SetFit Polarity Model with sentence-transformers/paraphrase-mpnet-base-v2

This is a SetFit model that can be used for Aspect Based Sentiment Analysis (ABSA). This SetFit model uses sentence-transformers/paraphrase-mpnet-base-v2 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:

Fine-tuning a Sentence Transformer with contrastive learning.
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:

Use a spaCy model to select possible aspect span candidates.
Use a SetFit model to filter these possible aspect span candidates.
Use this SetFit model to classify the filtered aspect span candidates.

Model Details

Model Description

Model Type: SetFit
Sentence Transformer body: sentence-transformers/paraphrase-mpnet-base-v2
Classification head: a LogisticRegression instance
spaCy Model: en_core_web_lg
SetFitABSA Aspect Model: tomaarsen/setfit-absa-paraphrase-mpnet-base-v2-restaurants-aspect
SetFitABSA Polarity Model: tomaarsen/setfit-absa-paraphrase-mpnet-base-v2-restaurants-polarity
Maximum Sequence Length: 512 tokens
Number of Classes: 4 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
negative	'But the staff was so horrible:But the staff was so horrible to us.' ', forgot our toast, left out:They did not have mayonnaise, forgot our toast, left out ingredients (ie cheese in an omelet), below hot temperatures and the bacon was so over cooked it crumbled on the plate when you touched it.' 'did not have mayonnaise, forgot our:They did not have mayonnaise, forgot our toast, left out ingredients (ie cheese in an omelet), below hot temperatures and the bacon was so over cooked it crumbled on the plate when you touched it.'
positive	"factor was the food, which was:To be completely fair, the only redeeming factor was the food, which was above average, but couldn't make up for all the other deficiencies of Teodora." "The food is uniformly exceptional:The food is uniformly exceptional, with a very capable kitchen which will proudly whip up whatever you feel like eating, whether it's on the menu or not." "a very capable kitchen which will proudly:The food is uniformly exceptional, with a very capable kitchen which will proudly whip up whatever you feel like eating, whether it's on the menu or not."
neutral	"'s on the menu or not.:The food is uniformly exceptional, with a very capable kitchen which will proudly whip up whatever you feel like eating, whether it's on the menu or not." 'to sample both meats).:Our agreed favorite is the orrechiete with sausage and chicken (usually the waiters are kind enough to split the dish in half so you get to sample both meats).' 'to split the dish in half so:Our agreed favorite is the orrechiete with sausage and chicken (usually the waiters are kind enough to split the dish in half so you get to sample both meats).'
conflict	'The food was delicious but:The food was delicious but do not come here on a empty stomach.' "The service varys from day:The service varys from day to day- sometimes they're very nice, and sometimes not."

Evaluation

Metrics

Label	Accuracy
all	0.7241

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(
    "tomaarsen/setfit-absa-paraphrase-mpnet-base-v2-restaurants-aspect",
    "tomaarsen/setfit-absa-paraphrase-mpnet-base-v2-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	6	21.3594	43

Label	Training Sample Count
conflict	2
negative	19
neutral	25
positive	82

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
0.0018	1	0.221	-
0.0923	50	0.1118	-
0.1845	100	0.0784	-
0.2768	150	0.0024	-
0.3690	200	0.0004	-
0.4613	250	0.0003	-
0.5535	300	0.0006	-
0.6458	350	0.0004	-
0.7380	400	0.0005	-
0.8303	450	0.0001	-
0.9225	500	0.0003	-

Environmental Impact

Carbon emissions were measured using CodeCarbon.

Carbon Emitted: 0.004 kg of CO2
Hours Used: 0.054 hours

Training Hardware

On Cloud: No
GPU Model: 1 x NVIDIA GeForce RTX 3090
CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
RAM Size: 31.78 GB

Framework Versions

Python: 3.9.16
SetFit: 1.0.0.dev0
Sentence Transformers: 2.2.2
spaCy: 3.7.2
Transformers: 4.29.0
PyTorch: 1.13.1+cu117
Datasets: 2.15.0
Tokenizers: 0.13.3

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}
}