InstaFoodRoBERTa-NER

Model description

InstaFoodRoBERTa-NER is a fine-tuned BERT model that is ready to use for Named Entity Recognition of Food entities on social media like informal text (e.g. Instagram, X, Reddit). It has been trained to recognize a single entity: food (FOOD).

Specifically, this model is a roberta-base model that was fine-tuned on a dataset consisting of 400 English Instagram posts related to food. The dataset is open source.

Intended uses

How to use

You can use this model with Transformers pipeline for NER.

from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline

tokenizer = AutoTokenizer.from_pretrained("Dizex/InstaFoodRoBERTa-NER")
model = AutoModelForTokenClassification.from_pretrained("Dizex/InstaFoodRoBERTa-NER")

pipe = pipeline("ner", model=model, tokenizer=tokenizer)
example = "Today's meal: Fresh olive poké bowl topped with chia seeds. Very delicious!"

ner_entity_results = pipe(example, aggregation_strategy="simple")
print(ner_entity_results)

To get the extracted food entities as strings you can use the following code:

def convert_entities_to_list(text, entities: list[dict]) -> list[str]:
        ents = []
        for ent in entities:
            e = {"start": ent["start"], "end": ent["end"], "label": ent["entity_group"]}
            if ents and -1 <= ent["start"] - ents[-1]["end"] <= 1 and ents[-1]["label"] == e["label"]:
                ents[-1]["end"] = e["end"]
                continue
            ents.append(e)

        return [text[e["start"]:e["end"]] for e in ents]

print(convert_entities_to_list(example, ner_entity_results))

This will result in the following output:

['olive poké bowl', 'chia seeds']

Performance on InstaFoodSet

metric val
f1 0.91
precision 0.89
recall 0.93
Downloads last month
1,384
Safetensors
Model size
124M params
Tensor type
I64
·
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train Dizex/InstaFoodRoBERTa-NER