User Profiling Approaches, Modeling, and Personalization
Marina Farid
Faculty of Computer and Information Sciences
Ain Shams University
Cairo, Egypt
marina.farid.12@gmail.com
Rania Elgohary
Faculty of Computer and Information Sciences
Ain Shams University
Cairo, Egypt
rania.elgohary@cis.asu.edu.eg
Ibrahim Moawad
Faculty of Computer and Information Sciences
Ain Shams University
Cairo, Egypt
ibrahim_moawad@cis.asu.edu.eg
Mohamed Roushdy
Faculty of Computer and Information Sciences
Ain Shams University
Cairo, Egypt
mroushdy@cis.asu.edu.eg
ABSTRACT
The growth of the available information on the internet and the
huge diversity of users give high priority to personalization. User
profiling is the crucial issue for both information and service per-
sonalization. Building an automated user profile is the main chal-
lenge in developing an adaptive personalized application. Recent
research works are involved in developing systems with person-
alizing user’s interaction. This paper examines what information
is needed to be modeled for presenting various user models, how
this information is collected, how the user model is represented
and maintained, and finally how the user model is exploited to
deliver personalized services. The paper investigates the current
studies on user profiling and modeling so far. A novel classification
schema is proposed for the study efforts in user profile building
and usage. The proposed classification schema includes three main
classes: data collection, profile representation and construction,
and personalization. Finally, we debate the current research gaps
found in this research area.
KEYWORDS
User profiling, Ontology, Dynamic, Semantic similarity, Personal-
ization
ACM Reference Format:
Marina Farid, Rania Elgohary, Ibrahim Moawad, and Mohamed Roushdy.
2018. User Profiling Approaches, Modeling, and Personalization. In
Pro-
ceedings of ACM INFOS conference (INFOS’18),
(Ed.). ACM, New York, NY,
USA, Article 4, 7 pages. https://doi.org/10.1145/1234567890
1
INTRODUCTION
Today, the rapid development of information caused an overload on
the web, which increases the demand for personalized approached
for finding the required information in time-consuming for users.
Research in personalization is used in several applications such
Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored.
For all other uses, contact the owner/author(s).
INFOS’18, December 2018, CAIRO, EGYPT
© 2018 Copyright held by the owner/author(s).
ACM ISBN 123-4567-24-567/08/06.
https://doi.org/10.1145/1234567890
as recommendation, e-learning, and personalized information re-
trieval, etc.
Profiling of a Web user is the process of identifying the data
about a user’s interest domain that constitute the user model [35].
A user profile is generally used for personalization and user mod-
eling. It reflects personal information or interesting facts about
an individual user, demographic information, e.g., name, age, ed-
ucation level, country, etc. It should capture the user’s behavior
(interesting topics, ratings, patterns, goals, etc.) when interacting
with the Web. User modeling is the process of gathering infor-
mation about a user’s interests, constructing, maintaining and
using user profiles. For example, in e-business systems, it captures
online users’ characteristics, finds similar online users, provides
customized products and services, and therefore improves the user
satisfaction.
User profiling techniques have widely applied in the various web
search, user-adaptive software systems, web user identification,
personalization, recommendation, e-market analysis, intelligent
tutoring systems, intelligent agents, as well as personalized in-
formation retrieval and filtering. The content and amount of the
information within a user profile can differ depending on the appli-
cation area. The accuracy of the user profile based on how the user
information is gathered and organized. Moreover, how accurate
this information reflects the user while building the profile.
There are two approaches in user modeling process: The gener-
ation of an initial user profile for a new user and the continuous
update of the profile information to adapt to the countinous chang-
ing of the user’s interests, preferences, and needs [14]. A major
challenge within user profiling is how the user profile can be built
to accurately reflects users’ preferences.
This paper examines the challenges in current personalization
framework; which depend for the most part on static or low-level
dynamic profiles, and lack in representing the multi-dimensionality
of the user profiles. A static personalization leads to recommen-
dations of irrelevant services as the user’s needs and interests
change because they use the same user information over time [37].
Another challenge is dealing with the cold start problem in recom-
mendation systems, and how to implicitly gather the information
with a high acceptance of trust-based model. To overcome these
challenges, building a user model that dynamically captures and
learns the user’s interests and preferences is required,
Electronic copy available at: https://ssrn.com/abstract=3389811
INFOS’18, December 2018, CAIRO, EGYPT
M. Farid et al.
various clustering and classification algorithms have been uti-
lized to create more comprehensive user profiles.
This paper explores some aspects of the user profiling system
in the existing literature on the user profile; and proposes an in-
novative classification schema for the research done in user pro-
file construction and personalization. The proposed classification
schema includes three main phases: data collection process, user
profile construction, and the personalized applications.
This paper is organized as follows: Section 2 presents the pro-
posed classification schema for the user profiling research work,
whereas Section 3, 4, and 5 conduct surveys the research efforts for
data collection, building and modeling a user profile, and personal-
ization respectively. The main gaps found in the current research
works represented in section 6. Finally, the conducted study is
summarized in section 7.
2
CLASSIFICATION SCHEME OF USER
PROFILING
In this section, we provide a comprehensive study of recent re-
search on user profiling. Figure (1) presents the proposed classifi-
cation schema of research studies on user profiling. The modeling
process studies first the information constituting the user profile
and the gathering of raw data about the user and how it can be
extracted. Second, the building and maintaining process are investi-
gated, which way for representing a user model and the techniques
of its construction. Third, how to identify the user and how to ex-
plore the applications in the context of user profiling in order to
provide personalized services.
3
DATA COLLECTION
The principal phase of user profiling is to gather data about the
user. Obtaining and extracting user data, and knowing his or her
interests is a major task for making personalized systems. For this
reason, the nature of data accessible for a personalized system
determines the way and the recommending items that the system
can execute personalization.
3.1
User Profile Content
There are several types of data that can exist in a user profile: user
personal data, user background knowledge and skills; user needs
and goals, which are static data types [4]. Alternatively, user’s
behavior, user’s interests and preferences, and context are dynamic.
The following subsections are sequenced by the dynamism of data
from the low-level data to the high-level dynamic data.
3.1.1
Demographic data.
The demographic characteristics of a
user are the very basic features, like name, country, gender, age
or native language, education, family members, etc. Prajakta et
al. [30] compared some frameworks using different data mining
and machine learning approaches for modeling and prediction
of demographic data. Fernandez et al. [7] proposed a model that
captures the demographic information about the user as well as
his/her actions within online communities. Pazzani et al. [25],
group users based on their demographic data (such as location,
work, language, etc.) for building a classifier for learner.
3.1.2
Background, Knowledge and Skills.
These are especially im-
portant for modeling students in tutoring systems and adaptive
education systems. The student’s knowledge is the main charac-
teristic defining an adaptive system and his knowledge about the
subject taught is important to provide proper assistance to the stu-
dent or to adapt the content of courses according to it [8]. Different
systems manage differently the challenge of incorrect knowledge
or misconceptions. In some recommended systems, the skills a user
or employee has, the roles he takes within an organization and his
performance in these roles are required to provide the appropriate
jobs for the job applicant and recommend some suitable candidates
to a recruiter [12].
3.1.3
Goals and needs.
The goals of a user are important to detect
his purpose from the application he is using with, that is what
the user wants to achieve. The information need is the goal of
obtaining relevant information when the user is browsing the Web.
It is crucial to learn from for goal prediction in some intelligent
dialog systems. These modeling is nearby to dimensions is the
concept of needs, requirements and desires NRD which have been
applied in personalized service delivery systems to adapt user’s
behavior. Periklis et al [35], discussed that user’s interests specified
explicitly in the form of keywords, can be treated as goals to be
achieved during a browsing session, these goals are stored on the
server side.
3.1.4
Behavior.
Behavior is a type of information that is gathered
implicitly. It is important that the behavior is repetitive in order to
detect patterns that used by an adaptive system or an intelligent
agent to adopt a website or to assist the user’s behavior. Jie et
al. [36] used three types of user behaviors: attention time, click
through and mouse movement as potential indicators of users’
interests to represent users’ preference as implicit feedback. Other
systems as in [30] only keep track of click-through user’s behav-
ior, which comprises submitted queries and clicked documents. A
similar adaptive strategy also proposed based on user behaviors
[2],[13].
3.1.5
Interests and Preferences.
Interests and preferences of a user
are a main part of the personalization. Interests can signifie so-
cial topics, website topics, document topics or job topics. These
topics can be derived from various sources, such as purchasing
history or browsing history. Sometimes user interests are classi-
fied as short-term or long-term interests [15]. Some frameworks
perform personalization based on both, long-term interests as pref-
erences and short-term interests. Li et al. [17] Modeled different
representations for frequent and long-term user activities based
on preferences of the user. Long-term profiles are built from the
Google Directory using topics of clicked streams results, and short-
term models using a cache of recently clicked results.
3.1.6
Other Context.
Context is identified as any information;
used to characterize the state of any item. Jie et al. [36] used present
a supervised macine learning approach to extract a user profile
from social media like Twitter. They discussed some adaptive sys-
tems based on the user’s context where information is displayed
depending on the current location of the user. A modeling user
profile based on users’ interactions with social networks, tags, and
annotations in social bookmarking systems, and posts on the Web
Electronic copy available at: https://ssrn.com/abstract=3389811
User Profiling Approaches, Modeling, and Personalization
INFOS’18, December 2018, CAIRO, EGYPT
Figure 1: A New Classification Schema of user profiling.
such as the follow/followed connections in Twitter [18]. Knowledge
also about a user includes his likes, dislikes, even his emotional
state can be determined by using profiling. Emotions of a user have
a direct impact on how he uses the software application. Typing
speed and mouse clicks are the major parameters for the detection
and recognition of basic human emotions.
3.2
Data Collection Approaches
The data collection methods are explicit, implicit, and hybrid [28].
The data collection can be manual, semi-automatic, and automatic
profiles respectively. The accuracy of the user profile depends on
the amount of generating data through user-system interaction
Moreover, where the gathered or processed information is main-
tained; the personalization process in terms of when and where
the information is available to be exploited for personalization. We
discuss in Figure (1) the methods of how the data are gathered and
what the resources of each method are.
3.2.1
Explicit Data.
The explicit user information collection ap-
proaches are acquired using manual techniques through user in-
tervention such as registration forms, questionnaires, or user clas-
sified training sites, or by asking users to rate items, or by tracking
users’ query words. For example, eBay asks users to provide their
opinions and to give ratings for the services and products on offer.
The company then utilizes this information to improve the per-
sonalized recommendations that it gives to customers. My Yahoo
explicitly asks the user to provide personal information that is
stored to create a profile. In recommender systems, explicit rat-
ing data are widely used to profile users’ preferences. Tags are
also commonly used as direct interest keywords when they are
attached by the user to web content or using social websites [14].
Some researchers use users’ comments and posts directly to extract
keywords to represent users’ interests [19], while others directly
use the rated items as an indication of users’ interest [20]. The
explicit profiles have also a static nature and are valid only until the
user changes their interest and preference parameters explicitly.
To overcome this problem by extracting the information implicitly
by tracking the users’ activities and/or behavior [28].
3.2.2
Implicit Data.
The implicit method attempts to infer the
user’s interests or context from the processed logs or the infor-
mation gathered automatically without any effort from the user.
Unlike static profiling, automatic profiling uses the implicit method
and analyzes the user’s behavior pattern (e.g. Usage history) to de-
termine a user’s interests and developing user background knowl-
edge. The implicit user profiling can be either client-side or server-
side. Browser agents as used in [15] collect information interac-
tively, while the user browses. It must be installed on the user’s
desktop computer to capture user activity and actions performed
on the Web page such as tagging, bookmarking and download-
ing. Shen et al. [29] perfom an implicit feedback on an intelligent
client-side web search agent (UCAIR) based on query expansion
and click-through information. Fernandez et al. [7] predicted the
demographic attributes of the web users from their Web browsing
behaviors. Moawad et al. [21] proposed an agent-based system
for initialing the web search results. This model builds an implicit
user profile from bookmarks and browsing activities using weblogs
and search interactions using search logs. The user query semanti-
cally optimizes using user profile interests and WordNet. However,
several projects [17] , [21], [28], [14] and [27]] have been able to
provide personalized search successfully by building user profiles
based on this information. Drawbacks of this approach, much less
data is available and sometimes raise privacy issues for the users.
3.2.3
Hybrid Data.
Hybrid user profiles are gained by semi-
automated techniques with restricted user involvement. The
system collects user and usage information in a mixed approach
Electronic copy available at: https://ssrn.com/abstract=3389811
INFOS’18, December 2018, CAIRO, EGYPT
M. Farid et al.
of implicit as manual and explicit methods. This approach helps
to profile more efficient, and maintains the accuracy of tempo-
ral information as information gets updated temporally [18],[31].
Kobsa [16] created a keyword-based profiling by combining explicit
feedback as forms of personal information and a set of interest
keywords and implicit information sources as web bookmarks
and user’s browsing histories, respectively. Another hybrid pro-
filing system [30] generated both explicit and implicit profiles
independently and then combined into a single profile. In which
the explicit user profile is limited to asking the users to specify their
interests in terms of keywords. The implicit user profile is created
by constantly monitoring user’s activities by storing the browsing
history, the time spend on each page, and additional actions like
printing or saving. It combines explicitly state interests with the
observation of user behavior. Fakhfakh [6] used the explicit and
the implicit models for the acquisition of user preferences. In an
explicit way, they interviewed 30 users for their preference and in
the other way, user preferences are discovered through the Movie
Lens datasets collected from its website which represent a service
of movie recommendation.
4
BUILDING AND MODELING USER PROFILE
4.1
User Profile Representation
This section of the survey is concerned with the second phase of
user profiling, which is the storage and representation of the infor-
mation gathered in the previous phase. These profiles, summarized
in [9], are represented by a previously prepared collection of data
reflecting user interests. In this following section, we discuss three
types of user models which are vector-based, semantic-networks,
or concept-based.
4.1.1
Term-Based (Vector-Space Model).
A keywordśbased or
vector-based user model is the most common representation
of users’ interests. It is made up of a collection of data refered to
as a set of terms (or vector space of terms) by weighted vectors
of keywords that can be directly used to expand the users queries
[29]. Boolean, (TF) and (TF-IDF) are different methods to develop a
representation for each weighted term of the documents or queries.
Moreover, the similarity between two weighted term vectors is
calculated by Cosine Similarity (CS). User interests keywords are
extracted from visited documents during browsing. They are repre-
sented either by a single vector that includes all the interest or with
multiple vectors, which reflects interest in several domains [9]. In
this model the effectiveness of the user profiles depends on the
vectors’ degree of generalization. Its main drawbacks are polysemy
and that it may cause poor interpretation of user interests to the
users. Yasser et al. [22] improved their initial term-based model
(IUM) by building a semantically enhanced user model (SUM) in
which terms are mapped to related high-level concepts. A fuzzy
technique is employed to classify the learners according to their
interests and the Felder-Silverman model. Wang et al. [12] pro-
posed a new method of extraction algorithm based on combining
the Word2Vec which is a word vector training tool published by
Google in 2013, and the TF-IDF for representing the weight vector
for solving shortcomings of the vector representation method.
4.1.2
Semantic Network-Based (Ontology-based).
In order to ad-
dress the polysemy ambiguity and synonymy problems inherent
with keyword-based profiles, the profiles may be represented by a
weighted concept of each node in the semantic network. However,
this also places a burden to the ease of constructing such a system.
An existing mapping is required between words of interests and
concepts, such as: WordNet ontology, or through building a learn-
ing system method or manually. Liu [18] generated semantically
enhanced user profiles that reflect the users’ interests and inten-
tions in a set of specific categories. An ontology-based user model
is proposed in [31], which facilitates a vast number of different ap-
plications to semantically interact with the same knowledge base,
and also represents interests in the form of ontological concepts
interlinked with predefined source ontology by using concepts
from Wikipedia and WordNet. Skillen et al. [32] described the im-
provement of ontological user profile in mobile environments for
context-aware system. A method proposed in [34] for enhancing
presentation in the social web using ontological knowledge rep-
resentation including FOAF, SIOC, and DC that enables semantic
profiling in social websites.
4.1.3
Concept–Based (Hierarchy-based).
Concept-based profiles
are same as semantic network-based profiling in representing by
nodes of concepts and relationships between those nodes. Various
mechanisms are applied to associate a weight of how much the
user is interested in each topic. N. Nanas et al. [23] have used
different techniques for categorizing the user interests in the form
of some hierarchy and used a knowledge base for this purpose. The
researchers used library classification system (LCS) for finding user
interest hierarchies and (TF-IDF) mechanism used for calculating
the edge weight.
4.2
User Profile Construction
There are two aspects that need to be considered when creating a
keyword profile. The first is what information source(s) to use and
the second is how to create the profile with the aid of the selected
sources. User profiles techniques are constructed based on machine
learning or data mining or information retrieval. Depending on the
user profile representation desired, different techniques may be
appropriate. It is important that the chosen construction method,
keep the profile updated to reflect the user’s preferences accurately;
this has proven to be a very challenging task [14].
4.2.1
Information Retrieval.
The information retrieval systems
become more interactive, which require a continuous learning with
the system to improve retrieval. After the extraction of relevant
information is done there might be data preprocessing problems
like noisy data, duplicate data or it may look like duplicate data
but that might be a unique data. To resolve this issue, unique and
duplicate data are needed to make data cleaning. In Information
Retrieval approach, both the document and the user interests are
represented in the form of weighted term vectors or by presenting
topic vectors. This strategy suffers from issues of polysemy and
synonymy and furthermore the unrelated web page content such as
(homepage, contact, etc.). Fakhfakh et al. [27] developed association
rules based framework for personnel selection between personnel
characteristics and work activities. Enrique et al. [8] overviewed
Electronic copy available at: https://ssrn.com/abstract=3389811
User Profiling Approaches, Modeling, and Personalization
INFOS’18, December 2018, CAIRO, EGYPT
different data mining techniques that can be utilized to effectively
capture user’s behavior.
4.2.2
Machine Learning Approach.
Machine Learning is popularly
used for document classification in a collection by topics. The
resulting topics can be used as keywords or concepts to constitute
a user model. Some papers tackle the problem of automatic web
personalization using machine learning techniques to model the
users’ behavior and user interest learning. Decision trees, Naive
Bayesian classifier and K Nearest Neighbors used for identifying
the topics of documents by learning classifications from a labeled
set of training documents. Quickstep is a recommender system
which classified each research paper with one topic, derived from
DMOZ’s computer science categories using machine learning [27].
Some projects used Latent Semantic Indexing (LSI) and Linear Least
Squares Fit (LLSF) for creating the keyword-based feature vectors
[36]. Song et al. [33] used to predict some data about user tasks with
a set of latent parameters, and mapping documents with the term.
Liu [18] modeled a dynamic user preference for recommendation
by using a probabilistic topic model Latent Dirichlet Allocation
(PLDA) to represent a user’s temporal preference.
4.2.3
Ontology-based Approach.
Recent approaches manifest the
adaptive web and the semantic web [34] together by using stan-
dardized web ontology languages and their software base. In this
approach, user profiles are generated based on some reference
ontology, for example, Open Directory hierarchy as an ontology to
find user interests [38]. Reference ontology is able to resolve or at
least reduce the issue of ontology application specificity and might
provide important advantages over the domain and application
ontology previously used [41]. The use of ontology in acquiring
knowledge domains about a user’s preferences so this ontology is
constructed based on standard advertisements taxonomy for user
profiles, OpenDNS taxonomy, FOAF ontology, and DBpedia ontol-
ogy. The ontological user profile [18] allows the relevant topics of
interest that a user has can be easily identified and reasoned. A user
profile is demonstrated on RDF (Resource Description Framework)
/XML format or OWL (Ontology Web Language). User profile
has been automatically built by extracting URLs posted by users
in tweets, user profile ontology, OpenDNS, DBpedia and NoSQL
databases and Reasoner. Bouzeghoub [5] proposed a multi-agent
learner profile model joining IMS LIP and FOAF standards of con-
text and agenda information, collected from social networking
providers.
4.2.4
Agent-based Approach.
The agent-based system technique
used to improve the information retrieval recall and precision crite-
ria through constructing multi-agents for addressing the different
personalization challenges and phases. It allows creating experi-
ments in the form of models and running those models to give as
output the representation of the interaction between the agents.
An agent-based personalized web search (WSP) model in [10] pro-
posed for retrieving better search results for the user’s preferences
based on multi-agent system technique, in addition to solving the
problem of polysemy and synonymy for the keyword-based profile
using Word Net ontology and exploiting user preferences. There
are other methods such as agent-based model for context-aware
and personalized news recommendations were proposed by re-
searchers.
5
PERSONALIZATION
The third phase involves exploiting information in a user profile to
provide personalized services. After a user profile is constructed,
it is then used to provide personalized services in different areas,
such as personalized recommender systems, mobile services, social
network, web search, and browsing and etc.
5.1
User Identification
Accurate user identification is essential for any system that con-
structs profiles that represent individual users. There are five basic
approaches to user identification: software agents, logins, enhanced
proxy servers, cookies, and session ids. Qian Gao et al. [15], use
two option methods that are adapted to identify the users: one is
cookies for current sessions, and the other one is logins for users
who choose to register with a site. Web data mining can be used to
identify users and find learner preferences based on the weblogs
existing on servers [22]. Weblog records each transaction, which
was executed by the browser at each web access. With such data,
we follow sequences in visiting individual pages by the learner,
who is, under a certain condition, identified by the IP address.
5.2
Applications
5.2.1
Personalized Web Search and Browsing.
Personalized search
is the process of searching a set of documents or web pages by
giving preference to the user’s interest. User interest is identified
by from the user profile, which is created by analyzing the sites
searched by the user. The accuracy of the system is enhanced by
predicting the user’s interest better compared to an usual search.
A personalized search and browsing in [3] proposed a concept
network for retrieving web pages based on the user’s preferences,
and improving navigation with taking into consideration the dy-
namism changing of the preferences and the new concept while
searching. A recent research follows the ontology-based profil-
ing for personalized search approaches [28] by using the Open
Directory Project (ODP) taxonomy as the Web topic ontology.
5.2.2
Recommendation and Prediction Systems.
A recommender
system is a subclass of the information filtering system that look
for predicting the item-rating or user’s preference. A personalized
recommendation system has two important aspects; as efficiently,
knowing user and based on those recommending items of his in-
terest. Thorat et al. [37] Surveyed different techniques of a recom-
mender system which were based on demographic, content-based
filtering, collaborative filtering and hybrid filtering. User profiling
could obtain adequate and accurate information about a user’s
interests and characteristics and establish them with minimum
user intervention [2] to offer appropriate awareness with a rele-
vant mitigation recommendation based on the security position.
Hong et al [11] proposed a job recommendation system, based
on time and dimensionality. In this they dynamically updated the
candidate’s profile over the user’s interests feature sets, extracted
over a time period, through the information gain concept. Zhang
et al. [42] have explored a hierarchy-based profiling by creating
Electronic copy available at: https://ssrn.com/abstract=3389811
INFOS’18, December 2018, CAIRO, EGYPT
M. Farid et al.
taxonomies based on the terms in the item description. An inter-
esting technique in [1] is developed for solving the problem of
the tree of concepts user profile representation. A model Dynamic
Normalized Tree of Concepts (DNTC) is developed for user profiles
and a research paper recommender system using the 2012 ACM
ntology. Quercia et al. [26] investigated the correlation between
users’ personality and the attributes of their social web profiles.
They indicated that some personality traits such as Extraversion
and Neuroticism can be accurately predicted based on the user’s
profile on Twitter. An approach [18] based on the generation of
user profiles is proposed in order to provide advertisements and
services that really interests which are extracted and obtained
through URLs as they share their tweets.
5.2.3
Adaptive Learning Systems.
Adaptive learning is a branch
of e-learning, which helps the learners in their learning and study
path. Nguyen [24] proposed a user modeling framework called
Zebra which operates Triangular Learner Model (TLM) of students’
characteristics and provides a powerful inference mechanism for
reasoning out new information about students.
5.2.4
Visualization.
Yingze et al. [39] reported a usability study
of the Introspective-Views interface for visualizing semantic user
models in social networks. They described how the interface can be
used in a social network for visualizing and editing ontology-based
user interest models.
5.2.5
Personalized Online Social Network Services.
Personalized
social networking applications automatically classify the data into
different but most relevant categories; provide most relevant data
to the users of his/her interests. A user preference profiling tech-
nique [14] presented that generates classification model for user
preferences based on user behavior on Facebook such as like, share
and post. The model is created from three classification models and
concludes that the support vector machine has better performance
than ANN and NB algorithms. A proposed technique used Folk-
sonomies for gathering the user’s personal data and personomies,
via collaborative tagging system [40].
6
RESEARCH GAPS
During our extensive study of user profiling, several research gaps
have been deduced throughout the previous studies and literature
concerned with user profiling approaches and personalization.
These gaps include:
•
Need to improve the accuracy of retrieving the user percep-
tion with the physical charateristics of data. Therefore, it
should exist a learning approach to check the user input by
using relevance feedback [29].
•
With the power of information retrieval and machine learn-
ing, we need to implicitly discover the ontological user pro-
file, and deal with the Cold Start problem and sparsity of
user profiles [25].
•
Need for mining the uncertainty of inherent in profile data
by associating profiles with weights, through explicit rea-
soning techniques [9].
•
We need ontology-based model for describing the variety of
items of interest and user profiles. In order to increase the
acceptance of trust-based models of recommender systems
semantically [20].
•
There is a need for for expanded consideration regarding
topical/tag-based models of trust in the context of heteroge-
neous content for web processing [19]. Limited considera-
tion has been paid on exploiting link and graph structures
among resources and people especially in the context of
on-line social media [40].
7
CONCLUSION
In this paper, we we have given an extensive review of works
done on the aspects of user profiles including related concepts,
methods, techniques, in addition to the existing solutions in the
literature. In addition, it discusses the pros and cons of the user
profiling methods together with commonly used techniques and
user profile information. A dynamic user profiling approach which
is an exciting and promising approach that aims to better under-
stand user requirements semantically, to provide the user with an
adequate and suitable service. The personalized applications fall
into three main categories; interest discovery, visualization and
recommendation. Finally, we have highlighted research gaps in
those areas to call for future work.
ACKNOWLEDGMENTS
We are grateful to Professor Birgitta König-Ries, the Heinz-Nixdorf
Chair for Distributed Information Systems at Friedrich-Schiller-
Universität Jena, Germany, for her technical support and valuable
discussions. A part of this research was supported by Biodialog
project.
REFERENCES
[1] Modhi Al Alshaikh, Gulden Uchyigit, and Roger Evans. 2017. A research paper
recommender system using a Dynamic Normalized Tree of Concepts model for
user modelling. In
Research Challenges in Information Science (RCIS), 2017 11th
International Conference on
. IEEE, 200ś210.
[2] Bhushan S Atote, Taranpreet Singh Saini, Mangesh Bedekar, and Saniya Zahoor.
2016. Inferring emotional state of a user by user profiling. In
Contemporary
Computing and Informatics (IC3I), 2016 2nd International Conference on
. IEEE,
530ś535.
[3] KR Remesh Babu and Philip Samuel. 2015. Concept Networks for Personalized
Web Search Using Genetic Algorithm.
Procedia Computer Science
46 (2015),
566ś573.
[4] Nor Adzlan bin Jamaludin, Muthukkaruppan Annamalai, Nursuriati Jamil, and
Zainab Abu Bakar. 2013. A model for keyword Profile Creation using extracted
keywords and terminological ontology. In
e-Learning, e-Management and e-
Services (IC3e), 2013 IEEE Conference on
. IEEE, 136ś141.
[5] Amel Bouzeghoub and Alda Lopes Gancarski. 2011. An agent-based service
architecture for user profiles dynamic share. In
2011 Seventh International Con-
ference on Semantics, Knowledge and Grids
. IEEE, 137ś143.
[6] Rim Fakhfakh, Ghada Feki, Anis Ben Ammar, and Chokri Ben Amar. 2016.
Personalizing information retrieval: a new model for user preferences elicitation.
In
2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC)
.
IEEE, 002091ś002096.
[7] Miriam Fernandez, Arno Scharl, Kalina Bontcheva, and Harith Alani. 2014. User
profile modelling in online communities. (2014).
[8] Enrique Frias-Martinez, Sherry Y Chen, and Xiaohui Liu. 2006. Survey of data
mining approaches to user modeling for adaptive hypermedia.
IEEE Transactions
on Systems, Man, and Cybernetics, Part C (Applications and Reviews)
36, 6 (2006),
734ś749.
[9] Susan Gauch, Mirco Speretta, Aravind Chandramouli, and Alessandro Micarelli.
2007. User profiles for personalized information access. In
The adaptive web
.
Springer, 54ś89.
[10] Ahmad Hawalah and Maria Fasli. 2011. A multi-agent system using onto-
logical user profiles for dynamic user modelling. In
Proceedings of the 2011
IEEE/WIC/ACM International Conferences on Web Intelligence and Intelligent
Agent Technology-Volume 01
. IEEE Computer Society, 430ś437.
Electronic copy available at: https://ssrn.com/abstract=3389811
User Profiling Approaches, Modeling, and Personalization
INFOS’18, December 2018, CAIRO, EGYPT
[11] Wenxing Hong, Siting Zheng, and Huan Wang. 2013. Dynamic user profile-
based job recommender system. In
Computer Science & Education (ICCSE), 2013
8th International Conference on
. IEEE, 1499ś1503.
[12] Jianqiao Hu, Feng Jin, Guigang Zhang, Jian Wang, and Yi Yang. 2017. A User
Profile Modeling Method Based on Word2Vec. In
Software Quality, Reliability
and Security Companion (QRS-C), 2017 IEEE International Conference on
. IEEE,
410ś414.
[13] Rachsuda Jiamthapthaksin and Than Htike Aung. 2017. User preferences pro-
filing based on user behaviors on Facebook page categories. In
Knowledge and
Smart Technology (KST), 2017 9th International Conference on
. IEEE, 248ś253.
[14] Diane Kelly and Jaime Teevan. 2003. Implicit feedback for inferring user prefer-
ence: a bibliography. In
Acm Sigir Forum
, Vol. 37. ACM, 18ś28.
[15] JeeHyun Kim, Qian Gao, and Young Im Cho. [n. d.]. A Context-Aware based
Dynamic User Preference Profile Construction Method. ([n. d.]).
[16] Alfred Kobsa. 2001. Generic user modeling systems.
User modeling and user-
adapted interaction
11, 1-2 (2001), 49ś63.
[17] Lin Li, Zhenglu Yang, Botao Wang, and Masaru Kitsuregawa. 2007. Dynamic
adaptation strategies for long-term and short-term user profile to personalize
search. In
Advances in Data and Web Management
. Springer, 228ś240.
[18] Xin Liu. 2015. Modeling Users’ Dynamic Preference for Personalized Recom-
mendation.. In
IJCAI
. 1785ś1791.
[19] Chunliang Lu, Wai Lam, and Yingxiao Zhang. 2012. Twitter user modeling and
tweets recommendation based on wikipedia concept graph. In
Workshops at the
Twenty-Sixth AAAI Conference on Artificial Intelligence
. 33ś38.
[20] Hao Ma, Dengyong Zhou, Chao Liu, Michael R Lyu, and Irwin King. 2011.
Recommender systems with social regularization. In
Proceedings of the fourth
ACM international conference on Web search and data mining
. ACM, 287ś296.
[21] Ibrahim F Moawad, Hanaa Talha, Ehab Hosny, and Mohamed Hashim. 2012.
Agent-based web search personalization approach using dynamic user profile.
Egyptian Informatics Journal
13, 3 (2012), 191ś198.
[22] Yasser A Nada and Khaled M Fouad. 2011. An approach to improve the repre-
sentation of the user model in the web-based systems.
IJACSA) International
Journal of Advanced Computer Science and Applications
2, 12 (2011), 152ś160.
[23] Nikolaos Nanas, Victoria Uren, and Anne De Roeck. 2003. Building and applying
a concept hierarchy representation of a user profile. In
Proceedings of the 26th
annual international ACM SIGIR conference on Research and development in
informaion retrieval
. ACM, 198ś204.
[24] Loc Nguyen. 2014. A user modeling system for adaptive learning. In
Interactive
Collaborative Learning (ICL), 2014 International Conference on
. IEEE, 864ś866.
[25] Michael J Pazzani and Daniel Billsus. 2007. Content-based recommendation
systems. In
The adaptive web
. Springer, 325ś341.
[26] Daniele Quercia, Michal Kosinski, David Stillwell, and Jon Crowcroft. 2011.
Our twitter profiles, our selves: Predicting personality with twitter. In
Privacy,
Security, Risk and Trust (PASSAT) and 2011 IEEE Third Inernational Conference on
Social Computing (SocialCom), 2011 IEEE Third International Conference on
. IEEE,
180ś185.
[27] Krishnan Ramanathan, Julien Giraudi, and Ajay Gupta. 2008. Creating hierar-
chical user profiles using Wikipedia.
HP Labs
6 (2008), 163ś170.
[28] Silvia Schiaffino and Analía Amandi. 2009. Intelligent user profiling. In
Artificial
Intelligence An International Perspective
. Springer, 193ś216.
[29] Xuehua Shen, Bin Tan, and ChengXiang Zhai. 2005. Implicit user modeling for
personalized search. In
Proceedings of the 14th ACM international conference on
Information and knowledge management
. ACM, 824ś831.
[30] P. Shitole and M. Potey. 2015. Focusing User Modeling For Age Specific Dif-
ferences.
International Journal of Emerging Trends & Technology in Computer
Science (IJETTCS)
4, 3 (2015).
[31] Ahu Sieg, Bamshad Mobasher, and Robin Burke. 2007. Web search personaliza-
tion with ontological user profiles. In
Proceedings of the sixteenth ACM conference
on Conference on information and knowledge management
. ACM, 525ś534.
[32] Kerry-Louise Skillen, Liming Chen, Chris D Nugent, Mark P Donnelly, William
Burns, and Ivar Solheim. 2012. Ontological user profile modeling for context-
aware application personalization. In
International Conference on Ubiquitous
Computing and Ambient Intelligence
. Springer, 261ś268.
[33] Rui Song, Enhong Chen, and Min Zhao. 2005. SVM based automatic user profile
construction for personalized search. In
International Conference on Intelligent
Computing
. Springer, 475ś484.
[34] Sergey Sosnovsky and Darina Dicheva. 2010. Ontological technologies for user
modelling.
International Journal of Metadata, Semantics and Ontologies
5, 1
(2010), 32ś71.
[35] Hamza Hydri Syed and Periklis Andritsos. 2007.
User preference modeling-a
survey report
. Technical Report. University of Trento.
[36] Jie Tang, Limin Yao, Duo Zhang, and Jing Zhang. 2010. A combination approach
to web user profiling.
ACM Transactions on Knowledge Discovery from Data
(TKDD)
5, 1 (2010), 2.
[37] Poonam B Thorat, RM Goudar, and Sunita Barve. 2015. Survey on collabo-
rative filtering, content-based filtering and hybrid recommendation system.
International Journal of Computer Applications
110, 4 (2015).
[38] Joana Trajkova and Susan Gauch. 2004. Improving ontology-based user pro-
files. In
Coupling approaches, coupling media and coupling languages for in-
formation retrieval
. LE CENTRE DE HAUTES ETUDES INTERNATIONALES
D’INFORMATIQUE DOCUMENTAIRE, 380ś390.
[39] Yingze Wang and Shi-Kuo Chang. 2014. User Profile Visualization to Facilitate
MSLIM-Model-Based Social Influence Analysis Based on Slow Intelligence Ap-
proach.
International Journal of Software Engineering and Knowledge Engineering
24, 10 (2014), 1573ś1587.
[40] Au Yeung, Ching Man, Nicholas Gibbins, and Nigel Shadbolt. 2008. A study of
user profile generation from folksonomies. (2008).
[41] Leila Zemmouchi-Ghomari and Abdessamed Reda Ghomari. 2009. Reference
Ontology. In
2009 Fifth International Conference on Signal Image Technology and
Internet Based Systems
. IEEE, 485ś491. https://doi.org/10.1109/SITIS.2009.81
[42] Yuchen Zhang, Amr Ahmed, Vanja Josifovski, and Alexander Smola. 2014. Tax-
onomy discovery for personalized recommendation. In
Proceedings of the 7th
ACM international conference on Web search and data mining
. ACM, 243ś252.
Electronic copy available at: https://ssrn.com/abstract=3389811
|
