Understanding Label Meanings in FLANG-BERT Model

[JY251]

Code:

I have been working with the FLANG-BERT model and have encountered an issue with understanding the labels returned by the model. Here is the code I am using:

from transformers import BertTokenizer, BertForSequenceClassification, pipeline

model_path = "FLANG-BERT"
# model_path = "FLANG-ELECTRA"

flang_bert = BertForSequenceClassification.from_pretrained(model_path, num_labels=3) # Adjust num_labels as needed
tokenizer = BertTokenizer.from_pretrained(model_path)

nlp = pipeline('text-classification', model=flang_bert, tokenizer=tokenizer)
results1 = nlp("I am happy")
results2 = nlp("I am angry")
results3 = nlp("I am sleepy")

print(results1)
print(results2)
print(results3)

Output:

The output I receive is as follows:

[{'label': 'LABEL_1', 'score': 0.3824489414691925}]
[{'label': 'LABEL_1', 'score': 0.38688215613365173}]
[{'label': 'LABEL_1', 'score': 0.3874432444572449}]

When I run the same code again multiple times, I obtain different outputs:

[{'label': 'LABEL_2', 'score': 0.33723291754722595}]
[{'label': 'LABEL_2', 'score': 0.3374021649360657}]
[{'label': 'LABEL_2', 'score': 0.33843377232551575}]

[{'label': 'LABEL_0', 'score': 0.3722850978374481}]
[{'label': 'LABEL_0', 'score': 0.37265387177467346}]
[{'label': 'LABEL_0', 'score': 0.3684918284416199}]

Issue:

I am unsure what the labels (e.g., 'LABEL_1') represent in the FLANG-BERT model. I have looked at the documentation for the model, such as the Hugging Face documentation, but I could not find any notes about the meaning of the labels.

As there are only 3 labels in this case (maybe due to the num_labels=3 parameter), I suspect these 3 correspond to 'Positive', 'Negative', and 'Neutral' sentiments. However, I am still uncertain about if "LABEL_1" corresponds to 'Positive' or 'Negative' sentiment.

Could you please provide information on what the labels correspond to or direct me to where I can find this information in the documentation?

Thank you for your assistance.