SciHIT IE
Scientific Human–AI Teaming for Information Extraction
Built by Mutahar Safdar, Jiarui Xie
Overview
SciHIT IE (Scientific Human–AI Teaming for Information Extraction) is a human-centered framework designed to extract structured scientific information from literature using large language models (LLMs). It combines automated natural language processing with expert-in-the-loop feedback to ensure accuracy, transparency, and adaptability.
The framework is composed of three main components:
SciHIT IE is designed for continuous improvement and domain adaptation, making it a reusable tool for accelerating literature mining in AM and other scientific fields.
Features
03- 01
Base IE System – Allows users to upload and parse scientific PDFs, retrieve semantically relevant paragraphs, and customize retrievals through interactive labeling (positive/negative).
- 02
Paragraph Classification Tier – Trains classifiers to categorize paragraphs into key information types (e.g., data, model, sensing, system), enabling scalable filtering across large document collections.
- 03
Query Tier – Uses a generative LLM to answer user-defined questions based on filtered content, supporting dynamic and focused information extraction.