Ailee52/PatentSBERTa_finetuned_green_multiagent-dataset

Assignment3 - Green Patent Detection: Advanced Architectures (Option: Agentic-CrewAI llama3:4b)

Green Patent Detection Dataset (Multi-Agent + HITL)

This dataset was created for Assignment 3: Advanced Architectures (Agents vs QLoRA). It contains patent claims used to train and evaluate models for **green technology classification.

The dataset combines silver labels, multi-agent generated labels, and human-in-the-loop (HITL) verification.

NEW MODEL

• —Ailee52/PatentSBERTafinetunedgreen_multiagent

Agreement report (HITL)

• —The agreement between the agent system and the human labels was 56%. This shows that the multi-agent system correctly matched human decisions in slightly more than half of the reviewed high-risk patent claims.

Dataset Structure

The repository contains the following files:

patent50kgreen.parquet

Balanced dataset of 50,000 patent claims used as the primary silver training data.

Columns:

• —doc_id – patent identifier
• —text – patent claim text
• —is_green_silver – silver label for green technology
• —split – dataset split (trainsilver / evalsilver / pool_unlabeled)

A3agentlabels100FINAL.csv

Contains 100 high-risk patent claims labeled using a Multi-Agent System (MAS) and verified through Human-in-the-Loop (HITL) review.

Columns:

• —doc_id
• —advocate_argument
• —skeptic_argument
• —judge_rationale
• —is_green_tech – final classification from the agent system

These claims were selected using uncertainty sampling from Assignment 2.

a3gold100_labeled.csv

Final human-verified gold labels for the 100 high-risk claims.

Columns:

• —doc_id
• —text
• —is_green_gold

These labels were produced after reviewing the agent system's output.

Label Definitions

Labeling Pipeline

The dataset was constructed using the following workflow:

1. 1.Silver dataset generated using automated labeling.
2. 2.Uncertainty sampling selected 100 high-risk patent claims.
3. 3.A Multi-Agent System analyzed each claim using three agents:
4. 4.Advocate
5. 5.Skeptic
6. 6.Judge
7. 7.The agent outputs were reviewed using Human-in-the-Loop (HITL)
8. 8.Final labels were stored as gold labels ---

Disclaimer

This project was developed for academic purposes only. The classification results are intended for research and educational use, and should not be interpreted as legal advice or professional patent evaluation. The Human-in-the-Loop (HITL) annotations were performed by students as part of a coursework assignment and do not represent expert legal judgment. The model may contain biases and errors inherited from both automated labeling (silver labels) and LLM-assisted human review. --- Video Link Assignment Explaination: https://aaudk-my.sharepoint.com/:v:/g/personal/sm42zmstudentaau_dk/IQAlr-XVaSjLTpLLW0MH5LZKAUhA5aISCR6qsWVu7CR9kuE