结构化与非结构化数据双向驱动的智能专利价值评估系统

Patent is an important part of intangible assets, patent valuation is the key link patent operations. Traditional intangible assets assessment method is widely used as a common method of patent value assessment. Practice shows that the analysis results obtained from structured data do not fully reflect the value of patents, and unstructured data analysis is an effective supplement to them. However, the combination of the two is a difficult problem to be solved urgently. For the first time, this project integrates structured and unstructured data into patent value assessment research, and establishes two kinds of bi-directional smart patent value evaluation systems driven by data..The main research contents include: according to the structured data, a.wrap-around attribute selection method based on prediction accuracy and cost sensitivity is proposed, and the key external indicators that affect the value are found out. For non-institutional data, this project proposes a feature extraction method based on text similarity of word vector model, and builds a quantitative internal index to measure innovation and protection. Based on the above indexes, an intelligent patent value evaluation model is established based on a random forest method that can handle multiple heterogeneous data..Furthermore, the pruning of the patent by designing the difference degree measurement results in a more simplified evaluation model. The research results of this project help to improve the intellectualization and reliability of patent value assessment and enrich the existing data mining methods.

专利是无形资产的重要组成部分，专利价值评估是专利运营的关键环节。传统的无形资产评估方法作为一种通用的专利价值评估方法被广泛采用，实践表明结构化数据得出的分析结果并不能完整地体现专利价值，非结构化数据分析是其有效补充，但如何将两者有机结合是亟待解决的难点问题。本项目首次将结构化与非结构化数据一并纳入专利价值评估研究中，建立两种数据双向驱动的智能专利价值评估系统。主要研究内容包括：针对结构化数据，提出基于预测准确率和代价敏感的包裹式属性选择方法，发现影响价值的关键外部指标；针对非机构化数据，提出基于词向量模型文本相似度的特征提取方法，构建衡量创新性和保护性的量化内部指标；利用上述指标，基于可处理多重异构数据的随机森林方法建立智能专利价值评估模型，更进一步地，通过设计差异度测度对其剪枝，得到更为简约的评估模型。本项目的研究成果有助于提高专利价值评估的智能化程度和可靠性，并丰富现有的数据挖掘方法。

随着国家知识产权发展的战略地位日渐突出，我国人均专利申请量居全球首位，其巨大数量及多领域复合性为专利审查、评估工作带来挑战。目前亟需有科学的方法协助审查评估机构对专利规避专家知识主观影响快速开展审查评估工作。传统的专利价值评价指标体系中仅使用单一结构化数据导致的信息不全、依赖专家经验导致的信息差异等问题，本项目通过引入专利的非结构化数据提高信息完整度，综合不同领域专家背景知识规避主观影响，提出构建结构化与非结构化数据双向驱动的智能专利价值评估系统。首先对结构化专利数据进行特征选择选取最优特征子集，其次对非结构化数据提取关键特征，最终通过提取的结构化专利数据指标与非结构化专利数据特征组合构建结构化与非结构化数据双向驱动的智能专利价值评价体系。项目研究以智慧芽专利数据库作为数据来源，研究结果表明，通过构建结构化与非结构化数据组合的专利价值评价体系帮助解决市场等非确定性因素造成的价值波动、及由于专利审查对不同领域专家大量需求和时间紧迫性问题，为专利价值评估提供新的思路新方法，为知识产权相关部门提供技术支撑、辅助企业在科技研发领域进行高效转化。