A Semi-Automatic Annotation Method of Effect Clue Words for Chinese Patents Based on Co-Training

Abstract

In the era of big data, the latest and most advanced technologies are usually revealed to the world in the form of patents. Patents include abundant technical, economic and legal information. A deep analysis and mining of patents can provide important support for enterprises. Patent effect annotation is an important step in patent analysis and mining, and the extraction of patent effect clue words can greatly improve the accuracy and recall rate of annotation. This article summarizes the classification and characteristics of effect clue words, and proposes a co-training-based method of extracting effect clue words from Chinese patents suitable for various fields. Through a strategy called self-filtering, this method can gradually enrich effect clue words thesaurus by iterations, not relying on any other third-party filters. The experiments give the detailed steps, comparisons and boosting of the method.