Multiobjective recommendation optimization via utilizing distributed parallel algorithm

With the development of information technologies, various big data problems are emerging. The recommendation problem can be seen as a big data problem. Traditionally, for a recommender system (RS), only the recommendation precision is considered. However, reflecting another aspect of RS, recommendat...

Full description

Saved in:
Bibliographic Details
Published in:Future generation computer systems Vol. 86; pp. 1259 - 1268
Main Authors: Cao, Bin, Zhao, Jianwei, Liu, Xin, Kang, Xinyuan, Yang, Shan, Kang, Kai, Yu, Ming
Format: Journal Article
Language:English
Published: Elsevier B.V 01.09.2018
Subjects:
Diversity
Multiobjective
Parallelism
Precision
Recommender system (RS)
Multiobjective
Recommender system (RS)
Parallelism
Precision
Diversity
ISSN:0167-739X, 1872-7115
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:With the development of information technologies, various big data problems are emerging. The recommendation problem can be seen as a big data problem. Traditionally, for a recommender system (RS), only the recommendation precision is considered. However, reflecting another aspect of RS, recommendation diversity is also important. In this paper, we adopt a multiobjective recommendation model to simultaneously consider recommendation precision and diversity, specifically, precision, novelty and coverage of recommendation are involved. To tackle the multiobjective recommendation optimization problem (MROP), based on distributed parallel cooperative coevolutionary multiobjective evolutionary algorithm (DPCCMOEA), a novel multiobjective evolutionary algorithm (MOEA), DPCCMOEA for RSs (DPCCMOEA-RecSys) is proposed. On the basis of cooperative coevolution (CC) framework, all users are allocated to several groups and are optimized simultaneously. Optimization strategy specific for RSs is put forward, the individual integration approach is explored and different grouping techniques are compared and analyzed. Compared to state-of-the-art cooperative coevolutionary MOEAs: cooperative coevolutionary generalized differential evolution 3 (CCGDE3), multiobjective evolutionary algorithm based on decision variable analyses (MOEA/DVA) and DPCCMOEA, DPCCMOEA-RecSys can achieve better optimization results; compared to serial algorithms: CCGDE3 and MOEA/DVA, DPCCMOEA and DPCCMOEA-RecSys significantly reduce the time consumption. •On the basis of a two-layer distributed parallel structure, a recommender system (RS)-specific distributed parallel multiobjective evolutionary algorithm is proposed to simultaneously optimize recommendation precision and diversity.•Users are allocated to several groups, and are optimized under the cooperative coevolutionary framework, for which, different grouping methods are examined.•The individual integration approach of the proposed algorithm is studied and verified.•A RS-specificoptimization strategy is designed to guarantee the availability of the recommendation list.
ISSN:0167-739X
1872-7115
DOI:10.1016/j.future.2017.09.005