Large-Scale One-Class Collaborative Filtering:The Impact of Weighting Schemes
Date Issued
2015
Date
2015
Author(s)
Wang, Yu-Ting
Abstract
Recommendation systems have been widely used in e-commerce applications. With the development of information technology, users can easily reach enormous products. However, users have limited ability to evaluate their choices. Therefore, it’s important for content providers and e-retailers to recommend items which meet users’ taste to enhance user satisfaction and loyalty. Collaborative filtering is a popular way to implement a recommendation system. Collaborative filtering analyzes the relationships between users and items by users’ feedback which reflect users’ preferences. Then, it recommends user a ranked item list which is sorted by predicted preferences. This research focus on the One-class Collaborative Filtering (OCCF) approach. In OCCF, we only have positive examples to represent users’ actions. The data are ambiguous because unobserved data points can be interpreted as missing or negative cases. In this study, we treat unknown examples as negative examples with a confidence score, which is calculated by our weighting schemes. We apply our model on two large-scale movie rating datasets, and implement OCCF with gap-weighting Alternating Least Square (gALS). Then, we adjust weighting schemes to observe the impact on the model. Our result shows that gALS improves predicting performance. However, weighting strategies don’t make a dramatic impact.
Subjects
Recommendation systems
collaborative filtering
one-class collaborative filtering
alternative least square
matrix factorization
Netflix prize
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-104-R02725031-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):c3f90ac2a4365bf196d6a45599ae9a4b