基于Doc2vec的专利与行业类目映射研究

目的/意义]使用深度学习中Doc2vec文本向量化的方法进行专利与行业间类目相似度的计算,旨在为用计算机进行类目映射时提供新的方法和思路。方法/过程]实验通过《国际专利分类表》的小类及其下级类目大组与《国民经济行业分类表》中的小类展开,通过Doc2vec文本向量化和余弦相似度的方法求取三组相似值(专利小类与行业小类、专利大组与行业小类、每组专利小类下大组与行业小类相似度的平均值),并以农业类目为例进行解释说明。结果/结论]通过计算专利大组与行业小类相似度平均值的方法进行映射更具合理性。

Doc2vec-based Study on Mapping Between Patented and Industrial Categories

Purpose/significance]The paper uses Doc2vec which belongs to deep learning to calculate similarities between patented and industrial categories so as to provide a new method for category mapping while using computers.Method/process]The experiment is carried out with the subclasses as well as their main groups of International Patent Classification and the subclasses of National Industries Classification.And with the methods of Doc2vec text vectorization and cosine similarity it calculates three sets of similarities one set is the average similarity between patented and industrial subclasses another one is the average similarity between patented main groups and industrial subclasses and the third is the average similarity between patented subclass’main groups and industrial subclasses.Result/conclusion]It was proved that the third was the most efficient and reasonable one to category mapping.