DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Kim, Sung-Ho | - |
dc.contributor.advisor | 김성호 | - |
dc.contributor.author | Yoon, Heegeon | - |
dc.date.accessioned | 2019-09-03T02:44:47Z | - |
dc.date.available | 2019-09-03T02:44:47Z | - |
dc.date.issued | 2018 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=828530&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/266405 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 수리과학과, 2018.8,[iii, 19 p. :] | - |
dc.description.abstract | In this paper, we introduce a new matrix weighting scheme that is applied to a term-document matrix, which is an input matrix of documents required for running the $Na\"{i}ve$ Bayes method, as an effort to improve the accuracy of the $Na\"{i}ve$ Bayes method. We first examine two existing weighting strategies: Term Frequency - Inverse Document Frequency weighting and Golden Words weighting. Next, we present the new weighting method that incorporates the two existing methods with a slight modification in the algorithm. Then, we compare the accuracy of the $Na\"{i}ve$ Bayes method when the three different weighting schemes are applied to the term-document matrix. It is shown through simulation that the new method yields a greater degree of accuracy than the other two weighting methods. In addition, we set different values to the parameter in the new method and examine the change in accuracy. Finally, we find the optimal value of the parameter that maximizes the accuracy of the Na\"ive Bayes method. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | 나이브베이즈 방법▼a단어-문서 행렬▼a행렬 가중치▼a단어 빈도 - 역문서 빈도 가중치▼a핵심 단어 가중치 | - |
dc.title | (A) study on the Improvement of the $Na\"{i}ve$ Bayes method | - |
dc.title.alternative | 나이브베이즈 방법의 개선에 관한 고찰 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :수리과학과, | - |
dc.contributor.alternativeauthor | 윤희건 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.