A social recommender system using item asymmetric correlation

Recommender systems have been one of the most prominent information filtering techniques during the past decade. However, they suffer from two major problems, which degrade the accuracy of suggestions: data sparsity and cold start. The popularity of social networks shed light on a new generation of...

Full description

Saved in:
Bibliographic Details
Published in:Applied intelligence (Dordrecht, Netherlands) Netherlands), 2018-03, Vol.48 (3), p.527-540
Main Authors: Dakhel, Arghavan Moradi, Malazi, Hadi Tabatabaee, Mahdavi, Mehregan
Format: Article
Language:English
Subjects:
Artificial Intelligence
Asymmetry
Cold starts
Computer Science
Correlation
Filtration
Machines
Manufacturing
Mechanical Engineering
Processes
Recommender systems
Sparsity
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recommender systems have been one of the most prominent information filtering techniques during the past decade. However, they suffer from two major problems, which degrade the accuracy of suggestions: data sparsity and cold start. The popularity of social networks shed light on a new generation of such systems, which is called social recommender system. These systems act promisingly in solving data sparsity and cold start issues. Given that social relationships are not available to every system, the implicit relationship between the items can be an adequate option to replace the constraints. In this paper, we explored the effect of combining the implicit relationships of the items and user-item matrix on the accuracy of recommendations. The new Item Asymmetric Correlation (IAC) method detects the implicit relationship between each pair of items by considering an asymmetric correlation among them. Two dataset types, the output of IAC and user-item matrix, are fused into a collaborative filtering recommender via Matrix Factorization (MF) technique. We apply the two mostly used mapping models in MF , Stochastic Gradient Descent and Alternating Least Square, to investigate their performances in the presence of sparse data. The experimental results of real datasets at four levels of sparsity demonstrate the better performance of our method comparing to the other commonly used approaches, especially in handling the sparse data.
ISSN:0924-669X
1573-7497
DOI:10.1007/s10489-017-0973-5