metadata

pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - feature-extraction
  - sentence-similarity
language:
  - es
dataset:
  - hackathon-pln-es/ESnli
widget:
  - text: >-
      A ver si nos tenemos que poner todos en huelga hasta cobrar lo que
      queramos.
  - text: >-
      La huelga es el método de lucha más eficaz para conseguir mejoras en el
      salario.
  - text: Tendremos que optar por hacer una huelga para cobrar lo que queremos.
  - text: Queda descartada la huelga aunque no cobremos lo que queramos.

bertin-roberta-base-finetuning-esnli

This is a sentence-transformers model trained on a collection of NLI tasks for Spanish. It maps sentences & paragraphs to a 512 dimensional dense vector space and can be used for tasks like clustering or semantic search.

Based around the siamese networks approach from this paper.

Usage (Sentence-Transformers)

Using this model becomes easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer
sentences = ["Este es un ejemplo", "Cada oración es transformada"]

model = SentenceTransformer('hackathon-pln-es/bertin-roberta-base-finetuning-esnli')
embeddings = model.encode(sentences)
print(embeddings)

Evaluation Results

Our model was evaluated on the task of Semantic Textual Similarity using the SemEval-2015 Task for Spanish. We measure

	BETO STS	BERTIN STS (this model)	Relative improvement
cosine_pearson	0.609803	0.670862	+10.01
cosine_spearman	0.528776	0.598593	+13.20
euclidean_pearson	0.590613	0.675257	+14.33
euclidean_spearman	0.526529	0.604656	+14.84
manhattan_pearson	0.589108	0.676706	+14.87
manhattan_spearman	0.525910	0.606461	+15.32
dot_pearson	0.544078	0.586429	+7.78
dot_spearman	0.460427	0.495614	+7.64

Training

The model was trained with the parameters:

Dataset

We used a collection of datasets of Natural Language Inference as training data:

ESXNLI, only the part in spanish
SNLI, automatically translated
MultiNLI, automatically translated

The whole dataset used is available here.

DataLoader:

sentence_transformers.datasets.NoDuplicatesDataLoader.NoDuplicatesDataLoader of length 1127 with parameters:

{'batch_size': 64}

Loss:

sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss with parameters:

{'scale': 20.0, 'similarity_fct': 'cos_sim'}

Parameters of the fit()-Method:

{
    "epochs": 20,
    "evaluation_steps": 0,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'transformers.optimization.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 1127,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 514, 'do_lower_case': False}) with Transformer model: RobertaModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
  (2): Dense({'in_features': 768, 'out_features': 512, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
)

Authors