pipeline_tag: sentence-similarity
language: en
tags:
- sentence-similarity
- transformers
- Education
- en
- bert
- sentence-transformers
- feature-extraction
- xnli
- stsb_multi_mt
datasets:
- xnli
- stsb_multi_mt
inokufu/bertheo-en
A sentence-transformers model fine-tuned on course sentences. It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.
Details
This model is based on the English bert-base-uncased pre-trained model [1, 2].
It was first fine-tuned on our learning object (LO) sentences dataset. This dataset consists of a sample of 500k sentences of course descriptions. We used standard parameter settings for fine-tuning as mentioned in the original BERT paper [2]. This allows the model to improve its performance on the target task (Masked Language Model) for domain-specific sentences.
It was then fine-tuned on a natural language inference task (XNLI) [3]. This task consists in training the model to recognize relations between sentences (contradiction, neutral, implication).
It was then fine-tuned on a text semantic similarity task (on STS data) [4]. This task consists in training the model to estimate the similarity between two sentences.
This fine-tuning process allows our model to have a semantic representation of words that is much better than the one proposed by the base model.
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 = ["Learn to code in python", "Become an expert in accounting"]
model = SentenceTransformer('inokufu/bert-base-uncased-xnli-sts-finetuned-education')
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 = ["Learn to code in python", "Become an expert in accounting"]
# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('inokufu/bert-base-uncased-xnli-sts-finetuned-education')
model = AutoModel.from_pretrained('inokufu/bert-base-uncased-xnli-sts-finetuned-education')
# 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
STS (en) score: 84.61%
Model Architecture
SentenceTransformer(
(0): Transformer({'max_seq_length': 512, 'do_lower_case': True}) 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})
)
References
[1] https://huggingface.co./bert-base-uncased
[2] https://arxiv.org/abs/1810.04805
[3] https://arxiv.org/abs/1809.05053
[4] https://huggingface.co./datasets/stsb_multi_mt