E D R , A S I H C RSS
ZeroWiki

Collaborative Filtering

ν˜‘μ—… (μƒν˜Έν˜‘λ™) 필터링, Recommender System이라고도 뢈림. ProjectPrometheusμ—μ„œ μ‚¬μš©ν•œλ‹€.

Approaches to Collaborative Filtering

problem spaceκ°€ 2차원 matrix 의 ν˜•νƒœλ₯Ό 생각해본닀. 행에 λŒ€ν•΄μ„œλŠ” item을, 열에 λŒ€ν•΄μ„œλŠ” userλ₯Ό 두고, 그에 λ”°λ₯Έ rating 을 κ°’μœΌλ‘œ λ‘”λ‹€. 이 matrix λ₯Ό 이용, CollaborativeFiltering 은 νŠΉμ • μ‚¬μš©μž(user) i 에 λŒ€ν•΄μ„œ rating 을 μ˜ˆμΈ‘ν•˜κ³ , item 듀을 μΆ”μ²œν•œλ‹€.

보톡 λ‹€μŒμ˜ 과정을 가진닀.
  1. user인 i 와 λΉ„μŠ·ν•œ ν₯λ―Έλ‚˜ μ„ ν˜Έλ„λ₯Ό κ°€μ§€λŠ” μ‚¬μš©μž 집합(user set)을 μ„ νƒν•œλ‹€.
  2. 1 μ—μ„œ μ„ νƒλœ user set μœΌλ‘œλΆ€ν„° user i μ—κ²Œ μ ν•©ν•œ μΆ”μ²œ item 을 μ˜ˆμΈ‘ν•΄λ‚Έλ‹€.
    ex) 이 user set μ—μ„œ item j 에 λŒ€ν•΄μ„œ 높은 점수 (rating)을 μ£Όμ—ˆμ„ 경우, user i μ—κ²Œ item j λ₯Ό μΆ”μ²œν•œλ‹€.

Algorithms

  1. ν˜„μž¬ μ΄μš©μ€‘μΈ user 와 λΉ„μŠ·ν•œ μ·¨ν–₯의 μ‚¬μš©μž 집합을 선택 - calculate user correlation
  2. 1μ—μ„œ μ„ νƒλœ μ‚¬μš©μž 집합쀑 μ˜ˆμΈ‘μ„ μœ„ν•œ 뢀뢄집합 선택 - neighbourhood selection
  3. μ„ νƒλœ neighbours λ“€κ³Ό 자료λ₯Ό 근거둜 예츑 - generate a prediction

Calculate user correlation

  • Pearson correlation
  • Constrained Pearson correlation
  • The Spearman rank correlation
  • The Vector similarity
  • Emtropy-based uncertainty measure
  • Mean-square difference algorithm

Neighbourhood Selection

  • Correlation thresholding
  • Best-n correlations

Generate a prediction



Metrics

CollaborativeFiltering 의 μœ μš©μ„±μ„ ν‰κ°€ν•˜λŠ” κΈ°μ€€.
  • Coverage - μ‹œμŠ€ν…œμ΄ 좔어진 itemμœΌλ‘œλΆ€ν„° μΆ”μ²œ item 을 μ œκ³΅ν•΄μ£ΌλŠ” λŠ₯λ ₯. (μ•„λ§ˆλ„ μΆ”μ²œitemν’ˆλͺ©μˆ˜/전체itemν’ˆλͺ©μˆ˜ 에 λŒ€ν•œ νΌμ„ΌνŠΈμ΄λ €λ‚˜. μˆ˜μ • ν•„μš”)
  • Accuracy - μ‹œμŠ€ν…œμ΄ μΆ”μ²œν•œ item 에 λŒ€ν•œ μ •ν™•μ„± (μΆ”μ²œ item에 λŒ€ν•œ μ‚¬μš©μžμ˜ 점수)

κ΄€λ ¨ μžλ£Œλ“€

Valid XHTML 1.0! Valid CSS! powered by MoniWiki
last modified 2021-02-07 05:22:57
Processing time 0.0277 sec