A big-data oriented recommendation method based on multi-objective optimization
Introduction
As the volume of data on e-commerce platforms continues to grow at an exponential rate, it provides an unprecedented development for precision marketing and dynamic supply chain optimization. However, the big data also triggers increasingly serious problems such as resource-overload and information-mislead. Especially for mobile users, the accuracy and velocity of acquiring useful contents from massive information has become a prior issue. To a certain extent, the creation of personalized recommendation technology has solved the dilemma between information diversity and user requirements specialization. It is meaningful that recommender systems can help users to get useful and valuable objects from massive data. Most of e-commerce platforms, such as Amazon, Taobao, Jd., have applied various kinds of recommender systems to improve the service quality. In real applications, traditional recommender systems usually adopt collaborative filtering algorithms based on similarity between users or objects to generate results. This strategy guarantees the accuracy of recommendation results while causes diversity loss.
In this paper, we design a novel recommendation model which consists of an improved collaborative filtering algorithm and a multi-objective evolutionary algorithm under a big data processing framework. Firstly, we construct a similarity computational model according to resource diffusion principle which considers more factors including location information, popular degree, preference and social relationship. Then, we adopt a multi-objective evolutionary algorithm to handle the multiple conflicts between accuracy and diversity. Finally, the method produces a set of solutions called Pareto optimal solutions, and each solution make a trade-off between two conflicting objectives, meanwhile, each result is not superior to others. We generate these Pareto solutions as optional recommendation lists for target users.
The target of our work is to provide a recommendation method which can bring higher accuracy and more diversity. The advantage of our approach is that it models accuracy–diversitydilemma as a multi-objective optimization problem, then considers more factors and overcomes the shortcoming of parameter adjustment with traditional methods. Detailed numerical analysis on two benchmark data sets, MovieLens and Book-Crossing, and a real-world data set indicates that our algorithm outperforms other algorithms.
The main contributions of this paper are summarized as follows:
- (1)
Personalized recommendation based on an improved collaborative filtering model is refined with a novel similarity computational process which allows for many factors.
- (2)
By translating the procedure of generating personalized recommendation results into a multi-objective optimization problem, the requirement of the accuracy and diversity of recommendation is fulfilled.
- (3)
Similarity computational efficiency is substantiallyenhanced by adopting the massive-data oriented MapReduce framework which is widely used in computation-intensive field.
The remainder of this paper is organized as follows: Section 2 describes the related works; In Section 3, an improved CF method based on multi-objective optimization (ICF-MOA) is proposed; Section 4 provides experimental results of ICF-MOA on three data sets; Finally, conclusions are presented in Section 5.
Section snippets
Literature review
Recent years people have witnessed an increase in the use of information technology of recommender systems. Various kinds of successful algorithms have been proposed, among which are collaborative filtering (CF) [1], [2], [3], [4], [5], [6], content-based filtering [7], [8], [9], [10], [11], K-Nearest Neighbor (K-NN) [12], [13], [14], [15], [16], diffusion approach [17], [18], [19], [20], [21], and hybrid recommendation approaches [22], [23], [24], [25], [26], [27].
Firstly, we will introduce
Improved CF method based on multi-objective optimization
This paper is motivated to develop a big-data oriented recommendation method based on multi-objective optimization for e-commerce recommender systems. We show the framework of ICF-MOA, which is shorted for Improved CF Method Based on Multi-objective Optimization, as shown in Fig. 2. All solutions of Pareto front are considered as different recommendation lists for the target users.
The input is a user–object matrix and the outputs are different recommendation lists. At first, an improved CF
Experimental results and analysis
The program is written in Python running on Ubuntu 14.04. The tests are performed on three Servers which are powered by Xeon(R) X3323 2.5 GHz with 8 GB Memory and 1TB Hard disk. Performance profiling is done by running our method against two benchmark data sets and one real-world data set. A brief introduction of these data sets is as follows:
- (1)
MovieLens [39] is an international recognized commonly used and effective benchmark data set for testing the performance of a recommendation method and it
Conclusions
Recommender system has become an extremely common technique in recent years, and is adopted in a variety of applications. It collects information on the preferences of its users for a set of items and predicts possible future preferences or interests of users. An excellent recommendation algorithm meets high accuracy and certain diversity.
In order to solve the dilemma between accuracy and diversity and overcome the shortcoming of parameter adjustment with traditional methods, we proposed a
Acknowledgment
This research was supported by National Natural ScienceFoundation of China (71702164).
References (47)
- et al.
A survey of collaborative filtering based social recommender systems
Comput. Commun.
(2014) - et al.
A collaborative filtering approach for recommending OLAP sessions
Decis. Support Syst.
(2015) - et al.
Comparing user experiences on the search-based and content-based recommendation ranking on stroke clinical guidelines-a case study
Procedia Comput. Sci.
(2015) - et al.
On content-based recommendation and user privacy in social-tagging systems
Comput. Stand. Interfaces
(2015) - et al.
ML-kNN: a lazy learning approach to multi-label learning
Pattern Recognit.
(2007) - et al.
Fsknn: multi-label text categorization based on fuzzy similarity and k nearest neighbors
Expert Syst. Appl.
(2012) - et al.
A similarity metric designed to speed up, using hardware, the recommender systems k-nearest neighbors algorithm
Knowl.-Based Syst.
(2013) - et al.
Reversed CF: A fast collaborative filtering algorithm using a k-nearest neighbor graph
Expert Syst. Appl.
(2015) - et al.
Personal recommendation via unequal resource allocation on bipartite networks
Physica A
(2010) - et al.
Combining content-based and collaborative recommendations: A hybrid approach based on bayesian networks
Internat. J. Approx. Reason.
(2010)
Hybrid recommendation approaches for multi-criteria collaborative filtering
Expert Syst. Appl.
Personalized hybrid recommendation for group of users: Top-N multimedia recommender
Inf. Process. Manage.
Boosting the K-nearest-neighborhood based incremental collaborative filtering
Knowl. Based Syst.
A pattern mining approach to enhance the accuracy of collaborative filtering in sparse data domains
Physica A
Personal recommendation via modified collaborative filtering
Physica A
Enhancing collaborative filtering recommendations by utilizing multi-objective particle swarm optimization embedded association rule mining
Swarm Evol. Comput.
A hybrid online-product recommendation system: combining implicit rating-based collaborative filtering and sequential pattern analysis
Electron. Commerce Res. Appl.
Multiobjective e-commerce recommendations based on hypergraph ranking
Inform. Sci.
NNIA-RS: A multi-objective optimization based recommender system
Physica A
A real-coded genetic algorithm with a direction-based crossover operator
Inform. Sci.
Recommender systems
Phys. Rep.
A novel recommendation method based on social network using matrix factorization technique
Inform. Process. Manag.
Graph-based collaborative ranking
Expert Syst. Appl.
Cited by (39)
Multi-objective optimization with recommender systems: A systematic review
2023, Information SystemsA multi-objective artificial bee colony approach for profit-aware recommender systems
2023, Information SciencesThe use of multi-criteria decision-making methods in business analytics: A comprehensive literature review
2022, Technological Forecasting and Social ChangeA two-stage personalized recommendation based on multi-objective teaching–learning-based optimization with decomposition
2021, NeurocomputingCitation Excerpt :Geng et al presented a MOIA for personalized location recommendation so as to optimize the matching qualities of similarity and geographic properties [62]. Xu developed a new improved CF-based multi-objective PRS for big-data oriented recommendation based on MapReduce framework [63]. As a population-based optimization algorithm, teaching–learning-based optimization (TLBO) has characters of simple concept and without the specific parameters, and it has been successfully applied many areas [64].
A novel POI recommendation method based on trust relationship and spatial–temporal factors
2021, Electronic Commerce Research and ApplicationsBias Reduction News Recommendation System
2024, Digital