mdeberta-v3-base-name-classifier-v2

This model is a fine-tuned version of microsoft/mdeberta-v3-base on ele-sage/person-company-names-classification.

It achieves the following results on the evaluation set:

• Loss: 0.0732
• Accuracy: 0.9946
• Precision: 0.9989
• Recall: 0.9913
• F1: 0.9951

Model description

This model is a high-performance binary text classifier, fine-tuned from mdeberta-v3-base. Its purpose is to distinguish between a person's name and a company/organization name with high accuracy.

Direct Use

This model is intended to be used for text classification. Given a string, it will return a label indicating whether the string is a Person or a Company.

from transformers import pipeline

classifier = pipeline("text-classification", model="ele-sage/mdeberta-v3-base-name-classifier-v2")

results = classifier([
    "Satya Nadella",
    "Global Innovations Inc.",
    "Martinez, Alonso"
])

for result in results:
    print(f"Text: '{result['text']}', Prediction: {result['label']}, Score: {result['score']:.4f}")

Downstream Use

This model is a key component of a two-stage name processing pipeline. It is designed to be used as a fast, efficient "gatekeeper" to first identify person names before passing them to a more complex parsing model, such as ele-sage/distilbert-base-uncased-name-splitter.

Out-of-Scope Use

• This model is not a general-purpose classifier. It is highly specialized for distinguishing persons from companies and will not perform well on other classification tasks (e.g., sentiment analysis).

Bias, Risks, and Limitations

• Geographic & Cultural Bias: The training data is heavily biased towards North American (Canadian) person names and Quebec-based company names. The model will be less accurate when classifying names from other cultural or geographic origins.
• Ambiguity: Certain names can legitimately be both a person's name and a company's name (e.g., "Ford"). In these cases, the model makes a statistical guess based on its training data, which may not always align with the specific context.
• Data Source: The person name data is derived from a Facebook data leak and contains noise. While a rigorous cleaning process was applied, the model may have learned from some spurious data.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 128
• eval_batch_size: 512
• seed: 42
• 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: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 1
• label_smoothing_factor: 0.02

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
0.0914	0.0359	2000	0.0889	0.9907	0.9952	0.9882	0.9917
0.0796	0.0718	4000	0.0864	0.9907	0.9991	0.9843	0.9916
0.0808	0.1077	6000	0.0809	0.9919	0.9944	0.9910	0.9927
0.0828	0.1436	8000	0.0774	0.9930	0.9976	0.9899	0.9937
0.0787	0.1795	10000	0.0771	0.9931	0.9989	0.9886	0.9938
0.0761	0.2154	12000	0.0774	0.9935	0.9984	0.9899	0.9942
0.0779	0.2513	14000	0.0771	0.9935	0.9991	0.9892	0.9941
0.0833	0.2872	16000	0.0751	0.9937	0.9985	0.9903	0.9944
0.0812	0.3231	18000	0.0764	0.9935	0.9967	0.9915	0.9941
0.0763	0.3590	20000	0.0753	0.9940	0.9990	0.9902	0.9946
0.0753	0.3949	22000	0.0759	0.9936	0.9968	0.9917	0.9942
0.0749	0.4308	24000	0.0750	0.9940	0.9980	0.9912	0.9946
0.0755	0.4667	26000	0.0746	0.9939	0.9974	0.9917	0.9945
0.0755	0.5026	28000	0.0756	0.9937	0.9967	0.9919	0.9943
0.0753	0.5385	30000	0.0745	0.9942	0.9979	0.9916	0.9948
0.0791	0.5744	32000	0.0735	0.9943	0.9991	0.9908	0.9949
0.0789	0.6103	34000	0.0743	0.9939	0.9972	0.9918	0.9945
0.073	0.6462	36000	0.0741	0.9943	0.9985	0.9913	0.9949
0.0714	0.6821	38000	0.0738	0.9944	0.9989	0.9911	0.9950
0.0738	0.7180	40000	0.0733	0.9945	0.9989	0.9912	0.9950
0.0796	0.7539	42000	0.0732	0.9945	0.9987	0.9915	0.9951
0.0726	0.7898	44000	0.0734	0.9945	0.9988	0.9914	0.9951
0.0778	0.8257	46000	0.0733	0.9945	0.9988	0.9913	0.9951
0.0734	0.8616	48000	0.0733	0.9945	0.9989	0.9914	0.9951
0.0735	0.8975	50000	0.0732	0.9945	0.9988	0.9914	0.9951
0.0696	0.9334	52000	0.0732	0.9945	0.9989	0.9913	0.9951
0.0754	0.9693	54000	0.0732	0.9946	0.9989	0.9913	0.9951

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu128
• Datasets 4.4.1
• Tokenizers 0.22.1