A personalized patent recommendation system using knowledge maps to foster enterprise innovation

Area of Science:

• * Information Science and Technology
• * Intellectual Property Management
• * Artificial Intelligence in Innovation

Background:

• * Enterprises require efficient methods to leverage patent information for innovation.
• * Existing patent data management systems often lack personalization and effective knowledge integration.
• * The complexity of patent data necessitates advanced analytical tools for knowledge discovery.

Purpose of the Study:

• * To develop an effective recommendation system for patent information utilization in enterprises.
• * To construct a patent knowledge map and apply personalized recommendation algorithms.
• * To enhance enterprise efficiency in accessing relevant patent data and supporting innovation.

Main Methods:

• * Extraction of key patent information (titles, abstracts, dates) to build a patent knowledge map.
• * Utilizing the RippleNet algorithm for personalized patent recommendations based on user-patent interactions.
• * Employing the Bidirectional Encoder Representations from Transformers (BERT) model for relationship recognition in patent abstracts.

Main Results:

• * High accuracy in recognizing relationships within patent abstracts using BERT (e.g., composition: 0.973, function achievement: 0.907).
• * Demonstrated trade-off between accuracy and recall in the personalized recommendation system as the number of recommendations increases.
• * Empirical validation of the system's effectiveness and practicality using patent data from Chinese listed companies.

Conclusions:

• * The developed knowledge map-based recommendation system effectively integrates and manages patent information.
• * The system provides enterprises with an intuitive tool to enhance knowledge management capabilities.
• * Personalized patent recommendations improve access to relevant information, supporting strategic innovation.