File size: 11,444 Bytes
a12a6ae 522e605 a12a6ae 522e605 a12a6ae 522e605 a12a6ae 522e605 a12a6ae 0f55d8a a12a6ae 522e605 a12a6ae 522e605 a12a6ae 522e605 a12a6ae |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 |
---
library_name: span-marker
tags:
- span-marker
- token-classification
- ner
- named-entity-recognition
- generated_from_span_marker_trainer
datasets:
- imvladikon/nemo_corpus
metrics:
- precision
- recall
- f1
widget:
- text: >-
אלי ויזל, פרופסור ב אוניברסיטת בוסטון, ש סילבר התאמץ הרבה למען זכייתו ב פרס
נובל ל שלום, תמך בגלוי ב מועמדותו ל משרת ה מושל.
- text: >-
מאמרו של תום שגב, " ה קרב על סן סימון היה או לא היה " (" ה ארץ " 105), הגיע
ל ידי רק ב ימים אלה.
- text: >-
רק ב דבריו של ה רב אברהם טולדאנו, משגיח ב ישיבת ה רעיון ה יהודי ו מספר 4 ב
רשימת כך ל ה כנסת, היו כבר הוראות מעשיות: " אלוקים ייקום דמו ו אנו ניקום את
הוא.
- text: >-
מרכז ה מידע ל זכויות ה אדם ב ה שטחים, " בצלם ", מפרסם מ פעם ל פעם דפי מידע ו
ב המ פרטים על ה נעשה ב ה שטחים ב תחומים שונים.
- text: >-
גרוסבורד נהג לבדו ב ה מכונית, ב דרכו מ ה עיר מיניאפוליס ב אינדיאנה ל נמל ה
תעופה של היא.
pipeline_tag: token-classification
model-index:
- name: SpanMarker
results:
- task:
type: token-classification
name: Named Entity Recognition
dataset:
name: Unknown
type: imvladikon/nemo_corpus
split: test
metrics:
- type: f1
value: 0.7757111597374179
name: F1
- type: precision
value: 0.7912946428571429
name: Precision
- type: recall
value: 0.7607296137339056
name: Recall
language:
- he
---
# SpanMarker
This is a [SpanMarker](https://github.com/tomaarsen/SpanMarkerNER) model trained on the [imvladikon/nemo_corpus](https://huggingface.co/datasets/imvladikon/nemo_corpus) dataset that can be used for Named Entity Recognition.
## Model Details
### Model Description
- **Model Type:** SpanMarker
<!-- - **Encoder:** [Unknown](https://huggingface.co/unknown) -->
- **Maximum Sequence Length:** 512 tokens
- **Maximum Entity Length:** 100 words
- **Training Dataset:** [imvladikon/nemo_corpus](https://huggingface.co/datasets/imvladikon/nemo_corpus)
<!-- - **Language:** Unknown -->
<!-- - **License:** Unknown -->
### Model Sources
- **Repository:** [SpanMarker on GitHub](https://github.com/tomaarsen/SpanMarkerNER)
- **Thesis:** [SpanMarker For Named Entity Recognition](https://raw.githubusercontent.com/tomaarsen/SpanMarkerNER/main/thesis.pdf)
### Model Labels
| Label | Examples |
|:------|:------------------------------------------------|
| ANG | "יידיש", "אנגלית", "גרמנית" |
| DUC | "סובארו", "מרצדס", "דינמיט" |
| EVE | "מצדה", "הצהרת בלפור", "ה שואה" |
| FAC | "ברזילי", "תל - ה שומר", "כלא עזה" |
| GPE | "שפרעם", "רצועת עזה", "ה שטחים" |
| LOC | "חאן יונס", "גיבאליה", "שייח רדואן" |
| ORG | "ה ארץ", "מרחב ה גליל", "כך" |
| PER | "נימר חוסיין", "איברהים נימר חוסיין", "רמי רהב" |
| WOA | "ה ארץ", "קדיש", "קיטש ו מוות" |
## Evaluation
### Metrics
| Label | Precision | Recall | F1 |
|:--------|:----------|:-------|:-------|
| **all** | 0.7913 | 0.7607 | 0.7757 |
| ANG | 0.0 | 0.0 | 0.0 |
| DUC | 0.0 | 0.0 | 0.0 |
| FAC | 0.3571 | 0.4545 | 0.4 |
| GPE | 0.7817 | 0.7897 | 0.7857 |
| LOC | 0.5263 | 0.4878 | 0.5063 |
| ORG | 0.7854 | 0.7623 | 0.7736 |
| PER | 0.8725 | 0.8202 | 0.8456 |
| WOA | 0.0 | 0.0 | 0.0 |
## Uses
### Direct Use for Inference
```python
from span_marker import SpanMarkerModel
# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("iahlt/span-marker-xlm-roberta-base-nemo-mt-he")
# Run inference
entities = model.predict("גרוסבורד נהג לבדו ב ה מכונית, ב דרכו מ ה עיר מיניאפוליס ב אינדיאנה ל נמל ה תעופה של היא.")
```
## Training Details
### Training Set Metrics
| Training set | Min | Median | Max |
|:----------------------|:----|:--------|:----|
| Sentence length | 0 | 25.7252 | 117 |
| Entities per sentence | 0 | 1.2722 | 20 |
### Training Hyperparameters
- learning_rate: 1e-05
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 2
- mixed_precision_training: Native AMP
### Training Results
| Epoch | Step | Validation Loss | Validation Precision | Validation Recall | Validation F1 | Validation Accuracy |
|:------:|:----:|:---------------:|:--------------------:|:-----------------:|:-------------:|:-------------------:|
| 0.4393 | 1000 | 0.0083 | 0.7632 | 0.5812 | 0.6598 | 0.9477 |
| 0.8785 | 2000 | 0.0056 | 0.8366 | 0.6774 | 0.7486 | 0.9609 |
| 1.3178 | 3000 | 0.0052 | 0.8322 | 0.7655 | 0.7975 | 0.9714 |
| 1.7571 | 4000 | 0.0053 | 0.8008 | 0.7735 | 0.7870 | 0.9712 |
### Evaluation Results
| | precision | recall | f1 | number |
|:------------------------|------------:|------------:|------------:|-------------:|
| eval_loss | 0.00522302 | 0.00522302 | 0.00522302 | 0.00522302 |
| eval_ANG | 0 | 0 | 0 | 3 |
| eval_DUC | 0 | 0 | 0 | 2 |
| eval_EVE | 0 | 0 | 0 | 12 |
| eval_FAC | 0.333333 | 0.0833333 | 0.133333 | 12 |
| eval_GPE | 0.887931 | 0.85124 | 0.869198 | 121 |
| eval_LOC | 0.703704 | 0.678571 | 0.690909 | 28 |
| eval_ORG | 0.719298 | 0.689076 | 0.703863 | 119 |
| eval_PER | 0.889447 | 0.917098 | 0.903061 | 193 |
| eval_WOA | 0 | 0 | 0 | 9 |
| eval_overall_precision | 0.832244 | 0.832244 | 0.832244 | 0.832244 |
| eval_overall_recall | 0.765531 | 0.765531 | 0.765531 | 0.765531 |
| eval_overall_f1 | 0.797495 | 0.797495 | 0.797495 | 0.797495 |
| eval_overall_accuracy | 0.971418 | 0.971418 | 0.971418 | 0.971418 |
| eval_runtime | 34.3336 | 34.3336 | 34.3336 | 34.3336 |
| eval_samples_per_second | 23.505 | 23.505 | 23.505 | 23.505 |
| eval_steps_per_second | 11.767 | 11.767 | 11.767 | 11.767 |
| epoch | 2 | 2 | 2 | 2 |
### Tests Results
| | precision | recall | f1 | number |
|:------------------------|------------:|------------:|------------:|-------------:|
| test_loss | 0.00604774 | 0.00604774 | 0.00604774 | 0.00604774 |
| test_ANG | 0 | 0 | 0 | 1 |
| test_DUC | 0 | 0 | 0 | 3 |
| test_FAC | 0.357143 | 0.454545 | 0.4 | 11 |
| test_GPE | 0.781726 | 0.789744 | 0.785714 | 195 |
| test_LOC | 0.526316 | 0.487805 | 0.506329 | 41 |
| test_ORG | 0.785354 | 0.762255 | 0.773632 | 408 |
| test_PER | 0.87251 | 0.820225 | 0.84556 | 267 |
| test_WOA | 0 | 0 | 0 | 6 |
| test_overall_precision | 0.791295 | 0.791295 | 0.791295 | 0.791295 |
| test_overall_recall | 0.76073 | 0.76073 | 0.76073 | 0.76073 |
| test_overall_f1 | 0.775711 | 0.775711 | 0.775711 | 0.775711 |
| test_overall_accuracy | 0.964642 | 0.964642 | 0.964642 | 0.964642 |
| test_runtime | 49.5152 | 49.5152 | 49.5152 | 49.5152 |
| test_samples_per_second | 23.286 | 23.286 | 23.286 | 23.286 |
| test_steps_per_second | 11.653 | 11.653 | 11.653 | 11.653 |
| epoch | 2 | 2 | 2 | 2 |
### Framework Versions
- Python: 3.10.12
- SpanMarker: 1.5.0
- Transformers: 4.35.2
- PyTorch: 2.1.0+cu118
- Datasets: 2.15.0
- Tokenizers: 0.15.0
## Citation
```
@article{10.1162/tacl_a_00404,
author = {Bareket, Dan and Tsarfaty, Reut},
title = "{Neural Modeling for Named Entities and Morphology (NEMO2)}",
journal = {Transactions of the Association for Computational Linguistics},
volume = {9},
pages = {909-928},
year = {2021},
month = {09},
abstract = "{Named Entity Recognition (NER) is a fundamental NLP task, commonly formulated as classification over a sequence of tokens. Morphologically rich languages (MRLs) pose a challenge to this basic formulation, as the boundaries of named entities do not necessarily coincide with token boundaries, rather, they respect morphological boundaries. To address NER in MRLs we then need to answer two fundamental questions, namely, what are the basic units to be labeled, and how can these units be detected and classified in realistic settings (i.e., where no gold morphology is available). We empirically investigate these questions on a novel NER benchmark, with parallel token- level and morpheme-level NER annotations, which we develop for Modern Hebrew, a morphologically rich-and-ambiguous language. Our results show that explicitly modeling morphological boundaries leads to improved NER performance, and that a novel hybrid architecture, in which NER precedes and prunes morphological decomposition, greatly outperforms the standard pipeline, where morphological decomposition strictly precedes NER, setting a new performance bar for both Hebrew NER and Hebrew morphological decomposition tasks.}",
issn = {2307-387X},
doi = {10.1162/tacl_a_00404},
url = {https://doi.org/10.1162/tacl\_a\_00404},
eprint = {https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl\_a\_00404/1962472/tacl\_a\_00404.pdf},
}
```
<!--
## Glossary
*Clearly define terms in order to be accessible across audiences.*
-->
<!--
## Model Card Authors
*Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.*
-->
<!--
## Model Card Contact
*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*
--> |