A personalized recommendation system for high-quality patent trading by leveraging hybrid patent analysis

Patents, as technological innovation with commercial values, play a significant role for increasing enterprise and national competitiveness. Personalized recommendation in online patent marketplace would help patent buyers effectively identify their demands from the deluge of patents. However, state-of-the-art patent recommendation methods focus mainly on increasing recommendation effectiveness while ignoring patent quality and recommendation explainability. The glut of low-quality products in patent trading platforms reduces patent buyers’ trust and would further damage the patent market due to their weak technological competitiveness and low market potential. Besides, there is a need to differentiate the varied semantics (e.g., textual content, patent classification and citation) enriched in patent documents when recommending patents to potential buyers. To solve the problems, this research proposed a framework of personalized patent recommendation system by leveraging hybrid patent analysis. The system designs an improved high-quality patent identification method by including patent inventors’ reputation as a new indicator. Moreover, an innovative patent preference analysis is proposed by analyzing a potential buyer’s intra-collection patent citation network. Last, the content-based strategy, classification-based strategy and citation-based strategy are separately introduced for transparent patent recommendation. The offline experiment on patent quality indicators reveals that inventor reputation as a new quality indicator helps identify high-quality patents besides the widely-used quality indicators. The simulation experiment of patent recommendation validates the effectiveness of different recommendation strategies. The proposed recommendation framework was implemented on a real-world patent trading platform and achieved good performance.