language:

• pt thumbnail: "葡萄牙法律领域BERT模型" pipeline_tag: sentence-similarity tags:
• sentence-transformers
• sentence-similarity
• transformers datasets:
• assin
• assin2
• rufimelo/PortugueseLegalSentences-v0 widget:
• source_sentence: "律师向法官提交了证据。" sentences:
  • "法官阅读了证据。"
  • "法官阅读了上诉书。"
  • "法官扔了一块石头。" example_title: "示例1" model-index:
• name: BERTimbau results:
  • task: name: STS type: STS metrics:
    • name: 皮尔逊相关系数 - assin数据集 type: Pearson Correlation value: 0.71457
    • name: 皮尔逊相关系数 - assin2数据集 type: Pearson Correlation value: 0.73545
    • name: 皮尔逊相关系数 - stsb_multi_mt pt数据集 type: Pearson Correlation value: 0.72383

rufimelo/Legal-BERTimbau-sts-base

这是一个sentence-transformers模型：它能将句子和段落映射到1024维的密集向量空间，可用于聚类或语义搜索等任务。
rufimelo/Legal-BERTimbau-sts-base基于Legal-BERTimbau-large模型，后者源自BERTimbau基础版本。
该模型针对葡萄牙语法律领域进行了适配，并在葡萄牙语数据集上进行了STS（语义文本相似度）训练。

使用方法（Sentence-Transformers）

安装sentence-transformers后即可轻松使用：

pip install -U sentence-transformers

使用示例：

from sentence_transformers import SentenceTransformer
sentences = ["这是一个示例", "这是另一个示例"]

model = SentenceTransformer('rufimelo/Legal-BERTimbau-sts-base')
embeddings = model.encode(sentences)
print(embeddings)

使用方法（HuggingFace Transformers）

from transformers import AutoTokenizer, AutoModel
import torch

# 均值池化 - 考虑注意力掩码进行正确平均
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0]  # 模型输出的第一个元素包含所有标记嵌入
    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)

# 待编码句子
sentences = ['这是一个示例句子', '每个句子都会被转换']

# 从HuggingFace Hub加载模型
tokenizer = AutoTokenizer.from_pretrained('rufimelo/Legal-BERTimbau-sts-base')
model = AutoModel.from_pretrained('rufimelo/Legal-BERTimbau-sts-base')

# 标记化句子
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# 计算标记嵌入
with torch.no_grad():
    model_output = model(**encoded_input)
# 执行池化（此处使用均值池化）
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
print("句子嵌入：")
print(sentence_embeddings)

STS评估结果

训练过程

rufimelo/Legal-BERTimbau-sts-base基于Legal-BERTimbau-large模型（源自BERTimbau基础版），通过assin和assin2数据集进行语义文本相似度的微调训练。

完整模型架构

SentenceTransformer(
  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel 
  (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, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
)

引用与作者

如果使用本作品，请引用：

@inproceedings{souza2020bertimbau,
  author    = {F{\'a}bio Souza and
               Rodrigo Nogueira and
               Roberto Lotufo},
  title     = {{BERT}imbau: pretrained {BERT} models for {B}razilian {P}ortuguese},
  booktitle = {9th Brazilian Conference on Intelligent Systems, {BRACIS}, Rio Grande do Sul, Brazil, October 20-23 (to appear)},
  year      = {2020}
}

@inproceedings{fonseca2016assin,
  title={ASSIN: Avaliacao de similaridade semantica e inferencia textual},
  author={Fonseca, E and Santos, L and Criscuolo, Marcelo and Aluisio, S},
  booktitle={Computational Processing of the Portuguese Language-12th International Conference, Tomar, Portugal},
  pages={13--15},
  year={2016}
}

@inproceedings{real2020assin,
  title={The assin 2 shared task: a quick overview},
  author={Real, Livy and Fonseca, Erick and Oliveira, Hugo Goncalo},
  booktitle={International Conference on Computational Processing of the Portuguese Language},
  pages={406--412},
  year={2020},
  organization={Springer}
}
@InProceedings{huggingface:dataset:stsb_multi_mt,
title = {Machine translated multilingual STS benchmark dataset.},
author={Philip May},
year={2021},
url={https://github.com/PhilipMay/stsb-multi-mt}
}