This model is a fine-tuned version of distilbert-base-uncased trained on processed Amazon review datasets from multiple categories. It is designed for sentiment classification across five classes. It achieves the following results on the evaluation set:

Loss: 0.9056
Accuracy: 0.6087
F1 Macro: 0.6101
F1 Weighted: 0.6085
Precision Class 1: 0.6762
Precision Class 2: 0.4900
Precision Class 3: 0.5177
Precision Class 4: 0.6173
Precision Class 5: 0.7535
Recall Class 1: 0.6781
Recall Class 2: 0.5293
Recall Class 3: 0.5049
Recall Class 4: 0.5606
Recall Class 5: 0.7781
F1 Class 1: 0.6772
F1 Class 2: 0.5089
F1 Class 3: 0.5112
F1 Class 4: 0.5876
F1 Class 5: 0.7656

Model description

This model is a distilled, lightweight transformer optimized for multi-class sentiment analysis. It is capable of processing text efficiently while maintaining competitive performance. The model was fine-tuned on Amazon product reviews spanning different categories such as electronics and home goods... It can classify reviews into five sentiment levels, from highly negative to highly positive.

Intended uses & limitations

Automated sentiment classification for Amazon reviews or similar e-commerce datasets.

Analysis of customer feedback trends to inform product decisions.

Training and evaluation data

The model was trained on a processed subset of Amazon review datasets covering multiple product categories. Text data was cleaned, tokenized, and labeled into five sentiment classes. Evaluation was performed on a held-out test set representing the same categories to ensure balanced coverage.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 32
eval_batch_size: 64
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 64
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 4
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 Macro	F1 Weighted	Precision Class 1	Precision Class 2	Precision Class 3	Precision Class 4	Precision Class 5	Recall Class 1	Recall Class 2	Recall Class 3	Recall Class 4	Recall Class 5	F1 Class 1	F1 Class 2	F1 Class 3	F1 Class 4	F1 Class 5
1.0806	0.16	500	1.0584	0.5392	0.5333	0.5316	0.5485	0.4359	0.4347	0.5297	0.7385	0.7797	0.3506	0.4185	0.4576	0.6959	0.6440	0.3886	0.4264	0.4910	0.7166
0.9979	0.32	1000	0.9753	0.5727	0.5741	0.5726	0.7303	0.4461	0.4691	0.5915	0.7095	0.5433	0.5949	0.4908	0.4406	0.8016	0.6231	0.5099	0.4797	0.5050	0.7528
0.9773	0.48	1500	0.9464	0.5869	0.5832	0.5814	0.6321	0.5040	0.4675	0.5748	0.7381	0.7328	0.3800	0.5110	0.5416	0.7785	0.6787	0.4333	0.4883	0.5577	0.7578
0.9311	0.64	2000	0.9356	0.5876	0.5923	0.5909	0.7000	0.4546	0.5063	0.5627	0.8035	0.6272	0.6081	0.4236	0.6077	0.6754	0.6616	0.5203	0.4613	0.5843	0.7339
0.9332	0.8	2500	0.9305	0.5924	0.5832	0.5815	0.6279	0.4773	0.5509	0.5754	0.6969	0.7421	0.4949	0.3492	0.5432	0.8437	0.6802	0.4860	0.4274	0.5589	0.7633
0.9193	0.96	3000	0.9068	0.6037	0.6024	0.6008	0.6696	0.4849	0.5144	0.6016	0.7372	0.6956	0.5188	0.4514	0.5519	0.8097	0.6823	0.5013	0.4809	0.5757	0.7717
0.8735	1.12	3500	0.9177	0.6032	0.6029	0.6012	0.6766	0.5004	0.4973	0.5865	0.7445	0.6859	0.4791	0.4817	0.5848	0.7941	0.6812	0.4895	0.4894	0.5857	0.7685
0.8656	1.28	4000	0.9069	0.6042	0.6011	0.5996	0.6823	0.4792	0.5197	0.6217	0.7149	0.6830	0.5532	0.4448	0.4968	0.8526	0.6826	0.5136	0.4794	0.5523	0.7777
0.8667	1.44	4500	0.9034	0.6059	0.6051	0.6035	0.6859	0.4958	0.5050	0.5923	0.7384	0.6877	0.4941	0.4807	0.5640	0.8124	0.6868	0.4950	0.4925	0.5778	0.7736
0.879	1.6	5000	0.9001	0.6067	0.6086	0.6071	0.6592	0.4874	0.5043	0.6120	0.7918	0.7239	0.5192	0.4876	0.5703	0.7381	0.6901	0.5028	0.4958	0.5904	0.7640
0.8709	1.76	5500	0.9005	0.6039	0.6045	0.6029	0.7008	0.4802	0.5250	0.5652	0.7736	0.6591	0.5636	0.4078	0.6366	0.7603	0.6793	0.5186	0.4590	0.5988	0.7669
0.8565	1.92	6000	0.9076	0.6084	0.6053	0.6036	0.6409	0.5094	0.5011	0.6234	0.7513	0.7529	0.4305	0.5253	0.5368	0.8057	0.6924	0.4666	0.5129	0.5769	0.7776
0.794	2.08	6500	0.9080	0.6091	0.6103	0.6086	0.6960	0.4837	0.5061	0.6135	0.7610	0.6749	0.5568	0.4676	0.5521	0.8022	0.6853	0.5177	0.4861	0.5812	0.7810
0.7857	2.24	7000	0.9131	0.6058	0.6062	0.6046	0.6521	0.4940	0.5055	0.5987	0.7828	0.7342	0.4886	0.4795	0.5853	0.7482	0.6907	0.4913	0.4922	0.5919	0.7651
0.7956	2.4	7500	0.9092	0.6074	0.6105	0.6090	0.6897	0.4941	0.4919	0.6113	0.7813	0.6776	0.5317	0.5251	0.5475	0.7613	0.6836	0.5122	0.5080	0.5776	0.7712
0.7721	2.56	8000	0.9153	0.6057	0.6080	0.6065	0.6641	0.5053	0.4927	0.5936	0.7982	0.7196	0.4777	0.5280	0.5911	0.7177	0.6908	0.4911	0.5098	0.5924	0.7558
0.7974	2.7200	8500	0.9019	0.6095	0.6110	0.6094	0.6790	0.4909	0.5019	0.6212	0.7680	0.6981	0.5208	0.5058	0.5483	0.7819	0.6884	0.5054	0.5038	0.5825	0.7749
0.7884	2.88	9000	0.9119	0.6074	0.6054	0.6037	0.6467	0.5007	0.5033	0.6101	0.7592	0.7431	0.4593	0.4868	0.5608	0.7958	0.6915	0.4791	0.4949	0.5844	0.7771
0.7359	3.04	9500	0.9220	0.6066	0.6100	0.6085	0.6829	0.4916	0.4923	0.6094	0.7899	0.6889	0.5139	0.5286	0.5661	0.7414	0.6859	0.5025	0.5098	0.5869	0.7649
0.7453	3.2	10000	0.9298	0.6092	0.6064	0.6047	0.6534	0.4958	0.5115	0.6233	0.7431	0.7369	0.4930	0.4696	0.5315	0.8238	0.6927	0.4944	0.4897	0.5738	0.7814

Framework versions

Transformers 4.57.3
Pytorch 2.9.0+cu126
Datasets 4.0.0
Tokenizers 0.22.1

CheritiA/distilbert-sentiment-5class

checkpoints