SongKun909/Lithium-Battery-IE-Dataset

Lithium-Ion Battery Patent Technical Indicator Dataset (锂离子电池专利技术指标精标数据集)

Introduction (简介)

This repository provides a highly specialized, bilingual (Chinese & English) instruction-tuning dataset designed for Fine-grained Information Extraction (IE) from Lithium-ion battery patents. It is the official data repository for our data paper: [A Dataset of Fine-Grained Technical Indicators from Lithium-Ion Battery Patents for Instruction Tuning of Large Language Models].

本项目为数据论文《面向大语言模型指令微调的锂离子电池专利细粒度技术指标数据集构建》的官方开源数据集。旨在解决通用大语言模型（LLMs）在处理高壁垒工业长文本时面临的“垂直常识缺失”与“跨语境知识不对齐”问题。

Key Features (核心亮点)

• —High-Quality SFT Data: 1887 instruction-tuning samples rigorously verified via a Human-In-The-Loop (HITL) pipeline.
• —Cross-lingual Knowledge Anchors: A carefully designed ~8:2 (CN:EN) disproportionate sampling ratio to inject English physical dimensions and syntax into the model.
• —Professional Benchmark: Includes a hold-out test set and a custom Unit-Aware Dynamic Evaluation Algorithm (Dynamic F1-Star).

Repository Structure (文件说明)

• —battery_alpaca_sft_dataset.json: 1887 instruction-tuning samples (SFT Training Data).
• —test_sft_100.json: 100 benchmark samples in Alpaca format for zero-shot evaluation.
• —gold_standard_final_clean.json: Raw gold standard annotations by domain experts.
• —eval_expert.ipynb: Evaluation script featuring the Unit-Aware Dynamic F1-Star algorithm.

Quick Start (快速开始)

You can easily load this dataset using the datasets library:

from datasets import load_dataset

# Load the SFT training dataset
dataset = load_dataset("SongKun909/Lithium-Battery-IE-Dataset", data_files="battery_alpaca_sft_dataset.json")
print(dataset['train'][0])