Elsevier

Applied Soft Computing

Volume 40, March 2016, Pages 187-198
Applied Soft Computing

A fuzzy model for managing natural noise in recommender systems

https://doi.org/10.1016/j.asoc.2015.10.060Get rights and content

Highlights

  • Focus on natural noise management in collaborative filtering recommender systems.

  • The natural noise management is improved through fuzzy tools.

  • The proposal detects the noisy ratings by analysing the user's and item's tendencies.

  • Noisy ratings are corrected by predicting their value with a noise free dataset.

  • The proposal is evaluated on three widely-used recommendation datasets on movies domain and it shows improvements compared with other techniques.

Abstract

E-commerce customers demand quick and easy access to products in large search spaces according to their needs and preferences. To support and facilitate this process, recommender systems (RS) based on user preferences have recently played a key role. However the elicitation of customers preferences is not always precise either correct, because of external factors such as human errors, uncertainty and vagueness proper of human beings and so on. Such a problem in RS is known as natural noise and can bias customers recommendations. Despite different proposals have been presented to deal with natural noise in RS none of them is able to manage properly the inherent uncertainty and vagueness of customers preferences. Hence, this paper is devoted to a new fuzzy method for managing in a flexible and adaptable way such uncertainty of natural noise in order to improve recommendation accuracy. Eventually a case study is performed to show the improvements produced by this fuzzy method regarding previous proposals.

Introduction

The development of the e-commerce has made available huge quantities of information and items that customers cannot filter effectively. Therefore, the facilitation of finding out quick and easily items that fits their preferences and needs is an important challenge nowadays. Recommender systems (RS) are probably the most successful tool to support personalised recommendations [1], [2]. Currently, they are widely used in different scenarios like e-commerce [3], e-learning [4], [5], e-government [6], tourism [7], [8], web pages [9], and digital games [10].

Different approaches have been used in RS, being the content-based (CBRS) [11] and the collaborative filtering (CFRS) [1] the most widespread. CBRS methods are based on items’ descriptions to generate the users’ recommendations, meanwhile CFRS have performed this task just using users ratings about items.

CFRS are currently the most popular type of RS in real world because they perform very well even when items descriptions are not available. In spite of this, the necessity of customers preferences has produced some problems that limit their performance, such as cold start and sparsity [1], [12], and more recently new related problems regarding the quality of the rating data [13], [14], [15]. Specifically, Ekstrand et al. [16] pointed out that the rating elicitation process is not error-free, hence the ratings can contain noise. They mentioned that such a noise, previously coined natural noise in [17], could be caused by human error, mixing of factors in the rating process, uncertainty and other factors. They stated that its detection and correction should provide more accurate recommendations.

Thereby several approaches have been introduced for managing these rating inconsistencies in recommendation scenarios depending on the information available, such as user dependent approaches [14], item-attributes dependent approaches [18], and also approaches that manage natural noise only using the rating values [15]. They all perform generally better that base algorithms, but they present an important limitation to deal with inherent uncertainty and vagueness of preferences because they solely represent and manage them by means of crisp values that may imply lack of robustness.

Therefore, this paper aims at developing a more flexible approach for dealing with natural noise in RS, that properly models such vagueness and uncertainty by means of fuzzy tools such as fuzzy sets [19], fuzzy linguistic approach [20], [21] and computing with words [22], [23] that have provided successful results modelling that type of uncertainty in other problems [24], [25], [26]. Hence, the main objective of this paper is to improve accuracy of RS by managing natural noise with a novel fuzzy-based method that manages the uncertainty present in natural noise by using fuzzy tools [19] that provide a greater flexibility in the characterisation process of elements in the recommendation system, leading to improvements in the recommendation accuracy.

The remainder of this paper is structured as follows. First, Section 2 provides the required background for the current research. Section 3 focuses on the novel fuzzy approach for dealing with natural noise in CFRS. Section 4 then presents an experimental procedure to evaluate the performance of the approach in relation to the previous works. Finally, Section 5 concludes the article and remarks further work.

Section snippets

Preliminaries

This section provides the required background for the current research, including basics about CFRS, a review of natural noise processing in CFRS, and a description of the fuzzy logic tools used in the proposal.

A fuzzy approach to detect noisy ratings

Natural noise in RS may bias recommendations, here it is introduced a new fuzzy approach to deal with it that unlike previous rigid proposals, this new approach uses fuzzy tools for managing in a more flexible way the inherent uncertainty of customers preferences/needs. The general scheme of this fuzzy approach for managing natural noise is depicted in Fig. 1 (amplified in Fig. 2, Fig. 5, Fig. 6) and consists of the following steps:

  • (a)

    Fuzzy profiling: To manage the uncertainty of ratings, they are

Case study

This section describes the experiments developed to evaluate the effect of our proposal on the improvement of the recommendation accuracy. Hence our proposal is specified and then compared with other methods that deal with noise in the ratings. First, the datasets (MovieLens, MovieTweeting and Netflix Tiny) in which experiments have been performed are detailed in Section 4.1. Then a specification of the fuzzy method applied in the case study is provided in Section 4.2. After that, the

Concluding remarks

Natural noise is denominated to the noise unintentionally introduced by human beings when they are eliciting preferences. Its management in recommender systems has attracted the attention of many researchers and several proposals have been done. However, such proposals deal with natural noise in a rigid way such an small variation can make change the classification of ratings as noisy or not noisy.

Therefore, to avoid such a strict managing process, this paper has introduced a more flexible

Acknowledgements

This research work was partially supported by the Research Project TIN-2012-31263, the Eureka SD Project (agreement number 2013-2591), which is supported by the Erasmus Mundus Programme of the European Union, and also the Spanish Ministry of Education, Culture and Sport FPU fellowship (FPU13/01151).

References (61)

  • L. Martínez et al.

    An overview on the 2-tuple linguistic model for computing with words in decision making: extensions, applications and challenges

    Inf. Sci.

    (2012)
  • L. Martínez et al.

    Linguistic decision making: tools and applications

    Inf. Sci.

    (2009)
  • Y. Caises et al.

    Combining instance selection methods based on data characterization: an approach to increase their effectiveness

    Inf. Sci.

    (2011)
  • R. Nicolas et al.

    Multi-label classification based on analog reasoning

    Expert Syst. Appl.

    (2013)
  • Y.B. Ruiz-Blanco et al.

    A Hooke's law-based approach to protein folding rate

    J. Theor. Biol.

    (2015)
  • F. Chiclana et al.

    A statistical comparative study of different similarity measures of consensus in group decision making

    Inf. Sci.

    (2013)
  • G. Adomavicius et al.

    Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions

    IEEE Trans. Knowl. Data Eng.

    (2005)
  • J.A. Konstan et al.

    Recommender systems: from algorithms to user experience

    User Model. User-Adapt. Interact.

    (2012)
  • B. Xiao et al.

    E-commerce product recommendation agents: use, characteristics, and impact

    Manag. Inf. Syst. Q.

    (2007)
  • R.Y. Toledo et al.

    An e-learning collaborative filtering approach to suggest problems to solve in programming online judges

    Int. J. Distance Educ. Technol.

    (2014)
  • X. Guo et al.

    Intelligent e-government services with personalized recommendation techniques

    Int. J. Intell. Syst.

    (2007)
  • M. Pazzani et al.

    Content-based recommendation systems

    Adapt. Web

    (2007)
  • R. Rodríguez et al.

    Using linguistic incomplete preference relations to cold start recommendations

    Internet Res.

    (2010)
  • B. Li et al.

    Noisy but non-malicious user detection in social recommender systems

    World Wide Web

    (2013)
  • H.X. Pham et al.

    Preference-based user rating correction process for interactive recommendation systems

    Multimed. Tools Appl.

    (2013)
  • R. Yera et al.

    Correcting noisy ratings in collaborative recommender systems

    Knowl.-Based Syst.

    (2015)
  • M.D. Ekstrand et al.

    Collaborative filtering recommender systems

    Found. Trends Hum.-Comput. Interact.

    (2011)
  • M.P. O’Mahony et al.

    Detecting noise in recommender system databases

  • X. Amatriain et al.

    Data mining methods for recommender systems

    Recommender Systems Handbook

    (2011)
  • R.R. Yager et al.
    (2012)
  • Cited by (55)

    • Sampling and noise filtering methods for recommender systems: A literature review

      2023, Engineering Applications of Artificial Intelligence
    • A Mixture-of-Gaussians model for estimating the magic barrier of the recommender system[Formula presented]

      2022, Applied Soft Computing
      Citation Excerpt :

      Human inconsistency is often caused by the changes in human emotions and living environments [7]. Human error causes the malicious noise [8], while the latter two arise the natural noise [9]. We will discuss the natural noise throughout the paper.

    • An effective and efficient fuzzy approach for managing natural noise in recommender systems

      2021, Information Sciences
      Citation Excerpt :

      In our scheme, we devise a membership function to classify the ratings based on unfixed boundary points. To verify the effectiveness of unfixed boundary points in our membership function, we compare our membership function with the membership function in [16], which is based on fixed boundary points. The same schemes (i.e., the detecting scheme in Section 3.3 and the correction schemes in Section 3.4) are used here.

    View all citing articles on Scopus
    View full text