情报杂志2025，Vol.44Issue(6)：127-137,11.DOI:10.3969/j.issn.1002-1965.2025.06.017
基于时序聚合异构图的高价值专利识别方法
High-Value Patent Identification Method Based on Time-Series Aggregated Heterogeneous Graphs
邓娜 1喻卓群 1孙俊杰 1陈旭 2刘树栋 2孙湘怡3
作者信息
- 1. 湖北工业大学计算机学院 武汉 430068
- 2. 中南财经政法大学信息工程学院 武汉 430073
- 3. 湖北工业大学外国语学院 武汉 430068
- 折叠
摘要
Abstract
[Research purpose]This study aims to address the shortcomings of existing high-value patent identification methods in utili-zing heterogeneous patent associations and time-series features,to identify high-value patents more accurately.[Research method]We propose a high-value patent identification model based on time-series aggregated heterogeneous graphs.This model integrates multimodal patent information and designs a dynamic update mechanism for time-series citation influence,generating a time-series aggregated hetero-geneous graph that reflects changes in patent value.We construct a heterogeneous graph convolutional network model that incorporates a bidirectional attention mechanism to enhance the extraction capability of heterogeneous patent features,achieving precise identification of high-value patents.[Research result/conclusion]Experimental results show that our method achieves an accuracy of 84.61％and an F1 score of 84.59％on a patent dataset in the smart grid field,outperforming conventional methods.This validates the effectiveness of our approach and provides new perspectives and methodological references for patent screening and value assessment.关键词
高价值专利识别/异构图卷积网络/双向注意力机制/动态更新机制/多维特征Key words
high-value patent identification/heterogeneous graph convolutional network/bidirectional attention mechanism/dynamic up-dating mechanism/multi-dimensional features分类
社会科学引用本文复制引用
邓娜,喻卓群,孙俊杰,陈旭,刘树栋,孙湘怡..基于时序聚合异构图的高价值专利识别方法[J].情报杂志,2025,44(6):127-137,11.
