Semi-automatic construction method of Chinese patent corpus based on TRIZ

Abstract

The invention provides a semi-automatic construction method for the Chinese patent corpus based on the TRIZ. 40 TRIZ inventive principles are recombined, large ambiguity exists between categories of the regrouped inventive principles, the ambiguity in the categories is relatively small, and the quality of the corpus is further improved. The patent text contains many low-frequency domain terms, complete terms are destroyed by utilizing general Chinese word segmentation, and complete term keywords can be acquired by the method, so that a good foundation is provided for patent semantic analysis.According to the method, semantic information provided by keywords is limited, dependency syntactic analysis is carried out on sentences, more sufficient semantic information is obtained, machine recognition is more accurate, understanding of sentences by non-field professionals is facilitated, and then a small amount of annotation work is better completed. According to the method, sentences and dependency features of patent texts are extracted on the basis of a representation learning method, deeper and more abstract patent semantic representation is captured, and the most discriminative features can be extracted, so that clustering of the texts is facilitated.

Description

technical field [0001] The present invention relates to the technical field of patent text classification, and more specifically, relates to a semi-automatic construction method of a Chinese patent corpus based on TRIZ. Background technique [0002] In recent years, the number of patents in my country has shown a blowout growth, and there is still a large gap in the gold content of Chinese patents compared with developed countries, and there is a problem of "disharmony between the number and quality of patents". Therefore, people urgently need an effective means to effectively organize and manage massive patent data, and then identify key technologies with truly innovative invention principles, providing important basis for enterprise transformation and upgrading and formulation of national industrial policies. [0003] However, the traditional automatic classification of patents based on the International Patent Classification (IPC) cannot meet people's needs for fine-grain...

AI Technical Summary

Problems solved by technology

[0006] Academic paper (Author: Shi Cui, Title: Analysis and Solution of Patent Document Corpus Retrieval Problems, Journal of Liaoning Administration Institute) This document provides a certain solution from three aspects of patent document segmentation, part-of-speech tagging and dependency syntax analysis, but The representation of its patent is composed of feature engineering, which does not consider automatic learning of effective features, and has certain limitations

[0007] The corpus of traditional TRIZ-based patent text-related research relies on manual construction, which not only consumes a lot of manpower and material resources, but also has difficulty in guaranteeing its reliability due to differences in the understanding of knowledge contained in patents by annotators.

In addition, there are 40 original TRIZ invention principles, which are relatively large in number, and each principle is relatively abstract, and there is a phenomenon of overlapping text description information for some principles, which has caused certain troubles for annotators

Images