stormsidali2001/IMRAD-introduction-move-one-sub-moves-classifier

IMRaD Introduction Move 1 Sub-move Classifier

This model is a fine-tuned BERT model that classifies sentences from the "Establishing a Niche" (Move 1) section of scientific research paper introductions into their corresponding sub-moves:

• Claim something is wrong with the previous research: Pointing out limitations, flaws, or areas where past research falls short.
• Highlight a gap in the field: Identifying areas where knowledge or research is lacking.
• Raise a question where research in the field is unclear: Presenting an unanswered question or ambiguity in existing research.
• Extend prior research to add more information on the topic: Suggesting a new direction or contribution building on previous work.

Parent Classifier:

This model works in tandem with the main IMRaD Introduction Move Classifier: https://huggingface.co/stormsidali2001/IMRAD_introduction_moves_classifier

First, use the parent classifier to identify sentences belonging to "Establishing a Niche" (Move 1). Then, utilize this sub-move classifier to analyze the specific role each Move 1 sentence plays in establishing the research niche.

Intended Uses & Limitations

Intended Uses:

• Scientific Writing Assistance: Help researchers and students analyze and strengthen their "Establishing a Niche" section by precisely categorizing each sentence's sub-move.
• Literature Review Analysis: Gain a deeper understanding of how authors establish the need for their research by identifying the specific sub-moves used in Move 1.
• Educational Tool: Demonstrate the various sub-moves employed to establish a research niche in scientific writing.

Limitations:

• Domain Specificity: Trained on scientific research papers; accuracy may vary on other text types.
• Sentence-Level Classification: Focuses on individual sentences; does not provide a holistic analysis of the entire Move 1 section.
• Prediction Accuracy: While generally accurate, the model might misclassify complex or ambiguous sentences. Review predictions critically.

Training and Evaluation Data

Trained and evaluated on a subset of the "IMRAD Introduction Sentences Moves & Sub-moves Dataset": https://huggingface.co/datasets/stormsidali2001/IMRAD-introduction-sentences-moves-sub-moves-dataset

Specifically, the model uses sentences labeled as Move 1, further classified into the four sub-moves.

Training Details:

• Base Model: google/bert-base-cased
• Implementation: TensorFlow/Keras
• Evaluation Metrics: F1 score and accuracy

How to Use

from transformers import pipeline

# Parent classifier
move_classifier = pipeline("text-classification", model="stormsidali2001/IMRAD_introduction_moves_classifier")

# Move 1 sub-move classifier
submove_classifier_1 = pipeline("text-classification", model="stormsidali2001/IMRAD-introduction-move-one-sub-moves-classifier")

sentence = "This gap in research highlights the need for further investigation into [topic]." 

move_result = move_classifier(sentence)
move = move_result[0]['label']

if move == "Establishing a Niche":
    submove_result = submove_classifier_1(sentence)
    print(submove_result)