Authors:
Sofiane Abbar, Habibur Rahman, Saravanan Thirumuruganathan, Carlos Castillo, Gautam Das
Organization:
Qatar Computing Research Institute (卡達,阿拉伯半島中的國家)
University of Texas at Arlington
IEEE International Conference on Data Engineering, 2014
2. Paper title:
Ranking Item Features by Mining
Online User-Item Interactions
Authors:
Sofiane Abbar, Habibur Rahman, Saravanan
Thirumuruganathan, Carlos Castillo, Gautam Das
Organization:
Qatar Computing Research Institute (卡達,阿拉伯半島中的國家)
University of Texas at Arlington
IEEE International Conference on Data Engineering, 2014
4. Features
Features like “movie actor, Tom Hanks“, “movie
actor, Catherine Zeta-Jones”
Each element in feature matrix is a probability.
By the method with features, we can predict user
preference on new items.
5. Database with 3 features and 2
items.
To be given
Features Items Users
6. Terminology
H: User-Feature Preferences Matrix
h: for single user
W: Feature-Item Transition Matrix
Each element is a probability.
𝑊: Boolean feature-item matrix
ℎ =
100
202
101
202
1
202
𝑊 =
1 0.5 0
0 0.5 0.1
𝑊 =
1 1 0
0 1 1
7. User Interactions Model
𝑊ℎ = 𝑣
The most important case for me is that h is
unknown.
“𝑊ℎ” is not always equal to “v”.