從社群媒體挖掘以感測日常交通滿意度之研究

Sensing Daily Travel Satisfaction Of Commuters by Mining Social Media

陳翰(Chen Han)

指導教授 ： 陶治中

淡江大學/商管學院/運輸管理學系運輸科學碩士班/碩士(2018年)

https://doi.org/10.6846/TKU.2018.00924

摘要

本研究係針對日常交通下社群媒體進行文本挖掘，首先透過建置社群媒體文本資料庫，並將日常交通之社群媒體文本進行情感分析，以瞭解民眾對於日常交通下各公共運具評論之情緒，接著使用深度學習之CNN演算法建構多元情緒決策情感分析模型，再經由實證分析，運用文本挖掘及深度學習分類技術，分析民眾日常通勤通學搭乘公共運具之滿意度，並建構滿意度五個評級尺度，找出民眾搭乘不同公共運具之整體情緒，最後使用K-means集群演算法探討社群媒體輿情內部特徵，此為本研究建立之社群媒體挖掘與情感分析通用程序。
本研究之模式準確度在五個尺度下達79%，並可找出民眾搭乘不同公共運具所關注的變數。研究結果顯示：不論通勤或商務旅次皆關注是否能準時到達目的地，可見民眾最為重視搭乘公共運具之效率；而搭乘火車類、公車類、客運類之通勤族皆關注駕駛員及排班，可見民眾重視資訊的掌握與服務及乘車的安全性。根據自社群媒體挖掘而產生的公共運輸滿意度結果，本研究經由相關文獻評析而研擬相對應的改善策略。若特定時間的班次常有誤點情況時，業者應適時發佈到站時間資訊，則可減少民眾的負面情感。而駕駛員的素質、服務態度、應對能力等，對於民眾的影響亦相當直接，因此業者應對駕駛員進行系統性的在職訓練，亦有望提升民眾的正面情感；當前所有通勤族群對於費率的關注程度皆不明顯，顯示目前我國公共運具在費率制定尚受民眾肯定。本研究之建立之社群媒體挖掘與情感分析通用程序具有擴展性，此可供決策者藉由模型的修正與改良而能快速掌握網路輿情正、負面情感趨勢之即時資訊。

關鍵字

社群媒體挖掘 ； 情感分析 ； 日常交通滿意度 ； 卷積神經網路演算法

並列摘要

This study aims at proposing a generalized process of social media mining and sentiment analysis to sense commuters’ daily travel satisfaction. Firstly, available social media websites are chosen to perform text mining and filtered text database related to daily travel topics by using crawler systems. Secondly, a sentiment analysis is conducted to propose a multiple emotion recognition model which can be used to sense commuters’ emotions about public transportation vehicles including high speed rail, commuter rail, mass rapid transit, urban bus, intercity bus, specific bus and taxi by using Convolutional Neural Networks (CNN) algorithm from deep learning. Thirdly, an empirical study is performed to validate commuters’ daily travel satisfaction towards different public transportation vehicles with a five-grade-scale emotion recognition survey. Finally, influence factors depicting interrelationships among critical topics concerned public transportation services in social media mining are clustered with K-means algorithm and corresponding strategies to improve negative emotions against certain public transportation services are also provided. Empirical results show that the precision percentage of proposed model to verify critical variables of commuter’s public transportation satisfaction approximates 79% under five-grade scales. Either commuter or commercial trip purpose arrivals at destinations on time are much concerned by public transportation commuters. Driver behavior and timetable are also valued by public transportation commuters. It is recommended that operators should pay more attention to adjusting timetable and conducting systematical driver training programs for better emotions. In addition, fare topic is not significant for public transportation commuters that means current fare structures of public transportation are relatively acceptable by commuters. The proposed generalized process of social media mining and sentiment analysis in this study can be expanded with adequate modifications or improvements to grasp real-time information about net citizens’ emotion trends for decision makers.

並列關鍵字

Social media mining ； Sentiment analysis ； Daily travel satisfaction ； Convolutional Neural Networks algorithm

參考文獻

中英文文獻

1. 劉文良(2007)，網際網路行銷(第二版)，台北：碁峰資訊。

2. 張斐章、張麗秋(2015)，類神經網路導論原理與應用，蒼海圖書資訊股份有限公司(第二版) 。

3. 鄭捷(2016)，機器學習-算法原理與編程實踐，北京：電子工業出版社。

4. 陳亭愷(2015)，國道計程電子收費實施後之網路社群媒體文本情感分析研究，淡江大學運輸管理學系運輸科學碩士班碩士論文。