TF-IDF

From CS Wiki
Revision as of 08:26, 28 December 2019 by 김형교 (talk | contribs) (새 문서: 분류:인공지능 ;Term Frequency - Inverse Document Frequency == 예시 === === TF Score === * a new car, used car, car review ** TF Score를 통해 car가 중요한 단어라...)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Term Frequency - Inverse Document Frequency

예시 =

TF Score

  • a new car, used car, car review
    • TF Score를 통해 car가 중요한 단어라는 것을 확인하였음
단어 TF Score
a 1/7
new 1/7
car 3/7
used 1/7
review 1/7
  • a friend in need is a friend indeed
    • TF Score를 측정했는데 Friend와 a가 동일한 중요도로 산출됨
    • 이런 경우를 보완하기 위해 IDF 사용
단어 TF Score
a 2/8
friend 2/8
in 3/8
need 1/8
is 1/8
indeed 1/8

IDF

  • Log (이 단어가 사용된 문장의 수 / 총 문장의 수+1)
  • TF에 IDF를 적용하면, a/the/in/is 와 같은 의미 없는 불용어를 희석시킬 수 있다.
단어 TF Score IDF Score TF * IDF
a 1/7 Log(2/2) = 0 0
new 1/7 Log(2/1) = 0.3 0.04
car 3/7 Log(2/1) = 0.3 0.13
used 1/7 Log(2/1) = 0.3 0.04
review 1/7 Log(2/1) = 0.3 0.04
a 2/8 Log(2/2) = 0 0
friend 2/8 Log(2/1) = 0.3 0.08
in 1/8 Log(2/1) = 0.3 0.04
need 1/8 Log(2/1) = 0.3 0.04
is 1/8 Log(2/1) = 0.3 0.04
indeed 1/8 Log(2/1) = 0.3 0.04