237 0

베이지안 네트워크를 이용한 루머 확산 탐지 시스템

Title
베이지안 네트워크를 이용한 루머 확산 탐지 시스템
Other Titles
Rumor Propagation Detection System based on Bayesian Network
Author
양훈지
Alternative Author(s)
Hoonji Yang
Advisor(s)
오희국
Issue Date
2016-02
Publisher
한양대학교
Degree
Master
Abstract
최근 스마트폰이나 태블릿 PC 등의 스마트 기기의 사용이 증가하면서 소셜 네트워크 서비스(Social Network Service, SNS)의 사용자가 증가하고 있다. 소셜 네트워크 서비스는 빠른 전파성을 가지고 있어 정보를 전달하는 도구로 많이 사용되고 있으나, 이에 따른 부정적인 측면도 고려하지 않을 수 없다. 본 논문에서는 루머를 확산하고자 하는 계정의 실제 데이터를 분석하고 특징을 추출하여 이를 토대로 계정의 악성행위 여부를 판단하고자 한다. 먼저 실제 데이터를 크롤링하고 악의적인 계정과 일반 계정의 특징을 추출하고 분석하였다. 기존 연구를 바탕으로 3가지의 계정, 컨텐츠, 확산의 관점에서 특징을 선택할 수 있었고 새로운 특징 하나를 추가하였다. 이는 유명인의 트윗을 리트윗한 후 악의적인 내용으로 수정하는 행위에 대한 특징으로써, RT와 그 계정이 직접 올린 트윗의 비율을 살펴보았다. 그 후 분류된 특징을 기반으로 각 데이터의 평균값을 이용하여 분류 기준을 선택하고 이를 학습 시켰다. 학습 기법으로는 베이지안 네트워크를 이용하였고, 새로운 데이터가 입력되면 재학습될 수 있도록 설계하였다. 제안하는 시스템은 91.94%의 정확도를 보여주었고, F-measure 값은 93.76%를 보여주었다.|The growing use of the smart device such as smartphones and tablets has resulted in increasing number of social network service(SNS) users recently. SNS allows a fast propagation and it is used as a tool to send information. But its negative sides need to be considered. In this paper, we analyzed actual data of malicious accounts and extracted features. Based on this results, we detect the suspected accounts that spread rumors. Firstly, we crawled actual data and analyzed feature. And we selected feature as three approaches and added a new feature as propagation approach by existing work. That is user can re-tweet influencer’s tweet and edit it. We discussed it by ratio for RT. After that, we selected classification standard using average of data based on selected feature and trained it. Bayesian network is used for training. And the system may provide a new classification through re-analysis of the data. Proposed system is that the accuracy is 91.94% and F-measure is 93.76%.; The growing use of the smart device such as smartphones and tablets has resulted in increasing number of social network service(SNS) users recently. SNS allows a fast propagation and it is used as a tool to send information. But its negative sides need to be considered. In this paper, we analyzed actual data of malicious accounts and extracted features. Based on this results, we detect the suspected accounts that spread rumors. Firstly, we crawled actual data and analyzed feature. And we selected feature as three approaches and added a new feature as propagation approach by existing work. That is user can re-tweet influencer’s tweet and edit it. We discussed it by ratio for RT. After that, we selected classification standard using average of data based on selected feature and trained it. Bayesian network is used for training. And the system may provide a new classification through re-analysis of the data. Proposed system is that the accuracy is 91.94% and F-measure is 93.76%.
URI
https://repository.hanyang.ac.kr/handle/20.500.11754/126513http://hanyang.dcollection.net/common/orgView/200000428463
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > COMPUTER SCIENCE & ENGINEERING(컴퓨터공학과) > Theses (Master)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE