An instruction dataset for extracting quantum cascade laser properties from scientific text

Area of Science:

• Semiconductor physics
• Laser technology
• Materials science

Background:

• Quantum Cascade Lasers (QCLs) are compact, powerful semiconductor devices with complex designs.
• Extracting QCL properties from unstructured scientific text is crucial for understanding design-performance relationships.
• Existing methods face challenges due to the lack of quality training data for machine learning models.

Purpose of the Study:

• To address the scarcity of labeled data for training machine learning models in QCL property extraction.
• To present an original instruction dataset for training and evaluating Large Language Models (LLMs) on QCL data.
• To facilitate automated data mining and analysis of QCL properties from scientific literature.

Main Methods:

• Augmenting sample sentences from scientific articles using GPT-3.5 instruct with a few-shot strategy.
• Manual annotation of the generated data by Quantum Cascade Laser experts.
• Creation of a dataset comprising 1300 training examples, each with an Instruction, Input Text, and Output.

Main Results:

• Development of a novel, manually annotated instruction dataset specifically for QCL property extraction.
• The dataset is designed to train and evaluate LLMs, overcoming domain-specific challenges in QCL heterostructure and design extraction.
• The dataset provides a foundation for advancing machine learning applications in QCL research.

Conclusions:

• The presented dataset is a valuable resource for improving the accuracy and efficiency of extracting QCL properties from text.
• This work enables more robust data mining and analysis of QCL characteristics, accelerating research and development.
• The dataset supports the adaptation of LLMs for specialized scientific domains like Quantum Cascade Lasers.