A hybrid recommendation system with many-objective evolutionary algorithm
Introduction
Nowadays, with the increasing use of the network, the massive amounts of data are generated by people’s daily life, and it is a crucial issue that the rational use of data. Recommendation system (RS) (Bobadilla et al., 2013) is a crucial technique to analyze data reasonably and provide recommendations for users automatically (Jameson et al., 2013). At present, almost all applications use RS to provide more accurate recommendations for users to attract users to use their applications for a long time. These applications involve almost all areas of people's lives, including movies (Winoto and Tang, 2010), songs (Lee et al., 2010), books (Crespo and Ruben, 2011), news (Yeung and Yang, 2010), tourism (Cheng et al., 2012), and so on. The RS has gradually become an indispensable technology for people's online life.
Traditional recommendation algorithms mainly focus on providing accurate recommendations to users. And in recent years, some researchers have begun to study how to improve the accuracy and diversity of RS at the same time (Hurley and Mi, 2011, Vargas and Castells, 2011). However, with the increasing enrichment of people’s network life, user’s demand for RS is not limited to the accuracy and diversity of recommendations. In actual situations, RS usually needs to consider many factors, including accuracy, diversity, novelty, coverage, user satisfaction and so on. It can’t provide satisfactory recommendations for users when RS ignores the other indicators, such as novelty and coverage. At present, items with high ratings and popularity in the system are more likely to be recommended because of over-emphasis on the accuracy of recommendations. It means that the performance imbalances in recommendation diversity, novelty, coverage, etc. Therefore, it is a challenge for researchers to balance accuracy, diversity, novelty, and coverage at the same time.
Existing research using multi-objective evolutionary algorithms (MOEAs) to solve recommendation models only focuses on the accuracy and diversity of recommendations. However, the increase of the dimension of optimization objectives will lead to the poor performance of the MOEA in solving models with more than four optimization objectives (Wang and Huang, 2019). Because when the stranded MOEA solve the many-objective optimization problem (MaOP) with more than 4 optimization objectives, the selection pressure of the algorithm is not enough to support the solutions to move to the optimal position (Cui et al., 2019, Cui et al., 2019), and a large number of non-dominant solutions will be generated. Therefore, in order to simultaneously improve the accuracy, diversity, novelty and coverage of the recommendation, it is necessary to construct a many-objective optimization recommendation model and use MaOEAs to solve many-objective recommendation problems.
Generally, the RS needs to process an enormous amount of data and then make recommendations for users. It is time-consuming to solve recommendation problems with large amounts of data. Meanwhile, the generation of each recommendation list is accompanied by the running of the recommendation algorithm once. Therefore, it will take a lot of time for the algorithm runs many times to provide users with a variety of recommendations. The single recommendation technology is only suitable for solving simple recommendation problems, rather than complex MaOP. So, it’s necessary to use a hybrid RS to generate a variety of accurate recommendations for users.
In this paper, to address these issues of RS by using a new hybrid recommendation technology based on many-objective optimization, we use MaOEA (Li et al., 2014, Yang et al., 2013) to optimize coefficients to make the three basic recommendation algorithms, which have their applicable scenarios, combined in suitable proportions. This model can effectively balance accuracy, diversity, novelty, and coverage at the same time. Moreover, each run of the algorithm will produce a set of coefficients with good results, which means it will reasonably absorb the advantages of the three basic algorithms and produce a variety of recommendations with a brilliant performance. Besides, the proposed RS recommends items for users according to the predicted user's rating, so the result of recommendation is more objective.
The remainder of this paper is organized as follows. Section 2 reviews related work. In Section 3, we describe the framework and model of the proposed hybrid RS based on MaOEA. Section 4 shows the technologies and calculation methods used in this model in detail. In Section 5, we show the settings and results of the experiment. Finally, the conclusion is given in Section 6.
Section snippets
Related work
In this section, we introduce the related research regarding RS, including some basic recommendation technologies and hybrid recommendation algorithms, and then briefly introduce the related work of RS based on optimization algorithm and optimization algorithm.
Personalized recommendation method (Cui, Xu et al., 2020) is an important part of the recommendation algorithm. Personalized recommendation methods mainly include content-based recommendation method (Lops et al., 2011) and collaborative
The proposed hybrid RS based on MaOEA
In this section, we present the hybrid RS with MaOEA. This recommendation algorithm can combine three basic recommendation technologies and optimize the four objectives (accuracy, diversity, novelty, and coverage) to improve the recommended performance of the RS. We will describe the proposed hybrid RS based on MaOEA in detail, including the framework of hybrid RS and the proposed hybrid model.
Methods and implementation
In this section, we show the methods and techniques used to implement the model in detail, including the four optimization objectives, data reduction, and RVEA.
Experimental settings
In this paper, MovieLens as one of the famous classical benchmark data set is used to demonstrate the capability of the proposed method in RS. We choose “MovieLens 1 M Dataset” that contains 1,000,209 ratings from 6040 MovieLens users on 3952 movies (it can be downloaded from https://grouplens.org/datasets/movielens/). This benchmark data set contains ratings on a five-point scale. At the same time, according to a binary rating system (“like” or “dislike”), and user ratings of items more than 3
Conclusions
Traditional recommendation algorithms can’t provide users with more novel and diverse recommendations, so a hybrid many-objective recommendation model is presented to optimize accuracy, diversity, novelty, and coverage. And RVEA was employed to solve the modeled MaOP for getting better recommendations. Different from the previous Top-N strategy, a strategy based on prediction ratings were used to provide recommendations for users, which combines three basic recommendation technologies. In
CRediT authorship contribution statement
Xingjuan Cai: Conceptualization, Methodology, Writing - original draft. Zhaoming Hu: Methodology, Software, Writing - original draft. Peng Zhao: Methodology, Writing - original draft. WenSheng Zhang: Software, Supervision. Jinjun Chen: Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This work is supported by the National Natural Science Foundation of China under Grant No. 61806138, No. U1636220 and No. 61663028, Natural Science Foundation of Shanxi Province under Grant No. 201801D121127, Key R&D program of Shanxi Province (High Technology) under Grant No. 201903D121119. Special thanks should go to my partner Penghong Wang who has put considerable time and effort into their comments on this paper.
References (46)
- et al.
Improving collaborative filtering recommender system results and performance using genetic algorithms
Knowledge-Based Systems
(2011) - et al.
Recommender systems survey
Knowledge-Based Systems
(2013) - et al.
A novel multi-objective evolutionary algorithm for recommendation systems
Journal of Parallel and Distributed Computing
(2017) - et al.
Improved NSGA-III with selection-and-elimination operator
Swarm and Evolutionary Computataion
(2019) - et al.
Malicious code detection based on CNNs and multi-objective algorithm
Journal of Parallel and Distributed Computing
(2019) - et al.
Hybrid many-objective particle swarm optimization algorithm for green coal production problem
Information Sciences
(2020) - et al.
NNIA-RS: A multi-objective optimization based recommender system
Physica A
(2015) - et al.
Adaptive discrete cat swarm optimisation algorithm for the flexible job shop problem
International Journal of Bio-Inspired Computation
(2019) - et al.
A recommender system using GA-means clustering in an online shopping market
Expert Systems with Applications
(2008) - et al.
Collaborative filtering with ordinal scale-based implicit ratings for mobile music recommendations
Information Sciences
(2010)
The role of user mood in movie recommendations
Expert Systems with Applications
Using kernel-based collocation methods to solve a delay partial differential equation with application to finance
International Journal of Computing Science and Mathematics
A reference vector guided evolutionary algorithm for many-objective optimization
IEEE Transactions on Evolutionary Computation
The adaptive recommendation mechanism for distributed group in mobile environments
IEEE Transactions on Systems Man & Cybernetics Part C
Bat algorithm with principal component analysis
International Journal of Machine Learning and Cybernetics
A pigeon-inspired optimization algorithm for many-objective optimization problems. SCIENCE CHINA
Information Sciences
An evolutionary many-objective optimization algorithm using reference-point-based nondominated sorting approach
Part I: Solving Problems With Box Constraints.
Using collaborative filtering to weave an information tapestry
Communications of the ACM
Cited by (98)
Bandit algorithms: A comprehensive review and their dynamic selection from a portfolio for multicriteria top-k recommendation
2024, Expert Systems with ApplicationsMulti-objective optimization and integrated indicator-driven two-stage project recommendation in time-dependent software ecosystem
2024, Information and Software TechnologyA novel self-supervised graph model based on counterfactual learning for diversified recommendation
2024, Information SystemsProvably secure lightweight certificateless lattice-based authenticated key exchange scheme for IIoT
2024, Concurrency and Computation: Practice and ExperienceDesign of novel deformable mirror with large displacement
2024, Concurrency and Computation: Practice and Experience