graziasveva93/sbert-adapt-ub

Patent SBERT-adapt-ub

Patent SBERT-adapt-ub is a domain-adapted SentenceTransformer model for computing semantic similarity between patent texts. It maps patent abstracts, claims, and other patent-related text into a 384-dimensional dense vector space, enabling downstream tasks such as patent-to-patent similarity, semantic search, clustering, technology mapping, and patent landscaping.

The model is based on sentence-transformers/all-MiniLM-L6-v2 and was further adapted to the patent domain using a triplet-learning setup. It is designed to produce sentence-level and paragraph-level embeddings that are more suitable for patent similarity analysis than generic sentence embeddings.

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 = ["This is an example sentence", "Each sentence is converted"]

model = SentenceTransformer('graziasveva93/sbert-adapt-ub')
embeddings = model.encode(sentences)
print(embeddings)

Usage (HuggingFace Transformers)

Without sentence-transformers, you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.

from transformers import AutoTokenizer, AutoModel
import torch


#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)


# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('{MODEL_NAME}')
model = AutoModel.from_pretrained('{MODEL_NAME}')

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

Evaluation Results

Evaluation of this model is available in our paper: https://arxiv.org/pdf/2403.16630

Training

The model was trained using the SentenceTransformers framework with a triplet-learning objective.

DataLoader:

torch.utils.data.dataloader.DataLoader of length 13,456 with parameters:

{
    "batch_size": 8,
    "sampler": "torch.utils.data.sampler.RandomSampler",
    "batch_sampler": "torch.utils.data.sampler.BatchSampler"
}

Loss:

sentence_transformers.losses.TripletLoss.TripletLoss with parameters:

{
    "distance_metric": "TripletDistanceMetric.EUCLIDEAN",
    "triplet_margin": 5
}

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 1000,
    "evaluator": "sentence_transformers.evaluation.TripletEvaluator.TripletEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 1345,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel
  (1): Pooling({
      'word_embedding_dimension': 384,
      'pooling_mode_cls_token': False,
      'pooling_mode_mean_tokens': True,
      'pooling_mode_max_tokens': False,
      'pooling_mode_mean_sqrt_len_tokens': False
  })
)

The model uses sentence-transformers/all-MiniLM-L6-v2 as the base model. The underlying transformer is a BERT-style architecture with hidden size 384, 6 hidden layers, 12 attention heads, and a maximum sequence length of 512 tokens.

Citing & Authors

If you use this model, please cite the associated paper:

@misc{ascione2024comparative,
  title={A comparative analysis of embedding models for patent similarity},
  author={Ascione, Grazia Sveva and Sterzi, Valerio},
  year={2024},
  eprint={2403.16630},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

Authors

• Grazia Sveva Ascione, Bordeaux School of Economics, Université de Bordeaux