以專利挖掘、隨機森林與多屬性決策分析探勘自駕車技術

Patent Mining, Random Forest, and MCDM Techniques Based Explorations of Autonomous Vehicle Techniques

陳原本(CHEN Yuan-Pen)

指導教授 ： 黃啟祐

國立臺灣師範大學/科技與工程學院/工業教育學系/碩士(2022年)

https://doi.org/10.6345/NTNU202205658

摘要

汽車產業將迎來具有顛覆性的變革，其中自動駕駛為主要趨勢。專利揭露新型技術之細節，為探索自動駕駛的技術發展，專利挖掘為最有效的方式。雖然多有學者專家探勘各種專利，但少有研究探索專利技術之影響關係，自駕車技術彼此之影響關係更少。因此，本文研究擬定義挖掘自駕車專利之分析架構，跨越此論文之缺口。
首先，本研究依據美國專利資料庫中挖掘之自駕車專利，使用隱含狄利克雷分佈（Latent Dirichlet allocation，LDA）探勘主題，其次，對每一專利，導入隨機森林（Random Forest，RF）演算法，計算每一專利主題對應於其他主題的特徵重要性，之後，將特徵重要矩陣導入決策實驗室分析法（Decision Making Trial and Evaluation Laboratory，DEMATEL），作為初始影響矩陣。最後，以基於決策實驗室之網路流程法（DEMATEL based Analytic Network Process，DANP）推衍每一主題之權重。為探勘自駕車技術，本研究首先自美國專利局下載 26249 件與自駕車相關的專利，並以 LDA 法擷取 30 個主題後，透過群落分析，歸納九大類自動駕駛技術，並由隨機森林與 DANP 法，得知車輛控制系統為影響自動駕駛技術的最關鍵因素，其次為機器視覺與無線通訊，而道路與車輛安全是自駕車技術的基本要求。本分析結果能用來作為未來自駕車公司發展核心能耐的基礎。本研究透過驗證完善之分析架構，能成為傳統汽車公司或科技公司挖掘專利，訂定研發策略之依據。

關鍵字

自動駕駛技術 ； 專利探勘 ； 多準則決策分析 ； 隱含狄利克雷分佈 ； 主題建模 ； 隨機森林 ； 決策實驗室分析法 ； 基於決策實驗室分析法之網路流程

並列摘要

The automotive industry is facing disruptive changes, among which autonomous vehicle techniques are the main trend. Patents reveal the details of new techniques. To understand the development of autonomous vehicle techniques, exploring patent data would be an efficient way. Although many scholars and experts are exploring various patents, there are few studies to explore the influence relationship of patented technologies, and the influence relationship between autonomous vehicle techniques is even less. Therefore, this study intends to define an analysis framework for mining patents of autonomous vehicle techniques and cross the research gap.
First, this study explores autonomous vehicle techniques according to the patent data retrieved from the database of the United States Patent and Trademark Office (USPTO) and extract the topic model via Latent Dirichlet allocation (LDA).Afterward, for every one patent, the Random Forest (RF) algorithm is adopted to derive the feature importance of every one patent versusother topics. The feature importance matrix will be transformed into the initial influence matrix of the Decision-Making Trial and Evaluation Laboratory (DEMATEL). After that, by using the DEMATEL-based analytic network process (DANP), the influence weight versus each topic can be derived.
In order to explore the trajectory autonomous vehicle technologies, this study firstly downloaded 26,249 patents related to autonomous vehicle from the USPTO, and extracted 30 topics by LDA method. Then through the method of clustering and the confirmation of experts, this study obtained nine autonomous driving technologies. The research results demonstrate that the vehicle control system is the key factor affecting the development of autonomous vehicles, followed by machine vision and wireless technologies; Road & Vehicle Safety is the basic requirement for autonomous vehicles. The calculation results will be the basis for autonomous vehicle companies to develop core capabilities. A well-proven analysis framework can be used as a basis for autonomous vehicle companies to excavate patents and formulate research and development (R&D) strategies.

並列關鍵字

Autonomous Vehicle techniques ； Patent exploration ； Topic Modeling ； Random Forest ； Decision-Making Trial and Evaluation Laboratory(DEMATEL) ； DEMATEL-based analytic network process (DANP)

參考文獻

Reference

Abbasi, I. A., & Shahid Khan, A. (2018). A review of vehicle to vehicle communication protocols for VANETs in the urban environment. Future Internet, 10(2), 14.

Acheampong, R. A., & Cugurullo, F. (2019). Capturing the behavioural determinants behind the adoption of autonomous vehicles: Conceptual frameworks and measurement models to predict public transport, sharing and ownership trends of self-driving cars. Transportation Research Part F: Traffic Psychology and Behaviour, 62, 349-375.

Adhikari, M., Hazra, A., Menon, V. G., Chaurasia, B. K., & Mumtaz, S. (2021). A Roadmap of Next-Generation Wireless Technology for 6G-Enabled Vehicular Networks. IEEE Internet of Things Magazine, 4(4), 79-85.

Agarwal, S., Vora, A., Pandey, G., Williams, W., Kourous, H., & McBride, J. (2020). Ford multi-AV seasonal dataset. The International Journal of Robotics Research, 39(12), 1367-1376.