recommendation system - kungfumas.files.wordpress.com · i content data : analisis tekstual dari...
TRANSCRIPT
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommendation System
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Apa recommendation system (sistem rekomendasi)?Contoh?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Apa recommendation system (sistem rekomendasi)?Contoh?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommender system
Memberikan rekomendasi, biasanya barang, kepada orang (user)Misalnya :
I Kamera digital mana yang harus saya beli?
I Akun twitter mana yang akan harus saya follow?
I Universitas mana yang tepat bagi saya?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommender system
Memberikan rekomendasi, biasanya barang, kepada orang (user)Misalnya :
I Kamera digital mana yang harus saya beli?
I Akun twitter mana yang akan harus saya follow?
I Universitas mana yang tepat bagi saya?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommender system
Memberikan rekomendasi, biasanya barang, kepada orang (user)Misalnya :
I Kamera digital mana yang harus saya beli?
I Akun twitter mana yang akan harus saya follow?
I Universitas mana yang tepat bagi saya?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommender system
Memberikan rekomendasi, biasanya barang, kepada orang (user)Misalnya :
I Kamera digital mana yang harus saya beli?
I Akun twitter mana yang akan harus saya follow?
I Universitas mana yang tepat bagi saya?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommender system
Kegunaan recommender system :
I Mempersempit information overload
I Kegunaan bagi user : mendapatkan hal yang menarik,mempersempit pilihan, menemukan hal yang baru
I Kegunaan bagi provider : memberikan rekomendasi yang lebihpersonal kepada user-nya, meningkatkan loyalitas user,meningkatkan pembelian, peluang untuk promosi,mendapatkan pengetahuan tentang user-nya
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommender system
Kegunaan recommender system :
I Mempersempit information overload
I Kegunaan bagi user : mendapatkan hal yang menarik,mempersempit pilihan, menemukan hal yang baru
I Kegunaan bagi provider : memberikan rekomendasi yang lebihpersonal kepada user-nya, meningkatkan loyalitas user,meningkatkan pembelian, peluang untuk promosi,mendapatkan pengetahuan tentang user-nya
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommender system
Kegunaan recommender system :
I Mempersempit information overload
I Kegunaan bagi user : mendapatkan hal yang menarik,mempersempit pilihan, menemukan hal yang baru
I Kegunaan bagi provider : memberikan rekomendasi yang lebihpersonal kepada user-nya, meningkatkan loyalitas user,meningkatkan pembelian, peluang untuk promosi,mendapatkan pengetahuan tentang user-nya
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommenders vs Search Engines
I Search engines bukan sebuah sistem rekomendasi
I Query untuk mencari rekomendasi pada search enginemenghasilkan kumpulan sistem rekomendasi
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommenders vs Search Engines
I Search engines bukan sebuah sistem rekomendasi
I Query untuk mencari rekomendasi pada search enginemenghasilkan kumpulan sistem rekomendasi
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Recommenders vs Search Engines
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Kategori Recommendation System
Content based filtering: ”Rekomendasikan buku yang sesuaidengan tipe buku yang saya suka”. Biasanyamenggunakan fitur barang.
Collaborative filtering: ”Rekomendasikan buku yang disukai olehteman-teman saya”. Menggunakan preferensikomunitas.
Hybrid: Kombinasi dari CF dan content-based
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Kategori Recommendation System
Content based filtering: ”Rekomendasikan buku yang sesuaidengan tipe buku yang saya suka”. Biasanyamenggunakan fitur barang.
Collaborative filtering: ”Rekomendasikan buku yang disukai olehteman-teman saya”. Menggunakan preferensikomunitas.
Hybrid: Kombinasi dari CF dan content-based
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Kategori Recommendation System
Content based filtering: ”Rekomendasikan buku yang sesuaidengan tipe buku yang saya suka”. Biasanyamenggunakan fitur barang.
Collaborative filtering: ”Rekomendasikan buku yang disukai olehteman-teman saya”. Menggunakan preferensikomunitas.
Hybrid: Kombinasi dari CF dan content-based
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Input Recommendation System
(Vozalis & Margaritis, 2003) menyatakan input sistem rekomendasiada tiga:
I Demographic data : umur, jenis kelamin, dll
I Content data : analisis tekstual dari barang-barang yg pernahdibeli oleh user
I Ratings : scalar, binary, unary. Selain itu ada juga ratingimplisit dan eksplisit.
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Input Recommendation System
(Vozalis & Margaritis, 2003) menyatakan input sistem rekomendasiada tiga:
I Demographic data : umur, jenis kelamin, dll
I Content data : analisis tekstual dari barang-barang yg pernahdibeli oleh user
I Ratings : scalar, binary, unary. Selain itu ada juga ratingimplisit dan eksplisit.
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Input Recommendation System
(Vozalis & Margaritis, 2003) menyatakan input sistem rekomendasiada tiga:
I Demographic data : umur, jenis kelamin, dll
I Content data : analisis tekstual dari barang-barang yg pernahdibeli oleh user
I Ratings : scalar, binary, unary. Selain itu ada juga ratingimplisit dan eksplisit.
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Input Recommendation System
(Vozalis & Margaritis, 2003) menyatakan input sistem rekomendasiada tiga:
I Demographic data : umur, jenis kelamin, dll
I Content data : analisis tekstual dari barang-barang yg pernahdibeli oleh user
I Ratings : scalar, binary, unary. Selain itu ada juga ratingimplisit dan eksplisit.
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Collaborative Filtering
”wisdom of crowd”
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : Keuntungan
(Melville et al., 2002) menyebutkan dua keuntungan CF:
I bisa digunakan pada domains dimana di dalamnya terdapatcontent yangx tidak berhubungan dengan items
I serendipitious recommendations
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : Keuntungan
(Melville et al., 2002) menyebutkan dua keuntungan CF:
I bisa digunakan pada domains dimana di dalamnya terdapatcontent yangx tidak berhubungan dengan items
I serendipitious recommendations
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : Keuntungan
(Melville et al., 2002) menyebutkan dua keuntungan CF:
I bisa digunakan pada domains dimana di dalamnya terdapatcontent yangx tidak berhubungan dengan items
I serendipitious recommendations
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : InputInput dari CF : matriks user-item ratings
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : Output
Output dari CF : top-N list, prediksi rating score
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : Metode
Nearest Neighbour
I user-to-user
I item-to-item
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : Metode
Nearest Neighbour
I user-to-user
I item-to-item
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : User-to-user
Terdapat sebuah matriks berisi rating. Kira-kira berapa ratingAlice untuk item5?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF User-to-user : Pearson correlation coefficientBagaimana caranya mengukur kesamaan antar user?Solusi:
Pearson correlation coefficientsim(a, b) = ∑
p∈P(ra,p − ra)(rb,p − rb)√∑p∈P(ra,p − ra)2
√∑p∈P(rb,p − rb)2
Variables
I a, b: users
I P: items, yang sudah dirating oleh a dan b
I ra,p: rating dari user a untuk item p
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF User-to-user : Menentukan prediksi
Weighted normalized adjusted average
pred(a, b) =
ra +
∑b∈N sim(a, b) ∗ (rb,p − rb)∑
b∈N sim(a, b)
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF User-to-user : Menentukan prediksi
Lainnya :
I Simple average
I Weighted average
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF User-to-user : Menentukan prediksi
Jika kita ambil 2-Nearest-Neighbour
PAlice,item5 = 3+42 = 3.5
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF : Item-to-item
Memberikan prediksi berdasarkan kesamaan antar barang (items)
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF Item-to-item : Cosine similarity
Kesamaan dihitung dengan cosine similarity (yang paling seringdigunakan), dimana ratings dianggap sebagai vektor
Cosine similarity
sim(−→a ,−→b ) =
−→a .−→b
|−→a |∗|−→b |
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF Item-to-item : Cosine similarity
Opsi lain adalah adjusted cosine similarity yang menggunakanrata-rata ratings dari user untuk mengubah rating awal
Adjusted Cosine similarity
sim(−→a ,−→b ) = ∑
u∈U(ru,a − ru)(ru,b − ru)√∑u∈U(ru,a − ru)2
√∑u∈U(ru,b − ru)2
Variables
I U: satu set users yang sudah melakukan rating terhadapitems a dan b
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF Item-to-item : Menentukan prediksi
Prediction functionpred(u, p) = ∑
i∈ratedItems(u) sim(i , p) ∗ ru,i∑i∈ratedItem(u) sim(i , p)
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
CF Item-to-item : Menentukan prediksi
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Content-based
I Membutuhkan informasi barang, tidak perlu informasikomunitas
I Dibutuhkan : informasi tentangg konten barang informasitentang apa yg disukai oleh user
I Biasanya dipakai untuk text-based, misalnya berita ¿¿ textclassification
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Content-based
I Membutuhkan informasi barang, tidak perlu informasikomunitas
I Dibutuhkan : informasi tentangg konten barang informasitentang apa yg disukai oleh user
I Biasanya dipakai untuk text-based, misalnya berita ¿¿ textclassification
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Content-based
I Membutuhkan informasi barang, tidak perlu informasikomunitas
I Dibutuhkan : informasi tentangg konten barang informasitentang apa yg disukai oleh user
I Biasanya dipakai untuk text-based, misalnya berita ¿¿ textclassification
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Content-based : Keuntungan
Keuntungan dari content-based:
I komunitas tidak diperlukan
I lebih mudah daripada CF?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Content-based : Keuntungan
Keuntungan dari content-based:
I komunitas tidak diperlukan
I lebih mudah daripada CF?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Content-based : Keuntungan
Keuntungan dari content-based:
I komunitas tidak diperlukan
I lebih mudah daripada CF?
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Content-based : tf-idf
Karena merupakan text classification maka digunakan tf-idf danclassifiers klasik seperti Naive Bayes dan SVM
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Evaluasi Recommendation System
Statistical accuracy metrics: Mean Absolute Error (MAE), RootMean Squared Error (RMSE)
Decision support accuracy: biasanya untuk data binary,menunjukkan kualitas dari barang yangdirekomendasi : Precision, Recall, ROC
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Evaluasi Recommendation System
Statistical accuracy metrics: Mean Absolute Error (MAE), RootMean Squared Error (RMSE)
Decision support accuracy: biasanya untuk data binary,menunjukkan kualitas dari barang yangdirekomendasi : Precision, Recall, ROC
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Introduction
Recommenders vs Search Engines
Kategori recommendation system
Input recommendation system
Collaborative Filtering
Content-based
Evaluasi recommendation system
Masalah dalam recommendation system
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Masalah dalam Recommendation System
Data sparsity Solusi? SVD (Singular Value Decomposition)
Cold start First rater problem
Shilling attacks Biased ratings
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Masalah dalam Recommendation System
Data sparsity Solusi? SVD (Singular Value Decomposition)
Cold start First rater problem
Shilling attacks Biased ratings
Recommendation System
Introduction Recommenders vs Search Engines Kategori recommendation system Input recommendation system Collaborative Filtering Content-based Evaluasi recommendation system Masalah dalam recommendation system
Masalah dalam Recommendation System
Data sparsity Solusi? SVD (Singular Value Decomposition)
Cold start First rater problem
Shilling attacks Biased ratings
Recommendation System