For the following ratings given to items by the users, perform the tasks below:
Item 1 Item 2 Item 3 Item 4 Item 5
User 1 ? 3 ? 3 4
User 2 4 ? ? 2 ?
User 3 ? ? 3 ? ?
User 4 3 ? 4 ? 3
User 5 4 3 ? 4 4
Note: '?' means user has not rated that item.
For all of these tasks, use Pearson Correlation Coefficient as similarity measure and mean-centered prediction function.
Show your working clearly for :
a) Find nearest neighbors of user 1 for k=2.
b) Apply user-based Collaborative Filtering to predict R(U1, I3) (read: rating for user 1 and item 3).
c) Apply item-based Collaborative Filtering to predict R(U3, I2) and R(U4, I4). (HINT: USE CASCADING)
d) Say we had an option of using mode-centered prediction function instead of using mean. What do you think would be the effect of using mode as a baseline instead of average? Can using mode ever improve the performance of recommender system?
For the following ratings given to items by the users, perform the tasks below:
Item 1 Item 2 Item 3 Item 4 Item 5
User 1 ? 3 ? 3 4
User 2 4 ? ? 2 ?
User 3 ? ? 3 ? ?
User 4 3 ? 4 ? 3
User 5 4 3 ? 4 4
Note: '?' means user has not rated that item.
For all of these tasks, use Pearson Correlation Coefficient as similarity measure and mean-centered prediction function.
Show your working clearly for :
a) Find nearest neighbors of user 1 for k=2.
b) Apply user-based Collaborative Filtering to predict R(U1, I3) (read: rating for user 1 and item 3).
c) Apply item-based Collaborative Filtering to predict R(U3, I2) and R(U4, I4). (HINT: USE CASCADING)
d) Say we had an option of using mode-centered prediction function instead of using mean. What do you think would be the effect of using mode as a baseline instead of average? Can using mode ever improve the performance of recommender system?