How to select and weight context dimensions conditions for context-aware recommendation?

https://doi.org/10.1016/j.eswa.2021.115176Get rights and content

Highlights

  • •

    Exploits pertinent contextual information to improve the recommender system.

  • •

    Introduces a recommender prediction model based on context selection and weighting.

  • •

    Two rich context-aware datasets are built to upgrade the performance of our model.

  • •

    Experimental results show that our proposal outperforms the existing approaches.

Abstract

Contextual information plays a key role in Context-Aware Recommender Systems (CARS). The rating prediction in CARS focuses on improving recommendation accuracy attempting to form a personalized information recommendation for users. Three key problems that affect the performances of recommender systems: (i) context condition’s selection; (ii) context condition’s weighting; and (iii) users’ context conditions matching. Context-aware approaches have the assumption that all context conditions have the same weight. These approaches ignore that users have different preferences in different contexts. To address these three problems, we introduce a novel approach for Selecting, Weighting Context Conditions (SWCC) and measuring semantic similarity between users’ situations. Evaluation experiments show that the proposed approach is outperforming the pioneering context-aware recommendation approaches of the literature.

Section snippets

Introduction and motivation

Recommender Systems (RS) are specific types of information filtering to present items (movies, music, books, news, images, etc.) that likely to interest the user (Díaz et al., 2012, Chen et al., 2018, Usha Yadav and Bhatia, 2020). Typically, an RS compares the current situation of a user to some baseline characteristics (his/her history versus those of other users), and tries to predict the rating that a user would provide.

In this respect, context-awareness can solve many problems that affect

Formal definitions

We start this section by defining and modeling some notions like context, user context, etc.

  • •

    Context dimension: In order to use a context itself, the definition of its associated dimensions must be understood beforehand (Crane, Benachour, & Coulton, 2011). The dimension term is a measurable extent or quantity (such as amplitude, brightness, hue, frequency, length, width, height, time, mass, volume, weight) that denotes the degree to, or range over, which something extends. In our case, a

Related work

We basically built traditional RSs upon the knowledge of user preferences for a particular set of items, as well as the input data for traditional RSs. This input is typically based on the records of the form Users×Items→Ratings. In contrast, CARS are built based on the knowledge of context user preferences and typically deal with data records of the form Users×Items×Contexts→Ratings.

An RS relies on a content-based approach or Collaborative Filtering (CF) approach or even both. The operating

SWCC: A new prediction model based on the selection and the weighting of context conditions

From the related work, we notice that context condition selection and weighting are important for improving prediction in CARS. Nevertheless, they are used separately. In this paper, we introduce a prediction model (c.f, Fig. 1) that starts from a current situation and a user history to predict the rating of the current item. The first step in our approach comprises selecting the context conditions. We propose four selection methods (cf. sub-section 4.1.1). We base the first one on the variance

Experimental study

In this section, we usher by describing used datasets, and we specify how these datasets were split in terms of training and test. Then, we present and discuss the obtained results.

Conclusions and future work

Many studies have investigated contextual conditions weighting and contextual conditions selection independently, although these two topics seem closely related. In this paper, we introduce an algorithm that combines contextual conditions selection and weighting in context-aware recommendations. In fact, we address the problem of how to select contextual conditions in order to eliminate the irrelevant ones, how to associate importance degrees to a contextual conditions context and how to

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.

Acknowledgement

Sadok Ben Yahia is supported by the Astra funding program Grant 2014-2020.4.01.16-032.

References (96)

  • M. Akhmatnurov et al.

    Context-aware recommender system based on boolean matrix factorisation

  • B. Alhijawi et al.

    A collaborative filtering recommender system using genetic algorithm

    Information Processing & Management

    (2020)
  • Amit, L., Moshe, U., Bracha, S., & Lior, R. (2019). Deep context-aware recommender system utilizing sequential latent...
  • Baeza-Yates, R., & Ribeiro-Neto, B. (1999). Modern information...
  • Baltrunas, L., Ludwig, B., & Ricci, F. (2011). Matrix factorization techniques for context aware recommendation. In...
  • L. Baltrunas et al.

    Incarmusic: Context-aware music recommendations in a car

  • L. Baltrunas et al.

    Context relevance assessment and exploitation in mobile recommender systems

    Personal and Ubiquitous Computing

    (2012)
  • L. Baltrunas et al.

    Item weighting techniques for collaborative filtering

  • E. Blanchard et al.

    A generic framework for comparing semantic similarities on a subsumption hierarchy

  • Blum, A., & Langley, P. (1997). Selection of relevant features and examples in machine learning 97,...
  • D. Bouneffouf et al.

    Following the user’s interests in mobile context-aware recommender systems:the hybrid-∊-greedy algorithm

  • M. Braunhofer et al.

    Selective contextual information acquisition in travel recommender systems

    Information Technology & Tourism

    (2017)
  • J. Breese et al.

    Empirical analysis of predictive algorithms for collaborative filtering

  • H. Chaker et al.

    Business context information manager: An approach to improve information systems

    (2011)
  • W. Chang et al.

    Personalized e-tourism attraction recommendation based on context

  • Chen, X., Xu, K., & Pi, R. (2003). A popularity-based prediction model for web prefetching 36,...
  • Rui Chen et al.

    A survey of collaborative filtering-based recommender systems: From traditional methods to hybrid methods based on social networks

    IEEE Access

    (2018)
  • B. Chen et al.

    Conrec: A software framework for context-aware recommendation based on dynamic and personalized context

  • Chollet, F. (2015). Keras....
  • Cover, T., & Thomas, J. (1991). Elements of information...
  • L. Crane et al.

    A user study of the spatial and temporal dimensions of context to support virtual learning environments

  • F. de Rosaa et al.

    Analytical games for knowledge engineering of expert systems in support to situational awareness: The reliability game case study

    Expert Systems With Applications

    (2019)
  • B. Diallo et al.

    Mobile and context-aware geobi applications: A multilevel model for structuring and sharing of contextual information

    Journal of Geographic Information System

    (2012)
  • Díaz, A., Regina, M., Edelweis, R., & Libertad, T. (2012). Educational recommender systems and technologies: Practices...
  • A. Ferraro et al.

    Using offline metrics and user behavior analysis to combine multiple systems for music recommendation

  • N. Friedman et al.

    Bayesian network classifiers

    Machine Learning

    (1997)
  • R. Gupta et al.

    Contextual information based recommender system using singular value decomposition

  • F.M. Harper et al.

    The movielens datasets: History and context

    (2016)
  • J. Herlocker et al.

    An algorithmic framework for performing collaborative filtering

  • B. Hidasi et al.

    General factorization framework for context-aware recommendations

    CoRR

    (2014)
  • T. Huang

    A popularity-based neighbor selection model in p2p file-sharing system

  • Z. Huang et al.

    Context-aware recommendation using rough set model and collaborative filtering

    Artificial Intelligence Review

    (2011)
  • J. Jiang et al.

    Semantic similarity based on corpus statistics and lexical taxonomy

  • R. Jin et al.

    An automatic weighting scheme for collaborative filtering

  • C. Ji et al.

    Particle filter with swarm move for optimization

  • P. Jokinen et al.

    Two algorithms for approximate string matching in static texts

  • B. Joshi et al.

    Asynchronous distributed matrix factorization with similar user and item based regularization

  • A. Karatzoglou et al.

    Multiverse recommendation: n-dimensional tensor factorization for context-aware collaborative filtering

    • Cited by (5)

    View full text
    © 2021 Elsevier Ltd. All rights reserved.