Skip to content
Back to archive

TRIZ & Graph RAG-based Intelligent Patent Convergence (IPC) Service (Creative Award)

Published:

Project & Award: This team project (Junho Shin, Dongjae Kim, Miji Kang, Gaeun Kim) from Pukyong National University’s Data Engineering Dept. was presented at the K-Data Science (K-DS) Consortium, where it was awarded the Creative Award for the topic “Development of TRIZ-based Convergent Technology Idea Discovery Support Service”.

Framework of IPC System

Figure 1. Framework of IPC System

Predicting where a market is heading requires understanding how discrete technologies may converge in the future. This project introduces the Intelligent Patent Convergence (IPC) service, a data-driven system for supporting technology convergence analysis and R&D ideation.

📌 The Problem: Intuition-Dependent R&D and IP Risks

Innovation rarely happens in isolation; it occurs when different technological domains collide. However, finding these convergence opportunities hidden within millions of complex patent documents is cognitively overwhelming. Traditional R&D ideation heavily relies on human intuition, leading to severe inefficiency, high costs, and high failure rates. Furthermore, blind convergence often results in unexpected Intellectual Property (IP) disputes and patent infringements.

⚙️ The Method: TRIZ, Knowledge Graphs, and LangChain

To systematically simulate viable tech convergence, we integrated the TRIZ (Theory of Inventive Problem Solving) methodology with Graph Database technology and Agentic LLM workflows:

  1. TRIZ Contradiction Analysis: Every technical problem involves a “contradiction” (e.g., improving strength increases weight). The system leverages TRIZ’s universally proven ‘40 Inventive Principles’ and contradiction matrix to logically deduce how different technologies can resolve each other’s weaknesses.
  2. Knowledge Graph Construction (Neo4j): Processed vast amounts of patent data and mapped their complex relationships (citations, inventors, technical classifications, TRIZ parameters) into a Graph Database using Neo4j.
  3. Graph RAG via LangChain: Replaced standard vector-based RAG with Graph RAG utilizing LangChain. This allows the LLM to retrieve interconnected concepts and structural relationships within the technology landscape.

Contribution: Simulated Scenarios & IP Defense

This system supports technology opportunity exploration and IP risk review for R&D strategy.


Share this research:

Previous
[Submitted] Mapping Technological Convergence via LLM-enhanced Analytics
Next
Unsupervised Learning-Based Electrical Anomaly Detection in EV Batteries (KCI)