Data has growing significance in exploring cutting-edge materials, and the number of datasets has been generated either by hand or automated approaches. However, the materials science field struggles to effectively utilize the abundance of generated data, especially in applied disciplines where materials are evaluated based on device performance rather than their properties. This article presents a new NLP task called structured information inference (SII) to address the complexities of information extraction at the device level in materials science. We accomplished this task by tuning GPT-3 on an existing perovskite solar cell FAIR (Findable, Accessible, Interoperable, Reusable) dataset with 91.8% F1-score and we updated the dataset with all related scientific papers up to now. The produced data is formatted and normalized, enabling its direct utilization as input in subsequent data analysis. This feature will enable materials scientists to develop their own models by selecting high-quality review papers within their domain. Furthermore, we designed experiments to predict solar cells' electrical performance and design materials or devices with target parameters through LLM. We obtained comparable performance with traditional machine learning methods without feature selection, demonstrating the potential of LLMs to learn scientific knowledge and design new materials like a materials scientist.

翻译：数据在探索前沿材料中日益重要，无论是通过手工还是自动化方法生成的数据集数量都在增长。然而，材料科学领域难以有效利用这些海量生成数据，尤其是在根据器件性能而非材料性质评估材料的应用学科中。本文提出了一种名为结构化信息推理（SII）的新自然语言处理任务，旨在解决材料科学中器件级信息提取的复杂性。我们通过在现有钙钛矿太阳能电池FAIR（可查找、可访问、可互操作、可重用）数据集上对GPT-3进行微调，以91.8%的F1分数完成了该任务，并更新了该数据集以纳入截至目前所有相关科学论文。所生成的数据经过格式化和标准化处理，可直接用于后续数据分析。这一特性将使材料科学家能够通过选择其领域内高质量综述论文来开发自己的模型。此外，我们设计了实验来预测太阳能电池的电性能，并通过大规模语言模型设计具有目标参数的材料或器件。我们在无需特征选择的情况下获得了与传统机器学习方法相当的性能，展示了大规模语言模型像材料科学家一样学习科学知识并设计新材料的潜力。