Web数据驱动下的专利知识自动撮合模型研究

The implementation of patented technology is the last mile problem for improvement in productivity of the patented technology. Aiming at how to improve the implementation rate of patent, this project starts with the analysis of data characteristics, which draws the conclusion that the excessive expertise of domain knowledge, unknown demand-side of patent and invisible value of patent are important reasons for the low implementation rate of patent. Compared with the patent, the Web data has obvious complementary advantages in the above data characteristics. However, in the fusion of the two kindes of data sources, we face new challenge issues including unintelligent fusion between Web domain knowledge and patent knowledge, inaccurate interactive perception for discovering demand-side of patent and inefficient heterogeneous collaboration for patent evaluation. To solve the above issues, this project proposes a Web data-driven matching model of patent knowledge, including: 1) knowledge representation model based on domain knowledge fusion, which enriches the semantic information of patent through Web domain knowledge fusion to improve the accuracy of patent knowledge matching; 2) Knowledge matching model regardless of the patent value, which automatically discovers the patent counterparties and efficiently matches them with the patent; 3) Knowledge matching model based on patent evaluation, which efficiently discovers the high-valued patents and automatically matches them. The research can promote the transformation of scientific research and provide theoretical and technical support for innovation-driven development strategy in China.

专利技术实施是专利技术向生产力转化的“最后一公里”问题。围绕如何提高专利实施率这一重要问题，本项目从数据特征分析入手，发现专利数据的领域专业、对手潜在以及价值隐含等特征是导致专利实施率低的重要原因之一，Web数据在上述特征方面可以弥补专利数据的不足，与之形成互补。但在二者相互融合过程中我们又面临领域融合低智、交互感知低能和异构协同低效的技术新挑战。为此，本项目提出基于Web数据驱动的专利知识自动撮合模型，该模型包含三方面研究内容：1）基于领域知识融合的知识表征模型，通过领域知识融合来丰富专利知识的语义信息，提高专利知识撮合精度。2）基于对手交互感知的无价撮合模型，可自动准确发现专利对手方，提高专利撮合的速度。3）基于专利价值发现的有价撮合模型，通过专利价值度量高效地发现优质专利，保证专利知识撮合质量。通过以上模型的研究可促进科研成果转化，为我国实施创新驱动发展战略提供理论探索和技术支撑。

专利技术转化是专利技术向生产力转化的“最后一公里”问题。围绕如何提高专利转化率这一重要问题，本项目从数据特征分析入手，发现专利数据的领域专业、对手潜在以及价值隐 含等特征是导致专利转化率低的重要原因之一，Web数据在上述特征方面可以弥补专利数据的不足，与之形成互补。本项目提出基于Web数据驱动的专利知识自动撮合模型，该模型通过专利数据和Web数据相融合构建多源数据异构关联网络，分别从专利知识表示，专利价值发现，专利转化预测三个方面开展研究，以提高专利转化率。我们先后提出：1）基于技术功效的专利表示模型，将该表示模型作为本项目中专利表示的基础，可以提高专利知识撮合精度。2）基于异构关联网络的专利估值模型，将专利作为估值对象，将多源数据异构关联网络作为估值场景，分别提出有监督和无监督两种专利估值方法，高效地发现高价值专利，保证专利知识撮合质量。3）基于异构关联网络的可解释专利转化预测模型，将专利作为估值对象，将多源数据异构关联网络作为转化场景，分别提出专利转化路径预测方法和可解释的专利转化机会预测方法，来发现专利转化对手，预测专利转化机会，分析专利转化条件，提高专利撮合的效率。该研究为加速科技服务业发展、加快科技成果转化提供了支撑性的理论和方法。