Device and method for automated ethical essay scoring of harmful essays
The method and apparatus for ethical automatic scoring generate harmful essay datasets and classify essays ethically, addressing the detection of unethical content and dataset deficiencies in existing systems, providing fair and consistent grading.
WO2026142310A1PCT designated stage Publication Date: 2026-07-02KONKUK UNIV IND COOP CORP
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- KONKUK UNIV IND COOP CORP
- Filing Date
- 2025-12-24
- Publication Date
- 2026-07-02
AI Technical Summary
Technical Problem
Existing automated essay scoring systems fail to detect unethical or immoral content in essays, leading to improper feedback, and lack sufficient datasets for harmful essays, making them vulnerable during actual implementation.
Method used
A method and apparatus for ethical automatic scoring that generates a dataset of harmful essays by bypassing security policies of large language models (LLMs) and classifies essays considering ethical aspects, using a system with processors and memory to create and label harmful essays.
Benefits of technology
Enables appropriate scoring of essays with harmful content while generating a dataset to supplement the lack of existing harmful essay data, ensuring fair and consistent grading.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure KR2025022703_02072026_PF_FP_ABST
Abstract
A technique related to a device and method for automated ethical essay scoring of harmful essays is disclosed. An automated ethical essay scoring device which distinguishes between harmful essays and argumentative essays generates an instruction for instructing a large language model (LLM) to score an essay on the basis of a persona for essay evaluation, and receives a scoring result generated for the essay from the large language model. The instruction includes ethical scoring criteria, and the persona for evaluation may be determined on the basis of a toxicity evaluation result of the essay. Harmful essays may be generated by bypassing a safety policy of the LLM, and whether the generated essays correspond to harmful essays may be classified on the basis of toxicity evaluation results. According to the present invention, data in a dataset lacking harmful essays can be augmented, and accurate evaluation of harmful essays is possible by evaluating essays in consideration of ethical aspects.
Need to check novelty before this filing date? Find Prior Art