RecSys ’16: Local Item-Item Models for Top-N Recommendation
Recommendation system paper challenge (11/50)
What problem do they solve?
A top-n recommendation system
What others solve this problem?
There are two main stream to solve this problem: content-based and collaborative filtering.
Most of researchers applied SVD family to solve this problem. The best one is SVD++, SLIM.
What model do they propose?
LGSLIM: combining global SLIM and local SLIM together.
g_u lies in [0, 1] control what ratio to use global SLIM (s_li) and local SLIM (s_pu_li).
global SLIM computed as the following formula:
more detail is in this paper
local SLIM computed as the following formula:
The local SLIM is similar with global SLIM, only difference is that model assign each user to k subsets and optimized the formula in the subset.
We may ask, how to decide the subset user assignment ?
They initialized the assignment with CLUTO algorithm.
They apply ALS (Alternating Least Square)
- After they fixed the model, they update the user groups.
- Fixed the user groups, they update the model parameters.
What data do they test?
What metric do they measure?
HR: hit rate
ARHR: average-reciprocal hit rank
Here is an example for ARHR:
If we have 2 users
first user hit on position 2
second user hit on position 4
ARHR = (1/2) * (1/2 + 1/4) = 3/8
What is their baseline model?
LSLIMr0: local SLIM without updating user group
LSLIM: local SLIM with updating user group
GLSLIMr0: GLSLIM without updating user group
The result
Other related blogs:
Trust-aware recommender systems
Performance of recommender algorithms on top-n recommendation tasks
A Matrix Factorization Technique with Trust Propagation for Recommendation in Social Networks
Hidden Factors and Hidden Topics: Understanding Rating Dimensions with Review Text
Beyond Clicks: Dwell Time for Personalization
RecSys’15: Context-Aware Event Recommendation in Event-based Social Networks
RecSys’11: Utilizing related products for post-purchase recommendation in e-commerce
RecSys11: OrdRec: an ordinal model for predicting personalized item rating distributions
RecSys16: Adaptive, Personalized Diversity for Visual Discovery
Best paper in RecSys:
https://recsys.acm.org/best-papers/
My Website: