성능 분석 결과

LDA 성능 평가

수치로 평가되는 Coherence 및 Perplexity 지표는 다음과 같다.

Coherence: -2.4954, Perplexity: -8.6283

해당 모델의 토픽을 시각화하여 나타낸 그래프이다.

Topic 리스트

>> Topic_list
(0, '0.077*"학점" + 0.054*"전공" + 0.050*"신청" + 0.042*"학기" ,...
(1, '0.027*"공모전" + 0.024*"작품" + 0.021*"접수" + 0.020*"영상",...
(2, '0.081*"영어" + 0.071*"시험" + 0.040*"토익" + 0.033*"공부",...
(3, '0.038*"취업" + 0.029*"지원" + 0.026*"기업" + 0.023*"교육" ,...
(4, '0.028*"세종대" + 0.018*"연구" + 0.016*"교수" + 0.015*"학생" ,...
(5, '0.347*"커뮤니티" + 0.189*"자유" + 0.040*"새내기" ,...
(6, '0.024*"총학생회" + 0.019*"학술" + 0.017*"정보원" + 0.015*"학생" ,...
(7, '0.032*"university" + 0.028*"파견" + 0.022*"학생" +,...
(8, '0.033*"점검" + 0.026*"음악" + 0.024*"장터" + 0.020*"공연" ,...
(9, '0.058*"기숙사" + 0.031*"학생" + 0.029*"신청" + 0.029*"입사" +,...
(10, '0.083*"교수" + 0.056*"교양" + 0.042*"학과" + 0.037*"학부",...
(11, '0.056*"정보 산업" + 0.026*"스포츠" + 0.021*"운동" + 0.017*"서울시",..
(12, '0.010*"일본" + 0.008*"한국" + 0.007*"여행" + 0.006*"노력",...
(13, '0.091*"신청" + 0.059*"수강" + 0.055*"학기" + 0.040*"과목" ,...
(14, '0.011*"기말" + 0.011*"korea" + 0.008*"students" ,...
(15, '0.068*"추천" + 0.052*"공지" + 0.043*"학생" + 0.032*"설명회" + ,...
(16, '0.067*"봉사" + 0.051*"활동" + 0.037*"세종" + 0.029*"학생" ,...
(17, '0.113*"면접" + 0.038*"질문" + 0.022*"면접관" + 0.019*"지원자" ,...
(18, '0.031*"소식" + 0.031*"서울" + 0.029*"방송국" + 0.028*"news",...
(19, '0.072*"대회" + 0.042*"독서" + 0.038*"고전" + 0.024*"인증" ,...
(20, '0.072*"모집" + 0.047*"동아리" + 0.035*"동아리&모임" + 0.033*"활동" ,..
(21, '0.041*"호텔" + 0.040*"경영" + 0.038*"관광" + 0.031*"네이버카페",...
(22, '0.018*"철회" + 0.017*"사물함" + 0.016*"신청" + 0.015*"강철" ,...
(23, '0.044*"근무" + 0.032*"지원" + 0.032*"경력" + 0.028*"기업" ,...
(24, '0.060*"장학금" + 0.056*"학생" + 0.051*"학기" + 0.033*"장학",...
(25, '0.065*"제출" + 0.049*"서류" + 0.047*"지원" + 0.030*"휴학" ,...

토픽 분석 예시

>>> get_topics("아 장학금 받고 싶다 ㅠㅠ")
array([0.01923082, 0.01923082, 0.01923082, 0.01923082, 0.01923082,
       0.01923082, 0.01923082, 0.01923082, 0.01923082, 0.01923082,
       0.01923082, 0.01923082, 0.01923082, 0.01923082, 0.01923082,
       0.01923082, 0.01923082, 0.01923082, 0.01923082, 0.01923082,
       0.01923082, 0.01923082, 0.01923082, 0.01923082, 0.51922953,
       0.01923082])

>>> get_topics(doc)
array([0.        , 0.        , 0.        , 0.        , 0.        ,     "
       0.        , 0.        , 0.        , 0.        , 0.        ,
       0.        , 0.04859713, 0.        , 0.        , 0.05291525,
       0.        , 0.        , 0.        , 0.        , 0.        ,
       0.        , 0.        , 0.        , 0.86571997, 0.        ,
       0.        ])

Word2Vec(FastText) 성능 평가

워드투벡터의 경우, 수치로서 표현되는 결과가 존재하지 않기 때문에 모델의 시각화 및 각 단어별 케이스 테스트를 실시하여 성능을 확인하였다.

Visualization

유사 단어 측정 예시

Keyword: 파이썬

Keyword: 덮밥