SetFit Aspect Model with firqaaa/indo-sentence-bert-base

This is a SetFit model that can be used for Aspect Based Sentiment Analysis (ABSA). This SetFit model uses firqaaa/indo-sentence-bert-base 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:

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

Model Details

Model Description

Model Type: SetFit
Sentence Transformer body: firqaaa/indo-sentence-bert-base
Classification head: a LogisticRegression instance
spaCy Model: id_core_news_trf
SetFitABSA Aspect Model: firqaaa/indo-setfit-absa-bert-base-restaurants-aspect
SetFitABSA Polarity Model: firqaaa/indo-setfit-absa-bert-base-restaurants-polarity
Maximum Sequence Length: 512 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	'reservasi:restoran ini sangat kecil sehingga reservasi adalah suatu keharusan.' 'nyonya rumah:di sebelah kanan saya, nyo rumah berdiri di dekat seorang busboy dan mendesiskan rapido, rapido ketika dia mencoba membersihkan dan mengatur ulang meja untuk enam orang nyonya rumah' 'busboy:di sebelah kanan saya, nyo rumah berdiri di dekat seorang busboy dan mendesiskan rapido, rapido ketika dia mencoba membersihkan dan mengatur ulang meja untuk enam orang.'
no aspect	'restoran:restoran ini sangat kecil sehingga reservasi adalah suatu keharusan.' 'keharusan:restoran ini sangat kecil sehingga reservasi adalah suatu keharusan.' 'sebelah kanan:di sebelah kanan saya, nyo rumah berdiri di dekat seorang busboy dan mendesiskan rapido, rapido ketika dia mencoba membersihkan dan mengatur ulang meja untuk enam orang nyonya rumah'

Label

Examples

aspect

'reservasi:restoran ini sangat kecil sehingga reservasi adalah suatu keharusan.'
'nyonya rumah:di sebelah kanan saya, nyo rumah berdiri di dekat seorang busboy dan mendesiskan rapido, rapido ketika dia mencoba membersihkan dan mengatur ulang meja untuk enam orang nyonya rumah'
'busboy:di sebelah kanan saya, nyo rumah berdiri di dekat seorang busboy dan mendesiskan rapido, rapido ketika dia mencoba membersihkan dan mengatur ulang meja untuk enam orang.'

no aspect

'restoran:restoran ini sangat kecil sehingga reservasi adalah suatu keharusan.'
'keharusan:restoran ini sangat kecil sehingga reservasi adalah suatu keharusan.'
'sebelah kanan:di sebelah kanan saya, nyo rumah berdiri di dekat seorang busboy dan mendesiskan rapido, rapido ketika dia mencoba membersihkan dan mengatur ulang meja untuk enam orang nyonya rumah'

Evaluation

Metrics

Label	Accuracy
all	0.9087

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(
    "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	2	19.7819	59

Label	Training Sample Count
no aspect	2939
aspect	1468

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	0.3135	-
0.0001	50	0.3401	-
0.0001	100	0.3212	-
0.0002	150	0.3641	-
0.0003	200	0.3317	-
0.0004	250	0.2809	-
0.0004	300	0.2446	-
0.0005	350	0.284	-
0.0006	400	0.3257	-
0.0007	450	0.2996	-
0.0007	500	0.209	0.295
0.0008	550	0.2121	-
0.0009	600	0.2204	-
0.0010	650	0.3023	-
0.0010	700	0.3253	-
0.0011	750	0.233	-
0.0012	800	0.3131	-
0.0013	850	0.2873	-
0.0013	900	0.2028	-
0.0014	950	0.2608	-
0.0015	1000	0.2842	0.2696
0.0016	1050	0.2297	-
0.0016	1100	0.266	-
0.0017	1150	0.2771	-
0.0018	1200	0.2347	-
0.0019	1250	0.2539	-
0.0019	1300	0.3409	-
0.0020	1350	0.2925	-
0.0021	1400	0.2608	-
0.0021	1450	0.2792	-
0.0022	1500	0.261	0.2636
0.0023	1550	0.2596	-
0.0024	1600	0.2563	-
0.0024	1650	0.2329	-
0.0025	1700	0.2954	-
0.0026	1750	0.3329	-
0.0027	1800	0.2138	-
0.0027	1850	0.2591	-
0.0028	1900	0.268	-
0.0029	1950	0.2144	-
0.0030	2000	0.2361	0.2586
0.0030	2050	0.2322	-
0.0031	2100	0.2646	-
0.0032	2150	0.2018	-
0.0033	2200	0.2579	-
0.0033	2250	0.2501	-
0.0034	2300	0.2657	-
0.0035	2350	0.2272	-
0.0036	2400	0.2383	-
0.0036	2450	0.2615	-
0.0037	2500	0.2818	0.2554
0.0038	2550	0.2616	-
0.0039	2600	0.2225	-
0.0039	2650	0.2749	-
0.0040	2700	0.2572	-
0.0041	2750	0.2729	-
0.0041	2800	0.2559	-
0.0042	2850	0.2363	-
0.0043	2900	0.2518	-
0.0044	2950	0.1948	-
0.0044	3000	0.2842	0.2538
0.0045	3050	0.2243	-
0.0046	3100	0.2186	-
0.0047	3150	0.2829	-
0.0047	3200	0.2101	-
0.0048	3250	0.2156	-
0.0049	3300	0.2539	-
0.0050	3350	0.3005	-
0.0050	3400	0.2699	-
0.0051	3450	0.2431	-
0.0052	3500	0.2931	0.2515
0.0053	3550	0.2032	-
0.0053	3600	0.2451	-
0.0054	3650	0.2419	-
0.0055	3700	0.2267	-
0.0056	3750	0.2945	-
0.0056	3800	0.2689	-
0.0057	3850	0.2596	-
0.0058	3900	0.2978	-
0.0059	3950	0.2876	-
0.0059	4000	0.2484	0.2482
0.0060	4050	0.2698	-
0.0061	4100	0.2155	-
0.0061	4150	0.2474	-
0.0062	4200	0.2683	-
0.0063	4250	0.2979	-
0.0064	4300	0.2866	-
0.0064	4350	0.2604	-
0.0065	4400	0.1989	-
0.0066	4450	0.2708	-
0.0067	4500	0.2705	0.2407
0.0067	4550	0.2144	-
0.0068	4600	0.2503	-
0.0069	4650	0.2193	-
0.0070	4700	0.1796	-
0.0070	4750	0.2384	-
0.0071	4800	0.1933	-
0.0072	4850	0.2248	-
0.0073	4900	0.22	-
0.0073	4950	0.2052	-
0.0074	5000	0.2314	0.224
0.0075	5050	0.2279	-
0.0076	5100	0.2198	-
0.0076	5150	0.2332	-
0.0077	5200	0.1666	-
0.0078	5250	0.1949	-
0.0079	5300	0.1802	-
0.0079	5350	0.2496	-
0.0080	5400	0.2399	-
0.0081	5450	0.2042	-
0.0082	5500	0.1859	0.2077
0.0082	5550	0.2216	-
0.0083	5600	0.1227	-
0.0084	5650	0.2351	-
0.0084	5700	0.2735	-
0.0085	5750	0.1008	-
0.0086	5800	0.1568	-
0.0087	5850	0.1211	-
0.0087	5900	0.0903	-
0.0088	5950	0.1473	-
0.0089	6000	0.1167	0.1877
0.0090	6050	0.206	-
0.0090	6100	0.2392	-
0.0091	6150	0.116	-
0.0092	6200	0.1493	-
0.0093	6250	0.1373	-
0.0093	6300	0.1163	-
0.0094	6350	0.0669	-
0.0095	6400	0.0756	-
0.0096	6450	0.0788	-
0.0096	6500	0.1816	0.1838
0.0097	6550	0.1288	-
0.0098	6600	0.0946	-
0.0099	6650	0.1374	-
0.0099	6700	0.2167	-
0.0100	6750	0.0759	-
0.0101	6800	0.1543	-
0.0102	6850	0.0573	-
0.0102	6900	0.1169	-
0.0103	6950	0.0294	-
0.0104	7000	0.1241	0.1769
0.0104	7050	0.0803	-
0.0105	7100	0.0139	-
0.0106	7150	0.01	-
0.0107	7200	0.0502	-
0.0107	7250	0.0647	-
0.0108	7300	0.0117	-
0.0109	7350	0.0894	-
0.0110	7400	0.0101	-
0.0110	7450	0.0066	-
0.0111	7500	0.0347	0.1899
0.0112	7550	0.0893	-
0.0113	7600	0.0127	-
0.0113	7650	0.1285	-
0.0114	7700	0.0049	-
0.0115	7750	0.0571	-
0.0116	7800	0.0068	-
0.0116	7850	0.0586	-
0.0117	7900	0.0788	-
0.0118	7950	0.0655	-
0.0119	8000	0.0052	0.1807
0.0119	8050	0.0849	-
0.0120	8100	0.0133	-
0.0121	8150	0.0445	-
0.0122	8200	0.0118	-
0.0122	8250	0.0118	-
0.0123	8300	0.063	-
0.0124	8350	0.0751	-
0.0124	8400	0.058	-
0.0125	8450	0.002	-
0.0126	8500	0.0058	0.1804
0.0127	8550	0.0675	-
0.0127	8600	0.0067	-
0.0128	8650	0.0087	-
0.0129	8700	0.0028	-
0.0130	8750	0.0626	-
0.0130	8800	0.0563	-
0.0131	8850	0.0012	-
0.0132	8900	0.0067	-
0.0133	8950	0.0011	-
0.0133	9000	0.0105	0.189
0.0134	9050	0.101	-
0.0135	9100	0.1162	-
0.0136	9150	0.0593	-
0.0136	9200	0.0004	-
0.0137	9250	0.0012	-
0.0138	9300	0.0022	-
0.0139	9350	0.0033	-
0.0139	9400	0.0025	-
0.0140	9450	0.0578	-
0.0141	9500	0.0012	0.1967

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

@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}
}

firqaaa
/

setfit-indo-absa-restaurant-aspect