A Comprehensive Survey on Comparisons across Contextual Pre-Filtering, Contextual Post-Filtering and Contextual Modelling Approaches

TELKOMNIKA, Vol.15, No.4, December 2017, pp. 1865~1874
ISSN: 1693-6930, accredited A by DIKTI, Decree No: 58/DIKTI/Kep/2013
DOI: 10.12928/TELKOMNIKA.v15i4.6875  1865
Received August 8, 2017; Revised October 28, 2017; Accepted November 21, 2017
A Comprehensive Survey on Comparisons across
Contextual Pre-Filtering, Contextual Post-Filtering and
Contextual Modelling Approaches
Khalid Haruna
1
, Maizatul Akmar Ismail
*2
, Damiasih Damiasih
3
, Haruna Chiroma
4
and Tutut
Herawan
5
1,2
Department of Information Systems, Faculty of Computer Science and Information Technology,
University of Malaya, Malaysia
1
Department of Computer Science, Faculty of Computer Science and Information Technology, Bayero
University, Kano, Nigeria
5
Sekolah Tinggi Pariwisata Ambarrukmo, Yogyakarta, Indonesia
4
Federal College of Education (Technical), School of Science, Department of Computer Science Gombe,
Nigeria
5
Universitas Negeri Yogyakarta, Indonesia
6
Politeknik Negeri Malang, Indonesia
*Corresponding author, e-mail: Kharuna.cs@buk.edu.ng
1
, maizatul@um.edu.my
2
Abstract
Recently, there has been growing interest in recommender systems (RS) and particularly in
context-aware RS. Methods for generating context-aware recommendations are classified into pre-filtering,
post-filtering and contextual modelling approaches. In this paper, we present the several novel approaches
of the different variant of each of these three contextualization paradigms and present a complete survey
on the state-of-the-art comparisons across them. We then identify the significant challenges that require
being addressed by the current RS researchers, which will help academicians and practitioners in
comparing these three approaches to select the best alternative according to their strategies.
Keywords: Contextual Pre-filtering, Contextual Post-filtering, Contextual Modelling, Comparisons,
Contextualization Paradigms Context-Aware Recommender System
Copyright © 2017 Universitas Ahmad Dahlan. All rights reserved.
1. Introduction
With the help of the internet, users can now deliver, access and retrieve whatever web
resources they wish at any time and in anywhere according to their interest. However, the
excessive availability of web resources leads to the problem of information overload [2, 3], in
which users can easily be lost over the Cyber Ocean of information [4, 5]. Recommender
systems (RS) that personalise suggestions of various items and services to users emerged in
the mid of 1990s to remediate the problem of information overload [6-9]. At its emergence,
traditional 2D RS were predominantly used to predict users’ preferences. These approaches
utilised the items and users as the set of entities to predict the ratings that are either implicitly
deduced by the system [10] or are expressly provided by the users [11, 12].
In the early 2000s, researchers extend the research in recommender systems to
leverage contexts in the recommendation process [13, 14]. Context is an all-around concept
[15], that has been studied across numerous disciplines, such as linguistics, philosophy, and
cognitive and organisational science. In the late 1980s, computer science as a discipline adopts
the concept of context primarily in ubiquitous computing and AI [16].
Context has been researched across numerous disciplines [17], in which every
discipline tends to proffer its idiosyncratic view [18]. A hundred and fifty different views of
context from different fields of research have been presented and examined in [19]. The authors
concluded that it would be very much difficult to find a single definition of context that is
unanimously satisfying all research disciplines. However, the most reported view of the context
in the field of computer science is that of [20], which viewed context as:

 ISSN: 1693-6930
TELKOMNIKA Vol. 15, No. 4, December 2017 : 1865 - 1874
1866
“Any information that can be used to characterise the situation of an entity. An entity is a
person, place, or object that is considered relevant to the interaction between a user and an
application, including the user and application themselves”.
Context has additionally been classified in several ways. However, the foremost
reported categorisation within the field of computing is that of [21] that considers context to be
representational and interactional views. The interactional view assumes context to be some
specific factors that induced user behaviour and which may not necessarily be observable, while
representational view assumes context to be some specific factors that are priori identified, and
which are defined by some contextual factors with known hierarchical structures and which do
not change considerably over time. In the early work of [20, 22], the context was classified into
computing, user, and physical.
Most of the previous researchers focused on the representational view of context [23].
Authors in [15, 24] have further extended the view of representational context to include
contextual pre-filtering, contextual post-filtering and contextual modelling approaches as shown
in Figure 1. Then, they challenged researchers within the area of recommendation systems to
examine those strategies further and to compare them in numerous experimental settings and
with considering various recommendation factors to determine which approach is superior to the
others and under what situation. The comparison is essential not only to the researchers but
also to the practitioners [1].
In this paper, we determined to provide a complete survey of the state-of-the-art
comparisons between these three approaches. This is followed after the several novel
approaches of each of the pre-filtering, post-filtering and contextual modelling approaches are
presented. We then raised the significant challenges that require being addressed by the
present RS researchers. This paper is supposed to assist researchers in developing a deeper
understanding of those contextualisation paradigms, and their trade-offs and practitioners will
use it to decide on the most effective possibility of keeping with their market strategy. In
summary, the contributions of this paper are:
a. The paper presents the concept of context and also the three-contextualization paradigms
for incorporating contextual data in the recommendation process.
b. The paper provides a comprehensive summary of the many novel approaches of pre-
filtering, post-filtering, and contextual modelling approaches.
c. The paper presents state-of-the-art comparisons between pre-filtering, post-filtering and
contextual modelling approaches.
d. The paper also presents outstanding challenges that are not entirely addressed within the
literature and suggests available opportunities for future research directions.
The remaining sections of the paper are organised as follows. Section 2 presents the
adopted methodology for extracting the literature. Section 3 presents the three-contextualisation
paradigms, and the several novels approach for incorporating contexts in the recommendation
process. The comparisons between pre-filtering, post-filtering, and contextual modelling
approaches are presented in section 4. We then present outstanding challenges and suggest
possible opportunities for future research directions in section 5, and finally conclude the paper
in section 6.
2. Methodology
As defined by [25, 26], a review is a rigorous way of identifying, appraising, selecting,
synthesising, evaluating and interpreting all available researches relevant to particular research
questions of particular interest. It aims at critically making an appraisal on the previous
contributions based on specific research questions to draw attention to research gaps that need
to be addressed. In this paper, we target to provide a complete survey of the state-of-the-art
comparisons between pre-filtering, post-filtering, and contextual modelling approaches.
To perform a comprehensive search, we identified the bibliographic databases that
cover the majority of journals and conference proceedings papers published in the field of
computer science. These databases are ScienceDirect, ACM Digital Library, SpringerLink, Web
of Science, IEEE Xplore, Scopus, Dblp computer science bibliography, and Google Scholar
Portal.
The searching process was performed based on the Boolean search criteria. We
extracted all of the papers that mentioned any of the following words/phrase in the paper title;

TELKOMNIKA ISSN: 1693-6930 
A Comprehensive Survey on Comparisons across Contextual Pre-Filtering… (Khalid Haruna)
1867
Recommender system, context-aware, pre-filtering, post-filtering, contextual modelling. We then
reviewed all of the selected papers and included all papers that either compare any of the
contextualization paradigms or proposed a new variant of the three approaches.
In order to retrieve the highest number of publications, we followed the list of references
of each of the chosen papers and compared with our databases to search out any missing
paper that might satisfy our inclusion criteria. We also searched each of the selected papers in
scholar.google.com and had a brief overview over the title of all the papers in ‘cited by’, ‘related
articles’ and ‘all versions’.
Figure 1. Contextualization Paradigms (Figure from [12])
3. Contextualization Paradigms in Recommender Systems
As shown in Figure 1, three different algorithmic paradigms exist for incorporating
contexts in the recommendation process. The approaches are based on when the contextual
information is considered to be significant in conjunction with the classical two-dimensional
approaches. These approaches are pre-filtering, post-filtering, and contextual modelling
approaches. Since the challenge raised by [27], a substantial number of researchers have
proposed and compared novel ways of incorporating contexts in the recommender systems. In
this section, we described each of the paradigms and presented its corresponding novel
approaches from the literature.
3.1. Contextual Pre-Filtering
A pre-filtering approach depicted in Figure 1a is an approach that applies a context-
dependent criterion, which selects only items that are appropriate to a specific context. In
contextual pre-filtering, only the filtered items are considered for recommendations. One
advantage of this approach is its ability to employ any of the many classical recommendation
techniques. That is to say, all the approaches on traditional recommender system can be
applied to the reduced data. In what follows, we briefly described each of the several novel
contextual pre-filtering approaches proposed from the literature.
3.1.1. Item Splitting

 ISSN: 1693-6930
1868
Item splitting is a pre-filtering technique that tries to find a contextual condition on which
to split items. The approach was introduced in [28-30] and compared with a reduction-based
approach presented in [31]. In this approach, items are split based on the perceived context and
items with similar context are combined and processed together. This approach assumes that
the analysis of some certain items might result in a distinct outcome in several contextual
settings. A more comprehensive evaluation of these approaches has been presented in [32].
3.1.2. Micro Profiling
Micro profiling is a bit more similar to item splitting which was proposed by [33]. While
item splitting splits items, micro profiling split users instead. In this technique, user profiles are
split into numerous sub-profiles known as micro-profiles, each representing users in a selected
context. These micro-profiles are then used to make recommendations.
3.1.3. Distributional-Semantics Pre-Filtering (DSPF)
Distributional-semantic pre-filtering (DSPF) is an approach that aims at handling the
well-known data-sparsity problem of CARS. The novel method was presented in [11] to take
advantage of the distributional semantics of contextual conditions to build a context-aware rating
prediction model. DSPF is a reduction-based method that employs a situation-to-situation
similarity feature to select the right level of contextualisation for given information. Given a target
contextual situation, the ratings tagged with the contextual situations are used to construct a
predictive model primarily based on matrix factorisation. The model is then used to compute the
rating predictions and identify recommendations that are specific to the target contextual
situation.
3.1.4. Exact and Generalized Pre-Filtering
Exact and generalised pre-filtering are two forms of contextual pre-filtering proposed by
[34]. Exact pre-filtering (EPF) selects all the information mentioned the correctly given context
within the recommendation process, whereas generalised pre-filtering selects all the information
that mentioned a particular context which supports the generalisation of the contextual
information [24].
3.2. Contextual Post-Filtering
A contextual post-filtering approach depicted in Figure 1b is similar to contextual pre-
filtering approach only that it applies the filtering process after recommendations have been
computed. In other words, contextual post-filtering approach disregards the contextual
information in the input data when generating the list of the top-N recommendations. The list of
the top-N will then be refined by either filtering out recommendations with a smaller probability
of relevance or by adjusting the ranking of recommendations by weighting the predicted rating
score with the probability of relevance. Similar to contextual pre-filtering, contextual post-filtering
also allows using any of the numerous traditional recommendation techniques. In what follows,
we briefly described each of the several novel contextual post-filtering approaches proposed
from the literature.
3.2.1. Weight PoF and Filter PoF
Weight PoF and Filter PoF are two kinds of the post-filtering technique proposed by
[34]. In Weight PoF, the suggested items from 2D recommendations are re-ordered and
weighted based on the rating probability of relevance in the given context. While, in Filter PoF,
the suggested items with little probability relevance from the 2D recommendations process are
filtered out from the recommendation list.
The two approaches differ in how the recommendations are contextualised; Filter PoF
filters the traditional 2D ratings out based on a threshold value P.
),( jiRatingk {
𝑅𝑎𝑡𝑖𝑛𝑔(𝑖, 𝑗) 𝑖𝑓 ( PjiPk
),( )
0 𝑖𝑓 ( k
P < 𝑃)
(1)
Whereas Weight PoF multiplies each of the traditional 2D ratings by ),( jiPk

1869
),(*),(),( jiPjiRatingjiRating kk  (2)
Where P is the threshold value.
3.2.2. Context-RS
Context Recommender System (Context-RS) is a generic contextual post-filtering
technique proposed by [35] by mining association rules between contextual knowledge and item
characteristics to find their useful correlations. Context-RS combines traditional 2D
recommender systems, contextual knowledge and association rules to enhance the quality of
top-N recommendations.
3.3. Contextual Modelling
Contextual modelling approach depicted in Figure 1c takes contextual considerations
into the recommendation algorithm itself. That is to say, the contextual information that is
considered relevant is employed directly within the classical RS modelling method as part of the
recommendation process. In other words, the contextual information is employed directly within
the recommendation function as an explicit predictor of a user’s rating score for an item. In what
follows, we briefly described each of the several novel contextual modelling approaches
proposed from the literature.
3.3.1. Graph-Based Relevance Measure (CGR)
Graph-based relevance measure (CGR) is a contextual modelling approach proposed
by [36]. The approach aimed to model and incorporate contexts within the recommendation
method by assessing the potential relevance of a target user with the set of items for a better
recommendation.
3.3.2. Contextual-neighbors
Contextual-neighbours are new types of contextual modelling approach proposed by
[37], based on the notion of utilising contextual information to calculate the degree of the
neighbourhood using a user-based collaborative filtering approach. The authors introduced four
variant, in which each of the variants selects contextual neighbourhood differently (see [37]).
We summarised each of these approaches in Table 1.
Table 1. Summary of Novel Approaches for Incorporating Contexts
in the Recommendation Process
Research
reference
Approach(es) proposed
Pre-filtering Post-filtering Contextual modelling
[23-25] Item-Splitting
[28] Micro-Profiling
[8] Distributional semantic
pre-filtering (DSPF)
[29] Exact Pre-Filtering (EPF),
Generalized Pre-Filtering
Weight PoF, Filter PoF
[30] Context-RS
[31] Context-Aware Profiling Graph-Based Contextual Modeling (CGR)
[33] Context-Aware SVM
[32] Contextual-neighbors CM (Mdl1, Mdl2, Mdl3, Mdl4)
4. Comparisons Across Contextual Pre-Filtering, Contextual Post-Filtering and
Contextual Modelling Approaches
A number of researchers faced the challenge raised by [24] in comparing and proposing
the several variants of pre-filtering, post-filtering, and contextual modelling approaches. To be
specific, let us start with the work of [39], which undertake broad empirical experiments to study
the significance of contextual information in predicting customer behaviour and how best it can
be used when building customer models. The results from the experiments reveal that context
does matter in modelling customers’ behaviour and that user’s contexts can be inferred from an
existing data with acceptable accuracy. The results also show that contexts play an essential

 ISSN: 1693-6930
1870
role in personalisation and companies can utilise the opportunity to enhance the predictive
performance of customer’s behaviour.
The authors in [40] target to identify the effect of contextual information on
recommendation performance. In doing so, they utilised the collaborative recommender system
to compare a pre-filtering method to a post-filtering approach. Based on their experimental
results, the post-filtering approach has significantly outperformed the pre-filtering approach in all
experimental setup. This is because pre-filtering approach suffers the problem of sparsity-
homogeneity trade-off. To address the issues of homogeneity and sparsity, [41] studied the
interaction between sparseness and homogeneity to understand how including context with a
pre-filtering approach improves the performance of a recommender system using
transactional data.
In [42], experimental analysis has also been carried out to investigate whether context-
aware recommender systems (CARS) always surpass the classical RS. The authors assessed
the performance of the classical RS with the three forms of CARS using two different
recommendation tasks. The results from the experiments reveal that the type of
recommendation task has a significant effect in the comparison between RS and CARS and that
the primary conditions that affect the comparisons are the overall number of recommendable
items, number of items in the recommendation list and the specific performance that has to be
measured.
Also, [43] carried some series of experiments by varying market granularity, dependent
variable and context granularity to understand whether including the contextual information in a
predictive model reduces the misclassification costs and the conditions that trigger it to happen.
The results reveal that context leads to a decrease in the misclassification cost, especially when
the unit of analysis is a micro-segment.
To improve the target market, a conceptual framework has been proposed in [44]. The
framework has the potential of incorporating contextual information, which can use by a
segmentation model to construct a predictive model capable of identifying customers’ behaviour
in a segment.
A novel variant of contextual modelling approaches has been proposed in [45]. The
approach called contextual neighbours is based on the notion of utilising contexts to calculate
the degree of the neighbourhood in a collaborative filtering approach. In addition to introducing
four variants of the approach, the authors also compared pre-filtering, post-filtering, the four
variant and un-contextual recommendations between a wide range of experimental settings.
The results show that contextual post-filtering approaches proved more significant than any
other approach, but the comparison is time-consuming and laborious. On the other hand,
contextual modelling approach proved to be second-best and therefore may be a good
alternative.
The research presented in [34] compared the performance of two forms of post-filtering
approaches weight PoF and filter PoF to be precise and exact pre-filtering. The comparison was
to determine the circumstances in which one approach is preferable to others. Evaluation using
predictive accuracy shows that the comparison depends mostly on the forms of post-filtering
used. The comparison is extended further in [37] to include the different variant of contextual
modelling approaches. The resulting comparison reveals that contextual post-filtering yield the
best-of-breed contextual method when realised in the best way but may provide the worst result
if utilised poorly.
An empirical comparison of movie recommendation domain has been presented in [46].
The comparisons across the three contextualization paradigms reveal that none of the
techniques was superior to others in all situations. However, random forest-based contextual
approach and item splitting of pre-filtering tend to provide excellent performance.
While [34, 37, 39-46] performed the comparisons based on accuracy alone, [47] put into
consideration also the diversity of recommendation to compare the several pre-filtering, post-
filtering, and contextual modelling approaches to determine which method is superior to others
and under which situation.
The limitations of [34, 37, 39-46] were that; first, their comparisons solely consider
predictive accuracy that within the literature proven to be not sufficient for a utile and
satisfactory recommendations [48]. Secondly, the comparisons were strictly marginal and failed
to determine the region where one approach outperforms the others and under what situations.
Also, the comparisons mainly consider a domain of application such as e-commerce in the case

1871
of [34, 37, 39-45], and movie domain in the case of [46]. Such findings cannot be generalised
across other domains of applications because factors that are significant in one domain may be
entirely irrelevant in another and different dataset may contain different contextual information.
A more detailed and comprehensive analysis of pre-filtering, post-filtering, and
contextual modelling methods were carried out in [23]. In addition to considering the predictive
accuracy, the authors also take into account the diversity of recommendations. Furthermore, in
addition to the marginal comparison, they also rigorously present the regional analyses. We
summarised these comparisons in Table 2.
Table 2. Comparisons of the Recent Researches Across the Three Contextual Paradigms
S/
N
Research
Reference
Comparison Type Of
Comparison
Factors Considered Evaluation
Measures
Dataset
1 [34] Contextual
and un-
contextual
Marginal 1) data sets
2) The degree of contextual
information
3) The granularity of customer
segments
4) Types of predictive models
5) Types of dependent
variables used
6) Types of performance
measures.
Accuracy and
area
under the
ROC curve
(AUC)
E-commerce
2 [35] Un-contextual,
Pre-filtering
and post-
filtering
Marginal 1) Type of data (Contextual
and non-contextual)
Accuracy E-commerce
3 [29] Pre-filtering
and post-
filtering
Marginal 1) Type of data Accuracy E-commerce
4 [36] Un-contextual
and Pre-
filtering
contextual
Marginal 1) Type of data Accuracy E-commerce
Pre-filtering
and post-
filtering
Marginal (1) Recommendation
tasks
(2) The number of
overall recommendable items
(3) The performance
metrics
(4) The number of items
in the recommendations list
(5) The granularity of
the context
Accuracy E-commerce
pre-filtering,
post-filtering
and contextual
modelling
Marginal (1) Type of data
(2) Context granularity
(3) Different level of
item aggregation
(4) Different
neighbourhood sizes
(5) Seven
recommendation engines
(6) Several
performance measures
Accuracy E-commerce
7 [42] Pre-filtering,
post-filtering
and contextual
modelling
Marginal
and
Regional
(1) Type of the
recommendation task
(3) Type of
recommendation data.
Accuracy and
Diversity
E-commerce
pre-filtering,
post-filtering
and contextual
modelling
Marginal (1) recommendation
strategies
(2) Type of data set
Accuracy E-commerce
Pre-filtering,
post-filtering
and contextual
Marginal (1) Types of data sets
(2) The granularity of
contextual knowledge
(3) Dependent variables
(4) Types of predictive
models
(5) Performance
Accuracy E-commerce

 ISSN: 1693-6930
1872
S/
N
Research
Reference
Comparison Type Of
Comparison
Factors Considered Evaluation
Measures
Dataset
measures
(6) Clustering
algorithms
post-filtering
and contextual
modelling
Marginal Type of data Accuracy Movie
11 [38] Contextual
and un-
contextual
Marginal (1) Market granularity
(2) Dependent variable
accuracy and
the
misclassificati
on costs
E-commerce
post-filtering
and contextual
modelling
Marginal
and
Regional
(1) Type of the
recommendation task
(3) Type of
recommendation data.
Accuracy and
diversity
E-commerce
5. Challenges and Future Recommendations
Researchers are now developing an interest in the area of the recommender system,
especially with the emergence of context-aware recommender systems in the early 2000s. The
approaches used in generating context-aware recommendations are classified into pre-filtering,
post-filtering and contextual modelling approach as depicted in Figure 1. (See section 1).
Considering contexts in the recommendation process proved to enhance the general
performance of recommender systems in various application domains. However, the
improvement largely depends on the several factors considered in the recommendation
process. Therefore, companies need to identify the essential factors that will maximise their
profit before adopting a method.
As can easily be seen from Table 2, previous analysis between pre-filtering, post-
filtering, and contextual modelling approach mainly evaluate the recommendation algorithms
based on the performance accuracy alone. To enhance users’ satisfaction and utility of
recommendation, the comparisons across these three paradigms need to consider other
evaluation measures such as serendipity, novelty, and diversity.
From the literature, each recommendation strategy results in different recommendation
outcomes. Therefore, companies would like specific recommendations methods for various
business application settings betting on factors concerning their business applications. Thus,
there is a necessity for in-depth comparisons to spot the particular factors that affect
recommendation algorithms and also the contextualization approaches. Moreover, the existing
analyses are limited to e-commerce domain. The comparisons need to cover other vital
domains identified from the literature (see [49]).
We hope that the current RS researchers will aim at performing series of comparisons
between the different variant of each of the pre-filtering, post-filtering and contextual modelling
approaches across different domains of applications and different experimental settings and
conditions. It will also be imperative for this researchers to propose some hybrid metrics to
induce a stronger balance between different evaluations measures.
6. Conclusion
In this paper, we provided a complete survey of the state-of-the-art comparisons
between pre-filtering, post-filtering, and contextual modelling approaches. We presented and
described the various novel approaches of each of the three contextualization paradigms. We
then pointed out significant challenges that require being addressed by the existing RS
researchers.
This paper is meant to assist researchers and practitioners in comparing the three
contextualization paradigms to select the best alternative according to their strategies. We hope
that this paper will trigger the current RS researchers in moving towards performing series of
analysis between pre-filtering, post-filtering, and contextual modelling approaches by
considering several factors and different scenarios to ascertain a deeper understanding of their
tradeoffs.

1873
References
[1] K Haruna, MA Ismail, AB Bichi, S Danjuma, HA Kakudi, T Herawan. An Overview on the State-Of-
The-Art Comparison of the Three-Contextualization Paradigms, The 8th International Conference on
Information Technology (ICIT 2017), May 17-18, 2017 Amman, Jordan, 2017.
[2] K Haruna, M Akmar Ismail, D Damiasih, J Sutopo, T Herawan. A collaborative approach for research
paper recommender system, PLoS ONE, 2017; 12: e0184516.
[3] K Haruna, MA Ismail. Scholarly Paper Recommendation Using Publicly Available Contextual
Metadata: Conceptual Paper, Seminar on Information Retrieval and Knowledge Management
(SIRKM'17), 19th July 2017, Faculty of Computer Science and Information Technology, Universiti
Putra Malaysia. 2017.
[4] M Erritali, A Beni-Hssane, M Birjali, Y Madani. An Approach of Semantic Similarity Measure between
Documents Based on Big Data, International Journal of Electrical and Computer Engineering
(IJECE), 2016; 6: 2454.
[5] GR Krishna, GH Priya, V Sravani, YS Chandu, GS Srikanth. Provision of Relevant Results on web
search Based on Browsing History, Journal of Telematics and Informatics, 2014; 2: 74-80.
[6] Y Yang, X Duan. Cliques-based Data Smoothing Approach for Solving Data Sparsity in Collaborative
Filtering, Indonesian Journal of Electrical Engineering and Computer Science (IJEECS), 2014; 12:
6324-6331.
[7] C Veena, BV Babu. A User-Based Recommendation with a Scalable Machine Learning Tool,
International Journal of Electrical and Computer Engineering (IJECE), 2015; 5.
[8] KA Almohsen, H Al-Jobori. Recommender Systems in Light of Big Data, International Journal of
Electrical and Computer Engineering (IJECE), 2015; 5.
[9] N Rosmawarni, T Djatna, Y Nurhadryani. A mobile ecotourism recommendations system using cars-
context aware approaches, TELKOMNIKA (Telecommunication Computing Electronics and Control),
2013; 11: 845-854.
[10] C Zeng, D Jia, J Wang, L Hong, W Nie, Z Li, et al. Context-aware social media recommendation
based on potential group, in Proceedings of the 1st International Workshop on Context Discovery and
Data Mining, 2012.
[11] V Codina, F Ricci, L Ceccaroni. Distributional semantic pre-filtering in context-aware recommender
systems, User Modeling and User-Adapted Interaction, 2015; 1-32.
[12] ZA Baizal, DH Widyantoro, NU Maulidevi. Factors Influencing User’s Adoption of Conversational
Recommender System Based on Product Functional Requirements, TELKOMNIKA
(Telecommunication Computing Electronics and Control), 2016; 14: 1575-1585.
[13] K Haruna, MA Ismail. An Ontological Framework for Research Paper Recommendation, International
Journal of Soft Computing, 2016; 11: 96-99.
[14] Q Liu. Context-aware Mobile Recommendation System Based on Context History, Indonesian
Journal of Electrical Engineering and Computer Science (IJEECS), 2014; 12: 3158-3167.
[15] G Adomavicius, A Tuzhilin. Context-aware recommender systems, in Recommender systems
handbook, ed: Springer, 2011: 217-253.
[16] J-y Hong, E-h Suh, S-J Kim. Context-aware systems: A literature review and classification, Expert
Systems with Applications, 2009; 36: 8509-8522.
[17] K Haruna, MA Ismail. Evaluation Techniques For Context-Aware Recommender Systems: A
Systematic Mapping.
[18] K Haruna, M Akmar Ismail, S Suhendroyono, D Damiasih, AC Pierewan, H Chiroma, et al. Context-
Aware Recommender System: A Review of Recent Developmental Process and Future Research
Direction, Applied Sciences, 2017; 7: 1211.
[19] M Bazire, P Brézillon. Understanding context before using it, in Modeling and using context, ed:
Springer, 2005: 29-40.
[20] AK Dey. Understanding and using context, Personal and ubiquitous computing, 2001; 5: 4-7.
[21] P Dourish. What we talk about when we talk about context, Personal and ubiquitous computing,
2004; 8:19-30.
[22] B Schilit, N Adams, R Want. Context-aware computing applications, in Mobile Computing Systems
and Applications, 1994. WMCSA 1994. First Workshop on, 1994: 85-90.
[23] U Panniello, A Tuzhilin, M Gorgoglione, Comparing context-aware recommender systems in terms of
accuracy and diversity, User Modeling and User-Adapted Interaction, 2014; 24: 35-65.
[24] G Adomavicius, A Tuzhilin. Context-aware recommender systems, presented at the Proceedings of
the 2008 ACM conference on Recommender systems, Lausanne, Switzerland, 2008.
[25] S Keele. Guidelines for performing systematic literature reviews in software engineering, in Technical
report, Ver. 2.3 EBSE Technical Report. EBSE, ed, 2007.
[26] C Okoli, K Schabram. A guide to conducting a systematic literature review of information systems
research, Available at SSRN 1954824, 2010.

 ISSN: 1693-6930
1874
[27] G Adomavicius, A Tuzhilin, Extending recommender systems: A multidimensional approach, in
Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI-01), Workshop on
Intelligent Techniques for Web Personalization (ITWP2001), Seattle, Washington, August, 2001; 4-6.
[28] L Baltrunas, B Ludwig, F Ricci. Matrix factorization techniques for context aware recommendation, in
Proceedings of the fifth ACM conference on Recommender systems, 2011; 301-304.
[29] L Baltrunas, F Ricci. Context-based splitting of item ratings in collaborative filtering, in Proceedings of
the third ACM conference on Recommender systems, 2009; 245-248.
[30] L Baltrunas, F Ricci. Context-dependent items generation in collaborative filtering, in Proceedings of
the 2009 Workshop on Context-Aware Recommender Systems, 2009; 22-25.
[31] G Adomavicius, R Sankaranarayanan, S Sen, A Tuzhilin. Incorporating contextual information in
recommender systems using a multidimensional approach, ACM Transactions on Information
Systems (TOIS), 2005; 23:103-145.
[32] L Baltrunas, F Ricci. Experimental evaluation of context-dependent collaborative filtering using item
splitting, User Modeling and User-Adapted Interaction, 2014; 24:7-34.
[33] L Baltrunas, X Amatriain. Towards time-dependant recommendation based on implicit feedback, in
Workshop on context-aware recommender systems (CARS’09), 2009.
[34] U Panniello, A Tuzhilin, M Gorgoglione, C Palmisano, A Pedone. Experimental comparison of pre-vs.
post-filtering approaches in context-aware recommender systems, in Proceedings of the third ACM
conference on Recommender systems, 2009, pp. 265-268.
[35] P Cremonesi, P Garza, E Quintarelli, R Turrin. Top-n recommendations on unpopular items with
contextual knowledge, in 2011 Workshop on Context-aware Recommender Systems. Chicago, 2011.
[36] H Wu, K Yue, X Liu, Y Pei, B Li. Context-aware recommendation via graph-based contextual
modeling and postfiltering, International Journal of Distributed Sensor Networks, 2015.
[37] U Panniello, M Gorgoglione. Incorporating context into recommender systems: an empirical
comparison of context-based approaches, Electronic Commerce Research, 2012;12: 1-30.
[38] K Oku, S Nakajima, J Miyazaki, S Uemura. Context-aware SVM for context-dependent information
recommendation, in Proceedings of the 7th international Conference on Mobile Data Management,
2006; 09.
[39] C Palmisano, A Tuzhilin, M Gorgoglione. Using context to improve predictive modeling of customers
in personalization applications, IEEE transactions on knowledge and data engineering, 2008; 20:
1535-1549.
[40] U Panniello, M Gorgoglione, C Palmisano. Comparing pre-filtering and post-filtering approach in a
collaborative contextual recommender system: an application to e-commerce, E-Commerce and Web
Technologies, 2009: 348-359.
[41] M Gorgoglione, U Panniello. Including context in a transactional recommender system using a pre-
filtering approach: two real e-commerce applications, in Advanced Information Networking and
Applications Workshops, 2009. WAINA'09. International Conference on, 2009: 667-672.
[42] U Panniello, M Gorgoglione. Does the recommendation task affect a cars performance, in RecSys10:
Workshop on context-aware recommender systems (CARS-2010), 2010.
[43] S Lombardi, M Gorgoglione, U Panniello, The effect of context on misclassification costs in e-
commerce applications, Expert Systems with Applications, 2013; 40: 5219-5227.
[44] MF Faraone, M Gorgoglione, C Palmisano, U Panniello. Using context to improve the effectiveness
of segmentation and targeting in e-commerce, Expert Systems with Applications, 2012; 39: 8439-
8451.
[45] U Panniello, M Gorgoglione. Context-Aware Recommender Systems: A Comparison Of Three
Approaches, in DART@ AI* IA, 2011.
[46] PG Campos, I Fernández-Tobías, I Cantador, F Díez. Context-aware movie recommendations: an
empirical comparison of pre-filtering, post-filtering and contextual modeling approaches, in
International Conference on Electronic Commerce and Web Technologies, 2013: 137-149.
[47] U Panniello, A Tuzhilin, M Gorgoglione. Comparing context-aware recommender systems in terms of
accuracy and diversity: which contextual modeling, pre-filtering and post-filtering methods perform
the best, 2012.
[48] M Ge, C Delgado-Battenfeld, D Jannach. Beyond accuracy: evaluating recommender systems by
coverage and serendipity, in Proceedings of the fourth ACM conference on Recommender systems,
2010: 257-260.
[49] K Haruna, MA Ismail, SM Shuhidan. Domain of Application in Context-Aware Recommender
Systems: A Review, Knowledge Management International Conference (KMICe) 2016, 29-30 August
2016, Chiang Mai, Thailand 2016.

A Comprehensive Survey on Comparisons across Contextual Pre-Filtering, Contextual Post-Filtering and Contextual Modelling Approaches

More Related Content

What's hot (17)

Similar to A Comprehensive Survey on Comparisons across Contextual Pre-Filtering, Contextual Post-Filtering and Contextual Modelling Approaches (20)

More from TELKOMNIKA JOURNAL (20)

Recently uploaded (20)

A Comprehensive Survey on Comparisons across Contextual Pre-Filtering, Contextual Post-Filtering and Contextual Modelling Approaches