bltlab/queryner-augmented-data-bert-base-uncased

1---2license: cc-by-4.03datasets:4- bltlab/queryner5language:6- en7metrics:8- f19pipeline_tag: token-classification10inference:11  parameters:12    aggregation_strategy: "first"13---14 15# Model Card for Model ID16 17E-commerce query segmentation model in English.18This model is trained on QueryNER training dataset with the addition of augmentations so the model should be more robust to spelling mistakes and mentions unseen in the training data.19 20 21## Model Details22 23### Model Description24 25This is a token classification model using BERT base uncased as the base model.26The model is fine-tuned on the (QueryNER training dataset)[https://huggingface.co/datasets/bltlab/queryner] and augmented data as described in the QueryNER paper. 27 28 29- **Developed by:** [BLT Lab](https://github.com/bltlab) in collaboration with eBay.30- **Funded by:** eBay31- **Shared by:** (@cpalenmichel)[https://github.com/cpalenmichel]32- **Model type:** Token Classification / Sequence Labeling / Chunking33- **Language(s) (NLP):** English34- **License:** CC-BY 4.035- **Finetuned from model:** BERT base uncased36 37### Model Sources38 39Underlying model is based on [BERT base-uncased](https://huggingface.co/google-bert/bert-base-uncased). 40 41- **Repository:** [https://github.com/bltlab/query-ner](https://github.com/bltlab/query-ner)42- **Paper:** Accepted at LREC-COLING Coming soon43 44## Uses45 46### Direct Use47 48Intended use is research purposes and e-commerce query segmentation.49 50### Downstream Use51 52Potential downstream use cases include weighting entity spans, linking to knowledge bases, removing spans as a recovery strategy for null and low recall queries.53 54### Out-of-Scope Use55 56This model is trained only on the training data of the QueryNER dataset. It may not perform well on other domains without additional training data and further fine-tuning.57 58## Bias, Risks, and Limitations59 60See paper limitations section.61 62## How to Get Started with the Model63 64See huggingface tutorials for token classification and access the model using AutoModelForTokenClassification.65Note that we do some post processing to make use of only the first subtoken's tag unlike the inference API.66 67## Training Details68 69### Training Data70 71See paper for details.72 73 74### Training Procedure75 76See paper for details.77 78#### Training Hyperparameters79 80See paper for details.81 82 83## Evaluation84 85Evaluation details provided in the paper. 86Scoring was done using [SeqScore](https://github.com/bltlab/seqscore) using the conlleval repair method for invalid label transition sequences. 87 88### Testing Data, Factors & Metrics89 90#### Testing Data91 92QueryNER test set: [https://huggingface.co/datasets/bltlab/queryner](https://huggingface.co/datasets/bltlab/queryner)93 94 95#### Factors96Evaluation is reported with micro-F1 at the entity level on the QueryNER test set. 97We used conlleval repair method for invalid label transitions.98 99#### Metrics100We use micro-F1 at the entity level as this is fairly common practice for NER models.101 102### Results103 104[More Information Needed]105 106 107## Environmental Impact108Rough estimate109 110- **Hardware Type:** 1 RTX 3090 GPU 111- **Hours used:** < 2 hours112- **Cloud Provider:** Private113- **Compute Region:** northamerica-northeast1114- **Carbon Emitted:** 0.02115 116 117## Citation118 119Accepted at LREC-COLING coming soon120 121**BibTeX:**122 123Accepted at LREC-COLING coming soon124 125 126## Model Card Authors 127 128Chester Palen-Michel (@cpalenmichel)[https://github.com/cpalenmichel]129 130## Model Card Contact131 132Chester Palen-Michel (@cpalenmichel)[https://github.com/cpalenmichel]