基于密度峰值聚类的专利地图制作方法

Patent map making method based on CFSFDP

摘要:

目前国内对于专利地图的研究大部分仍停留在应用阶段，对其制作的基础理论研究较少。概述目前专利地图类别，分析现有专利地图制作方法的缺陷，从增强专利文献信息可信度和价值的角度，运用TF-IDF(term frequency-inverse document frequency)统计特征将非结构化的专利文献信息映射到低维空间中，采用密度峰值快速搜索聚类(clustering by fast search and find of density peaks，CFSFDP)算法进行聚类，对同一聚类中的专利文献特征进行分析，得到不同专利文献间的发展关系并映射为图表示，从而构建以有向图表示的专利地图。改进提出的这种专利地图制作方法，同时利用了结构化信息与非结构化信息，以使专利地图更为真实准确地反映目标技术领域的技术发展过程。

Abstract:

At present, most of the research on patent map is still in the application stage, the basic theory of its production is less studied. This paper summarize the current patent map categories, analyze the shortcomings of existing patent map constructing methods, improves and proposes a patent map constructing method to the point from the perspective of enhancing the credibility and value of patent map, i.e. mapping unstructured patent document information into low dimensional space by using the statistical features of TF-IDF (term frequency-inverse document frequency), clustering by using CFSFDP (clustering by fast search and find of density peaks) algorithm, and analyzes the characteristics of patent documents in the same cluster with unstructured information, obtains the development relationship between different patent documents and map it to a graph, constructs the patent map represented by directed graph. Using both structured information and unstructured information, the proposed method for constructing patent maps is improved, by which the resulting patent maps can more accurately and accurately reflect the technological development process of the target technical field.