model update

5ba8334 over 1 year ago

No virus

9.4 kB

	---
	datasets:
	- relbert/nell_relational_similarity
	model-index:
	- name: relbert/relbert-roberta-large-nce-a-nell
	results:
	- task:
	name: Relation Mapping
	type: sorting-task
	dataset:
	name: Relation Mapping
	args: relbert/relation_mapping
	type: relation-mapping
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.8019047619047619
	- task:
	name: Analogy Questions (SAT full)
	type: multiple-choice-qa
	dataset:
	name: SAT full
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.4411764705882353
	- task:
	name: Analogy Questions (SAT)
	type: multiple-choice-qa
	dataset:
	name: SAT
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.45103857566765576
	- task:
	name: Analogy Questions (BATS)
	type: multiple-choice-qa
	dataset:
	name: BATS
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.462479155086159
	- task:
	name: Analogy Questions (Google)
	type: multiple-choice-qa
	dataset:
	name: Google
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.75
	- task:
	name: Analogy Questions (U2)
	type: multiple-choice-qa
	dataset:
	name: U2
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.37280701754385964
	- task:
	name: Analogy Questions (U4)
	type: multiple-choice-qa
	dataset:
	name: U4
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.43287037037037035
	- task:
	name: Analogy Questions (ConceptNet Analogy)
	type: multiple-choice-qa
	dataset:
	name: ConceptNet Analogy
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.16526845637583892
	- task:
	name: Analogy Questions (TREX Analogy)
	type: multiple-choice-qa
	dataset:
	name: TREX Analogy
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.73224043715847
	- task:
	name: Analogy Questions (NELL-ONE Analogy)
	type: multiple-choice-qa
	dataset:
	name: NELL-ONE Analogy
	args: relbert/analogy_questions
	type: analogy-questions
	metrics:
	- name: Accuracy
	type: accuracy
	value: 0.8416666666666667
	- task:
	name: Lexical Relation Classification (BLESS)
	type: classification
	dataset:
	name: BLESS
	args: relbert/lexical_relation_classification
	type: relation-classification
	metrics:
	- name: F1
	type: f1
	value: 0.9028175380442971
	- name: F1 (macro)
	type: f1_macro
	value: 0.901815241446287
	- task:
	name: Lexical Relation Classification (CogALexV)
	type: classification
	dataset:
	name: CogALexV
	args: relbert/lexical_relation_classification
	type: relation-classification
	metrics:
	- name: F1
	type: f1
	value: 0.8288732394366197
	- name: F1 (macro)
	type: f1_macro
	value: 0.6346957553982132
	- task:
	name: Lexical Relation Classification (EVALution)
	type: classification
	dataset:
	name: BLESS
	args: relbert/lexical_relation_classification
	type: relation-classification
	metrics:
	- name: F1
	type: f1
	value: 0.6289274106175514
	- name: F1 (macro)
	type: f1_macro
	value: 0.6198817150706762
	- task:
	name: Lexical Relation Classification (K&H+N)
	type: classification
	dataset:
	name: K&H+N
	args: relbert/lexical_relation_classification
	type: relation-classification
	metrics:
	- name: F1
	type: f1
	value: 0.9608402309243931
	- name: F1 (macro)
	type: f1_macro
	value: 0.8839063766078612
	- task:
	name: Lexical Relation Classification (ROOT09)
	type: classification
	dataset:
	name: ROOT09
	args: relbert/lexical_relation_classification
	type: relation-classification
	metrics:
	- name: F1
	type: f1
	value: 0.8843622688812285
	- name: F1 (macro)
	type: f1_macro
	value: 0.8844905148218848

	---
	# relbert/relbert-roberta-large-nce-a-nell

	RelBERT based on [roberta-large](https://huggingface.co/roberta-large) fine-tuned on [relbert/nell_relational_similarity](https://huggingface.co/datasets/relbert/nell_relational_similarity) (see the [`relbert`](https://github.com/asahi417/relbert) for more detail of fine-tuning).
	This model achieves the following results on the relation understanding tasks:
	- Analogy Question ([dataset](https://huggingface.co/datasets/relbert/analogy_questions), [full result](https://huggingface.co/relbert/relbert-roberta-large-nce-a-nell/raw/main/analogy.forward.json)):
	- Accuracy on SAT (full): 0.4411764705882353
	- Accuracy on SAT: 0.45103857566765576
	- Accuracy on BATS: 0.462479155086159
	- Accuracy on U2: 0.37280701754385964
	- Accuracy on U4: 0.43287037037037035
	- Accuracy on Google: 0.75
	- Accuracy on ConceptNet Analogy: 0.16526845637583892
	- Accuracy on T-Rex Analogy: 0.73224043715847
	- Accuracy on NELL-ONE Analogy: 0.8416666666666667
	- Lexical Relation Classification ([dataset](https://huggingface.co/datasets/relbert/lexical_relation_classification), [full result](https://huggingface.co/relbert/relbert-roberta-large-nce-a-nell/raw/main/classification.json)):
	- Micro F1 score on BLESS: 0.9028175380442971
	- Micro F1 score on CogALexV: 0.8288732394366197
	- Micro F1 score on EVALution: 0.6289274106175514
	- Micro F1 score on K&H+N: 0.9608402309243931
	- Micro F1 score on ROOT09: 0.8843622688812285
	- Relation Mapping ([dataset](https://huggingface.co/datasets/relbert/relation_mapping), [full result](https://huggingface.co/relbert/relbert-roberta-large-nce-a-nell/raw/main/relation_mapping.json)):
	- Accuracy on Relation Mapping: 0.8019047619047619


	### Usage
	This model can be used through the [relbert library](https://github.com/asahi417/relbert). Install the library via pip
	```shell
	pip install relbert
	```
	and activate model as below.
	```python
	from relbert import RelBERT
	model = RelBERT("relbert/relbert-roberta-large-nce-a-nell")
	vector = model.get_embedding(['Tokyo', 'Japan']) # shape of (n_dim, )
	```

	### Training hyperparameters

	- model: roberta-large
	- max_length: 64
	- epoch: 10
	- batch: 31
	- random_seed: 0
	- lr: 5e-06
	- lr_warmup: 10
	- aggregation_mode: average_no_mask
	- data: relbert/nell_relational_similarity
	- data_name: None
	- exclude_relation: None
	- split: train
	- split_valid: validation
	- loss_function: nce
	- classification_loss: False
	- loss_function_config: {'temperature': 0.05, 'num_negative': 300, 'num_positive': 10}
	- augment_negative_by_positive: True

	See the full configuration at [config file](https://huggingface.co/relbert/relbert-roberta-large-nce-a-nell/raw/main/finetuning_config.json).

	### Reference
	If you use any resource from RelBERT, please consider to cite our [paper](https://aclanthology.org/2021.emnlp-main.712/).

	```

	@inproceedings{ushio-etal-2021-distilling,
	title = "Distilling Relation Embeddings from Pretrained Language Models",
	author = "Ushio, Asahi and
	Camacho-Collados, Jose and
	Schockaert, Steven",
	booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
	month = nov,
	year = "2021",
	address = "Online and Punta Cana, Dominican Republic",
	publisher = "Association for Computational Linguistics",
	url = "https://aclanthology.org/2021.emnlp-main.712",
	doi = "10.18653/v1/2021.emnlp-main.712",
	pages = "9044--9062",
	abstract = "Pre-trained language models have been found to capture a surprisingly rich amount of lexical knowledge, ranging from commonsense properties of everyday concepts to detailed factual knowledge about named entities. Among others, this makes it possible to distill high-quality word vectors from pre-trained language models. However, it is currently unclear to what extent it is possible to distill relation embeddings, i.e. vectors that characterize the relationship between two words. Such relation embeddings are appealing because they can, in principle, encode relational knowledge in a more fine-grained way than is possible with knowledge graphs. To obtain relation embeddings from a pre-trained language model, we encode word pairs using a (manually or automatically generated) prompt, and we fine-tune the language model such that relationally similar word pairs yield similar output vectors. We find that the resulting relation embeddings are highly competitive on analogy (unsupervised) and relation classification (supervised) benchmarks, even without any task-specific fine-tuning. Source code to reproduce our experimental results and the model checkpoints are available in the following repository: https://github.com/asahi417/relbert",
	}

	```