелка , действительно оказалась большой и вкусной !'
Negative	'Потом официантка как будто пропала:Потом официантка как будто пропала , было не дозваться , чтобы что - то дозаказать , очень долго приходилось ждать , в итоге посчитали неправильно , в счет внесли на 2 пункта больше , чем мы заказывали .' 'Обслуживание не впечатлило .:Обслуживание не впечатлило .' 'приятно удивлена " китайским интерьером " - диванчики:Была приятно удивлена " китайским интерьером " - диванчики как в бистро , скатерти на столах по типу а - ля столовая , европейские светильники / люстры , в общем в плане интерьера китайского никакого абсолютно !'

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(
    "isolation-forest/setfit-absa-aspect",
    "isolation-forest/setfit-absa-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	28.4766	92

Label	Training Sample Count
Negative	128
Positive	128

Training Hyperparameters

• batch_size: (16, 2)
• num_epochs: (1, 16)
• 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: False
• warmup_proportion: 0.1
• seed: 42
• eval_max_steps: -1
• load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0005	1	0.2196	-
0.0242	50	0.2339	-
0.0484	100	0.2258	-
0.0727	150	0.246	-
0.0969	200	0.1963	-
0.1211	250	0.18	-
0.1453	300	0.1176	-
0.1696	350	0.0588	-
0.1938	400	0.0482	-
0.2180	450	0.1131	-
0.2422	500	0.0134	-
0.2665	550	0.0415	-
0.2907	600	0.0144	-
0.3149	650	0.012	-
0.3391	700	0.0091	-
0.3634	750	0.0055	-
0.3876	800	0.0054	-
0.4118	850	0.0055	-
0.4360	900	0.0072	-
0.4603	950	0.0094	-
0.4845	1000	0.0054	-
0.5087	1050	0.0045	-
0.5329	1100	0.003	-
0.5572	1150	0.0067	-
0.5814	1200	0.0041	-
0.6056	1250	0.0048	-
0.6298	1300	0.0053	-
0.6541	1350	0.0048	-
0.6783	1400	0.0038	-
0.7025	1450	0.0037	-
0.7267	1500	0.0031	-
0.7510	1550	0.0038	-
0.7752	1600	0.0032	-
0.7994	1650	0.0039	-
0.8236	1700	0.0032	-
0.8479	1750	0.0023	-
0.8721	1800	0.0029	-
0.8963	1850	0.0041	-
0.9205	1900	0.0026	-
0.9448	1950	0.0027	-
0.9690	2000	0.0035	-
0.9932	2050	0.003	-

Framework Versions

• Python: 3.10.13
• SetFit: 1.0.3
• Sentence Transformers: 2.7.0
• spaCy: 3.7.2
• Transformers: 4.39.3
• PyTorch: 2.1.2
• 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}
}