(SSCI+EI收录)Forecasting emerging technologies A supervised learni
需积分: 0 49 浏览量
2022-08-03
20:25:49
上传
评论
收藏 482KB PDF 举报
Contents lists available at ScienceDirect
Technological Forecasting & Social Change
journal homepage: www.elsevier.com/locate/techfore
Forecasting emerging technologies: A supervised learning approach through
patent analysis
Moses Ntanda Kyebambe
a
, Ge Cheng
b,⁎
, Yunqing Huang
a
, Chunhui He
a
, Zhenyu Zhang
a
a
Department of Mathematics and Computational Science, Xiangtan University, Xiangtan, Hunan, China
b
College of Information Engineering, Xiangtan University, Xiangtan City, Hunan Province, China
ARTICLE INFO
Keywords:
Technology forecasting
Industrial technology roadmap
R & D planning
Patent analysis
Citation analysis
ABSTRACT
Both private and public enterprises have great interest in prior knowledge of emerging technologies to enable
them make strategic investments. Technology forecasting offers a relevant opportunity in this direction and is
currently a hot upcoming area of research. However, accurate forecasting of emerging technologies is still
problematic mainly due to absence labeled historical data to use in training of learners. Previous studies have
approached the technological forecasting problem through unsupervised learning methods and, as such, are
missing out on potential benefits of supervised learning approaches such as full automation. In this study, we
propose a novel algorithm to automatically label data and then use the labeled data to train learners to forecast
emerging technologies. As a case study, we used patent citation data provided by the United States Patent and
Trademark Office to test and evaluate the proposed algorithm. The algorithm uses advanced patent citation
techniques to derive useful predictors from patent citation data with a result of forecasting new technologies at
least a year before they emerge. Our evaluation reveals that our proposed algorithm can retrieve as high as 70%
of emerging technologies in a given year with high precision.
1. Introduction
Due to a very fast pace at which technology is evolving, enterprises
are faced with a hard decision of the best and most suitable technology
to invest in. In this paper, we propose an algorithm for predicting
emerging technologies to support enterprises to make data driven de-
cisions over which technologies to invest in. The model is designed to
detect signals of a technology likely to cause a significant disruption in
an industry at least a year before the technology fully emerges. This
way, the model has a potential of reducing the risk of being late to
adopt a technology by an enterprise. Automatic forecasting of tech-
nologies remains a difficult task largely due to scarcity of labeled data
to train reliable classifiers; traditional approaches have relied on un-
supervised learning methods. Patent databases offer a huge source of
technological inventions data that many researchers have exploited
with unsupervised learning methods to forecast technologies mostly
relying on citations. In the context of patent studies, a citation is re-
ference to previous work (also known as prior art) that is relevant to the
current patent application. For a specific granted patent, the patents it
cites are known as backward citations while future patents that cite it
are known as forward citations. All methods that base on forward ci-
tations to forecast technological trends suffer one major limitation of a
large time lag between the date a patent is published to the date it
begins attracting citations. In this study, rather than relying on forward
citations which take long to build, we use backward citations to derive
several features most capable of discriminating a high-impact patent of
technology likely to disrupt business in a given industry from patents of
just incremental technology. Furthermore, we propose an algorithm for
labeling emerging technology patent clusters based on new classes
progressively established in the United States Patent Classification
(USPC) system overtime. Besides the USPC, the proposed method is
extensible to make use of other data sources such as blogs, conferences
and social networks in labeling emerging technology patent clusters.
This study is part of a growing number of studies (Érdi et al., 2013;
Fleming et al., 2006; Karvonen and Kässi, 2013; Sorenson et al., 2006)
that have employed patent citation analysis in predictive analytics,
particularly technological trends.
2. Literature review
Use of citations in analytics dates far back in the 1970s with
Garfield's (Garfield, 1979) extensive article on citation index theory and
its application to patent literature analysis, scientific journal analysis
and many other areas. The Science Citation Index is indeed still widely
http://dx.doi.org/10.1016/j.techfore.2017.08.002
Received 22 November 2016; Received in revised form 15 July 2017; Accepted 1 August 2017
⁎
Corresponding author.
E-mail addresses: [email protected] (M.N. Kyebambe), [email protected] (G. Cheng), [email protected] (Y. Huang), [email protected] (Z. Zhang).
Technological Forecasting & Social Change 125 (2017) 236–244
Available online 15 August 2017
0040-1625/ © 2017 Elsevier Inc. All rights reserved.
MARK
方2郭
- 粉丝: 26
- 资源: 324
最新资源
- Picasso_v3.1 2.ipa
- chromedriver-mac-arm64.zip
- 蓝zapro.apk
- chromedriver-linux64.zip
- UCAS研一深度学习实验-MNIST手写数字识别python源码+详细注释(高分项目)
- 基于Python和PyTorch框架完成的一个手写数字识别实验源码(带MINIST手写数字数据集)+详细注释(高分项目)
- 基于Matlab在MNIST数据集上利用CNN完成手写体数字识别任务，并实现单层CNN反向传播算法+源代码+文档说明(高分项目)
- NVIDIA驱动、CUDA和Pytorch及其依赖
- 基于SVM多特征融合的微表情识别python源码+项目说明+详细注释(高分课程设计)
- html动态爱心代码一（附源码）
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
评论0