metadata

base_model: BAAI/bge-large-en-v1.5
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:1024
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: >-
      After rescue, survivors may require hospital treatment. This must be
      provided as quickly as possible. The SMC should consider having ambulance
      and hospital facilities ready.
    sentences:
      - What should the SMC consider having ready after a rescue?
      - What is critical for mass rescue operations?
      - >-
        What can computer programs do to relieve the search planner of
        computational burden?
  - source_sentence: >-
      SMCs conduct communication searches when facts are needed to supplement
      initially reported information. Efforts are continued to contact the
      craft, to find out more about a possible distress situation, and to
      prepare for or to avoid a search effort. Section 3.5 has more information
      on communication searches.MEDICO Communications
    sentences:
      - >-
        What is generally produced by dead-reckoning navigation alone for search
        aircraft?
      - >-
        What should be the widths of rectangular areas to be covered with a PS
        pattern and the lengths of rectangular areas to be covered with a CS
        pattern?
      - What is the purpose of SMCs conducting communication searches?
  - source_sentence: >-
      SAR facilities include designated SRUs and other resources which can be
      used to conduct or support SAR operations. An SRU is a unit composed of
      trained personnel and provided with equipment suitable for the expeditious
      and efficient conduct of search and rescue. An SRU can be an air,
      maritime, or land-based facility. Facilities selected as SRUs should be
      able to reach the scene of distress quickly and, in particular, be
      suitable for one or more of the following operations:– providing
      assistance to prevent or reduce the severity of accidents and the hardship
      of survivors, e.g., escorting an aircraft, standing by a sinking vessel;–
      conducting a search;– delivering supplies and survival equipment to the
      scene;– rescuing survivors;– providing food, medical or other initial
      needs of survivors; and– delivering the survivors to a place of safety. 
    sentences:
      - >-
        What are the types of SAR facilities that can be used to conduct or
        support SAR operations?
      - >-
        What is the scenario in which a simulated communication search is
        carried out and an air search is planned?
      - What is discussed in detail in various other places in this Manual?
  - source_sentence: >-
      Support facilities enable the operational response resources (e.g., the
      RCC and SRUs) to provide the SAR services. Without the supporting
      resources, the operational resources cannot sustain effective operations.
      There is a wide range of support facilities and services, which include
      the following:Training facilities Facility maintenanceCommunications
      facilities Management functionsNavigation systems Research and
      developmentSAR data providers (SDPs) PlanningMedical facilities
      ExercisesAircraft landing fields Refuelling servicesVoluntary services
      (e.g., Red Cross) Critical incident stress counsellors Computer resources
    sentences:
      - How many ways are there to train SAR specialists and teams?
      - What types of support facilities are mentioned in the context?
      - What is the duration of a prolonged blast?
  - source_sentence: >-
      Sound funding decisions arise out of accurate assessments made of the SAR
      system. To measure the performance or effectiveness of a SAR system
      usually requires collecting information or statistics and establishing
      agreed-upon goals. All pertinent information should be collected,
      including where the system failed to perform as it should have; failures
      and successes provide valuable information in assessing effectiveness and
      determining means to improve. 
    sentences:
      - >-
        What is required to measure the performance or effectiveness of a SAR
        system?
      - What is the purpose of having an SRR?
      - >-
        What is the effect of decreasing track spacing on the area that can be
        searched?
model-index:
  - name: SentenceTransformer based on BAAI/bge-large-en-v1.5
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 768
          type: dim_768
        metrics:
          - type: cosine_accuracy@1
            value: 0.7719298245614035
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.9298245614035088
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.956140350877193
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 1
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7719298245614035
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3099415204678363
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1912280701754386
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.1
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7719298245614035
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.9298245614035088
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.956140350877193
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 1
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8884520476480379
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.8524470899470901
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.85244708994709
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 512
          type: dim_512
        metrics:
          - type: cosine_accuracy@1
            value: 0.7543859649122807
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.9122807017543859
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.956140350877193
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9912280701754386
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7543859649122807
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.304093567251462
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1912280701754386
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09912280701754386
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7543859649122807
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.9122807017543859
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.956140350877193
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9912280701754386
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8791120820747885
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.8425438596491228
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.8431704260651629
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.7456140350877193
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8947368421052632
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.9385964912280702
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9649122807017544
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7456140350877193
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2982456140350877
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.18771929824561406
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09649122807017543
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7456140350877193
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8947368421052632
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.9385964912280702
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9649122807017544
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8623224236283672
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.8287628794207742
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.8310819942011893
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.7017543859649122
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8245614035087719
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8771929824561403
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9385964912280702
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7017543859649122
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.27485380116959063
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17543859649122803
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09385964912280703
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7017543859649122
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8245614035087719
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8771929824561403
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9385964912280702
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8146917044508328
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7757031467557786
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7788889950899075
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.6228070175438597
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.7543859649122807
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.7894736842105263
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8596491228070176
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6228070175438597
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.25146198830409355
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.15789473684210523
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08596491228070174
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6228070175438597
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.7543859649122807
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.7894736842105263
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8596491228070176
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7406737402395112
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.703104984683932
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.71092932980045
            name: Cosine Map@100

SentenceTransformer based on BAAI/bge-large-en-v1.5

This is a sentence-transformers model finetuned from BAAI/bge-large-en-v1.5 on the json dataset. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: BAAI/bge-large-en-v1.5
Maximum Sequence Length: 512 tokens
Output Dimensionality: 1024 tokens
Similarity Function: Cosine Similarity
Training Dataset:
- json

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("tessimago/bge-large-repmus-cross_entropy")
# Run inference
sentences = [
    'Sound funding decisions arise out of accurate assessments made of the SAR system. To measure the performance or effectiveness of a SAR system usually requires collecting information or statistics and establishing agreed-upon goals. All pertinent information should be collected, including where the system failed to perform as it should have; failures and successes provide valuable information in assessing effectiveness and determining means to improve. ',
    'What is required to measure the performance or effectiveness of a SAR system?',
    'What is the effect of decreasing track spacing on the area that can be searched?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

Dataset: dim_768
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7719
cosine_accuracy@3	0.9298
cosine_accuracy@5	0.9561
cosine_accuracy@10	1.0
cosine_precision@1	0.7719
cosine_precision@3	0.3099
cosine_precision@5	0.1912
cosine_precision@10	0.1
cosine_recall@1	0.7719
cosine_recall@3	0.9298
cosine_recall@5	0.9561
cosine_recall@10	1.0
cosine_ndcg@10	0.8885
cosine_mrr@10	0.8524
cosine_map@100	0.8524

Information Retrieval

Dataset: dim_512
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7544
cosine_accuracy@3	0.9123
cosine_accuracy@5	0.9561
cosine_accuracy@10	0.9912
cosine_precision@1	0.7544
cosine_precision@3	0.3041
cosine_precision@5	0.1912
cosine_precision@10	0.0991
cosine_recall@1	0.7544
cosine_recall@3	0.9123
cosine_recall@5	0.9561
cosine_recall@10	0.9912
cosine_ndcg@10	0.8791
cosine_mrr@10	0.8425
cosine_map@100	0.8432

Information Retrieval

Dataset: dim_256
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7456
cosine_accuracy@3	0.8947
cosine_accuracy@5	0.9386
cosine_accuracy@10	0.9649
cosine_precision@1	0.7456
cosine_precision@3	0.2982
cosine_precision@5	0.1877
cosine_precision@10	0.0965
cosine_recall@1	0.7456
cosine_recall@3	0.8947
cosine_recall@5	0.9386
cosine_recall@10	0.9649
cosine_ndcg@10	0.8623
cosine_mrr@10	0.8288
cosine_map@100	0.8311

Information Retrieval

Dataset: dim_128
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7018
cosine_accuracy@3	0.8246
cosine_accuracy@5	0.8772
cosine_accuracy@10	0.9386
cosine_precision@1	0.7018
cosine_precision@3	0.2749
cosine_precision@5	0.1754
cosine_precision@10	0.0939
cosine_recall@1	0.7018
cosine_recall@3	0.8246
cosine_recall@5	0.8772
cosine_recall@10	0.9386
cosine_ndcg@10	0.8147
cosine_mrr@10	0.7757
cosine_map@100	0.7789

Information Retrieval

Dataset: dim_64
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6228
cosine_accuracy@3	0.7544
cosine_accuracy@5	0.7895
cosine_accuracy@10	0.8596
cosine_precision@1	0.6228
cosine_precision@3	0.2515
cosine_precision@5	0.1579
cosine_precision@10	0.086
cosine_recall@1	0.6228
cosine_recall@3	0.7544
cosine_recall@5	0.7895
cosine_recall@10	0.8596
cosine_ndcg@10	0.7407
cosine_mrr@10	0.7031
cosine_map@100	0.7109

Training Details

Training Dataset

json

Dataset: json
Size: 1,024 training samples
Columns: positive and anchor
Approximate statistics based on the first 1000 samples:
positive anchor
type string string
details
min: 10 tokens
mean: 133.58 tokens
max: 512 tokens

min: 7 tokens
mean: 17.7 tokens
max: 39 tokens

	positive	anchor
type	string	string
details	min: 10 tokens mean: 133.58 tokens max: 512 tokens	min: 7 tokens mean: 17.7 tokens max: 39 tokens

Samples:

positive	anchor
The debriefing helps to ensure that all survivors are rescued, to attend to the physical welfare of each survivor, and to obtain information which may assist and improve SAR services. Proper debriefing techniques include:– due care to avoid worsening a survivor’s condition by excessive debriefing;– careful assessment of the survivor’s statements if the survivor is frightened or excited;– use of a calm voice in questioning;– avoidance of suggesting the answers when obtaining facts; and– explaining that the information requested is important for the success of the SAR operation, and possibly for future SAR operations.	`What are some proper debriefing techniques used in SAR services?`
`Communicating with passengers is more difficult in remote areas where phone service may be inadequate or lacking. If phones do exist, calling the airline or shipping company may be the best way to check in and find out information. In more populated areas, local agencies may have an emergency evacuation plan or other useful plan that can be implemented.IE961E.indb 21 6/28/2013 10:29:55 AM`	`What is a good way to check in and find out information in remote areas where phone service may be inadequate or lacking?`
Voice communication is the basis of telemedical advice. It allows free dialogue and contributes to the human relationship, which is crucial to any medical consultation. Text messages are a useful complement to the voice telemedical advice and add the reliability of writing. Facsimile allows the exchange of pictures or diagrams, which help to identify a symptom, describe a lesion or the method of treatment. Digital data transmissions (photographs or electrocardiogram) provide an objective and potentially crucial addition to descriptive and subjective clinical data.	`What are the types of communication methods used in telemedical advice?`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 32
per_device_eval_batch_size: 16
gradient_accumulation_steps: 16
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
tf32: True
load_best_model_at_end: True
optim: adamw_torch_fused

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 32
per_device_eval_batch_size: 16
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 16
eval_accumulation_steps: None
learning_rate: 2e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 4
max_steps: -1
lr_scheduler_type: cosine
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
warmup_steps: 0
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
save_safetensors: True
save_on_each_node: False
save_only_model: False
restore_callback_states_from_checkpoint: False
no_cuda: False
use_cpu: False
use_mps_device: False
seed: 42
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: True
fp16: False
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: True
local_rank: 0
ddp_backend: None
tpu_num_cores: None
tpu_metrics_debug: False
debug: []
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_prefetch_factor: None
past_index: -1
disable_tqdm: False
remove_unused_columns: True
label_names: None
load_best_model_at_end: True
ignore_data_skip: False
fsdp: []
fsdp_min_num_params: 0
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
fsdp_transformer_layer_cls_to_wrap: None
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: True
dataloader_persistent_workers: False
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: False
resume_from_checkpoint: None
hub_model_id: None
hub_strategy: every_save
hub_private_repo: False
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_64_cosine_map@100	dim_768_cosine_map@100
1.0	2	0.7770	0.8173	0.8316	0.6838	0.8448
2.0	4	0.7858	0.8221	0.8326	0.6993	0.8478
3.0	6	0.7801	0.8297	0.8412	0.7101	0.8517
4.0	8	0.7789	0.8311	0.8432	0.7109	0.8524

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.14
Sentence Transformers: 3.1.0
Transformers: 4.41.2
PyTorch: 2.1.2+cu121
Accelerate: 0.34.2
Datasets: 2.19.1
Tokenizers: 0.19.1

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}