PaECTER - a Patent Similarity Model

PaECTER (Patent Embeddings using Citationinformed TransformERs) is a patent similarity model. Built upon Google's BERT for Patents as its base model, it generates 1024-dimensional dense vector embeddings from patent text. These vectors encapsulate the semantic essence of the given patent text, making it highly suitable for various downstream tasks related to patent analysis.

Paper: https://arxiv.org/pdf/2402.19411

Applications

• Semantic Search
• Prior Art Search
• Clustering
• Patent Landscaping

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('mpi-inno-comp/paecter')
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('mpi-inno-comp/paecter')
model = AutoModel.from_pretrained('mpi-inno-comp/paecter')

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt', max_length=512)

# 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, PaECTER: Patent-level Representation Learning using Citation-informed Transformers

Training

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 318750 with parameters:

{'batch_size': 4, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.CustomTripletLoss.CustomTripletLoss with parameters:

{'distance_metric': 'TripletDistanceMetric.EUCLIDEAN', 'triplet_margin': 1}

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 4000,
    "evaluator": "sentence_transformers.evaluation.TripletEvaluator.TripletEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 1e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 31875.0,
    "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': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
)

Citing & Authors

@misc{ghosh2024paecter,
      title={PaECTER: Patent-level Representation Learning using Citation-informed Transformers}, 
      author={Mainak Ghosh and Sebastian Erhardt and Michael E. Rose and Erik Buunk and Dietmar Harhoff},
      year={2024},
      eprint={2402.19411},
      archivePrefix={arXiv},
      primaryClass={cs.IR}
}