Scholarlink

<p>The rapid changes in the technological composition of industries drive companies to seek external patents to foster production innovation. Existing patent recommendation methods primarily analyze the textual and bibliographic features of a company’s historical patents but often overlook industry relevance of patents and the technology development trends. To address this, this paper proposes a multi-view learning-based patent recommendation (MVLPR) approach that integrates industry relevance and technology trends. A multi-label classification model is employed to identify the industrial sectors applicable to a patent, while a Long Short-Term Memory (LSTM) network layer is designed to capture technology development trends. Offline experiments demonstrate that MVLPR achieves a better balance between accuracy and diversity than baseline methods. Crucially, the industry sector view alleviates the cold-start problem for new companies by leveraging inter-company similarities within the same industry sector to provide prior signals of technological interest independent of individual historical patents. Furthermore, the technology trend analysis within the International Patent Classification (IPC) view significantly improves both recommendation accuracy and diversity.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing patent recommendations for product innovation: integrating industry relevance and technology trends with multi-view learning

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Yuwen Chen,
  • Peihu Zhu,
  • Weiwei Deng,
  • Xiaoming Huang,
  • Jian Ma,
  • Jin Qin

摘要

The rapid changes in the technological composition of industries drive companies to seek external patents to foster production innovation. Existing patent recommendation methods primarily analyze the textual and bibliographic features of a company’s historical patents but often overlook industry relevance of patents and the technology development trends. To address this, this paper proposes a multi-view learning-based patent recommendation (MVLPR) approach that integrates industry relevance and technology trends. A multi-label classification model is employed to identify the industrial sectors applicable to a patent, while a Long Short-Term Memory (LSTM) network layer is designed to capture technology development trends. Offline experiments demonstrate that MVLPR achieves a better balance between accuracy and diversity than baseline methods. Crucially, the industry sector view alleviates the cold-start problem for new companies by leveraging inter-company similarities within the same industry sector to provide prior signals of technological interest independent of individual historical patents. Furthermore, the technology trend analysis within the International Patent Classification (IPC) view significantly improves both recommendation accuracy and diversity.