A hybrid recommendation system with many-objective evolutionary algorithm

https://doi.org/10.1016/j.eswa.2020.113648Get rights and content

Highlights

  • Recommend the more and novel items based on accurate and diverse recommendations.

  • Mixing multiple recommendation technologies to improve recommendation performance.

  • The system is based on rating, which makes the recommendation more objective.

  • Clustering strategies are used to reduce recommended consumption.

Abstract

Recommendation system (RS) is a technology that provides accurate recommendations to users. However, it is not comprehensive to only consider the accuracy of the recommendation because users have different requirements. To improve the comprehensive performance, this paper presents a hybrid recommendation model based on many-objective optimization, which can simultaneously optimize the accuracy, diversity, novelty and coverage of recommendation. This model enhances the robustness of recommendations by mixing three different basic recommendation technologies. Additionally, we solve it with many-objective evolutionary algorithm (MaOEA) and test it extensively. Experimental results demonstrate the effectiveness of the presented model, which can provide the recommendations with more and novel items on the basis of accurate and diverse.

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)

  • P. Winoto et al.

    The role of user mood in movie recommendations

    Expert Systems with Applications

    (2010)
  • H. Azari et al.

    Using kernel-based collocation methods to solve a delay partial differential equation with application to finance

    International Journal of Computing Science and Mathematics

    (2019)
  • BILGE, Alper, POLAT, & Huseyin. (2013). A scalable privacy-preserving recommendation scheme via bisecting k-means...
  • Cai X., Niu Y., Geng S., Zhang J. , Cui Z. , Li J. , & J., C. (2019). An under-sampled software defect prediction...
  • Cai X., Wang P., Du L., Cui Z., Zhang W., & J., C. (2019). Multi-Objective Three-Dimensional DV-Hop Localization...
  • R. Cheng et al.

    A reference vector guided evolutionary algorithm for many-objective optimization

    IEEE Transactions on Evolutionary Computation

    (2016)
  • S.T. Cheng et al.

    The adaptive recommendation mechanism for distributed group in mobile environments

    IEEE Transactions on Systems Man & Cybernetics Part C

    (2012)
  • Crespo, G., Ruben, Martinez, S., Oscar, Lovelle, C., Manuel, J., . . . Emilio, J. (2011). Recommendation System based...
  • Z. Cui et al.

    Bat algorithm with principal component analysis

    International Journal of Machine Learning and Cybernetics

    (2019)
  • Z. Cui et al.

    A pigeon-inspired optimization algorithm for many-objective optimization problems. SCIENCE CHINA

    Information Sciences

    (2019)
  • Cui Z., Xu X., Xue F., Cai X., Cao Y., Zhang W., & J., C. (2020). Personalized Recommendation System based on...
  • K. Deb et al.

    An evolutionary many-objective optimization algorithm using reference-point-based nondominated sorting approach

    Part I: Solving Problems With Box Constraints.

    (2014)
  • D. Goldberg et al.

    Using collaborative filtering to weave an information tapestry

    Communications of the ACM

    (1992)
  • Cited by (98)

    • Design of novel deformable mirror with large displacement

      2024, Concurrency and Computation: Practice and Experience
    View all citing articles on Scopus
    View full text