Smart mobiles as the most affordable and practical ubiquitous devices participate heavily in the enhancement of our daily life by the use of many convenient applications. However, the significant number of mobile users in addition to their heterogeneity (different profiles and contexts) obligates developers to enhance the quality of their apps by making them more intelligent and more flexible. This is realized mainly by analyzing mobile user’s data. Machine learning (ML) technology provides the methodology and techniques needed to extract knowledge from data to facilitate decision-making. Therefore, both developers and researchers affirm the benefits of combining ML techniques and mobile technology in several application fields as e-health, e-learning, e-commerce, and e-coaching. Thus, the purpose of this paper is to have an overview of the use of ML techniques in the design and development of mobile applications. Therefore, we performed a systematic mapping study of papers published on this subject in the period between 1 January 2007 and 31 December 2019. A total number of 71 papers were selected, studied, and analyzed according to the following criteria, year, sources and channel of publication, research type, and methods, kind of collected data, and finally adopted ML models, tasks, and techniques.
1. Introduction
Today, no one can deny that the most used ubiquitous systems are mobile devices. At the first of their apparition, they were mainly used to send and receive calls, SMS, MMS, and emails. However, nowadays mobiles are widely used in several challenging domains such as e-learning, e-health, e-commerce, e-travel, and e-coaching. Consequently, from a simple container of thin clients, mobiles are transformed to a container of a huge number of mobile applications offering various services of our daily life.
The total number of mobile app downloads has increased from 63 billion in 2015 to 2054 billion in 2018. Also, it is expected that in 2022 this number will reach 2582 billion [1]. Moreover, since 2015, mobile phone usage exceeded desktop Internet usage, and by 2025, connected users will reach 75% of the world’s population [2]. Therefore, it is expected that by 2025 mobile data will constitute 18% of the global Datasphere [2].
This transformation and this variation on targeted business areas in addition to the large number of targeted users make mobile apps design and development one of the hottest topics in software engineering [3–6]. Recently, mobile user satisfaction and assistance became also a primary interest of both researchers and developers. The objective is to develop very attractive and easier mobile apps to facilitate and improve the quality of several services [7–9].
Therefore, following a user-centric approach from the beginning of the mobile app design and development process became inevitable. One of the ultimate objectives of this approach is to adapt mobile apps (interface and/or logics) to diversifications in user profile, context changes, and also the history of previous actions [8, 10]. This adaptation to user’s data must last over time to provide an intelligent interaction with mobile users and to continuously improve the quality of provided services [10].
Machine Learning (ML) techniques are widely used to extract knowledge from data and can be used in the field of mobile application design and development to ensure intelligent interactions with mobile users. Therefore, many works present the benefits of combining ML techniques and mobile technology in several application domains. For instance, in [11] the authors present the need in developing smarter, more personalized and efficient patient-centric m-Health models. Also, they present the current and future roles of machine learning in making intelligent decisions based on the patient’s data. In [12] the authors affirm that the future of learning is the synergy of mobility, interaction, and machine learning because by combining them it is possible to create intelligent and interactive mobile models that provide innovative learning scenarios. In [13] the authors define the analysis of mobile phone datasets and networks as one of the most relevant topics in today’s m-commerce research. The objective is to improve the quality of provided services by developing user-centric m-commerce apps that exploit the power of ML techniques in the analysis of user-profiles and behavior.
Recently, the introduction of machine learning techniques in the creation of intelligent mobile apps is the subject of several research works of which the number keeps increasing year by year. This new orientation towards mobile applications is encouraged by the growth in performance of smartphones in terms of CPU power, RAM capacity, and energy storage, and also it is due to advances in the field of cloud computing that provide an on-demand cloud services of data storage and computing power [14]. Thus, to structure this research axis and to facilitate the information extraction about various available publications, it becomes imperative to carry out a literature study.
To the best of the authors’ knowledge, there is no existing literature review that focuses on ML-based intelligent mobile apps. Thus, the goal of this work is to have an overview of research studies done in the area of the design and the development of mobile applications that provide intelligent interactions with mobile users using machine learning techniques. Mainly, we want to discover the level of maturity of works concerning this research axis, the frequency of publications, the most targeted application fields, the most used machine learning techniques, and for which purposes they were used. To this end, we carry out a Systematic Mapping Study (SMS) as a review approach that structures available researches and results in a particular research area [15, 16]. Therefore, the identified works will be classified and categorized according to some research questions, which will frame the literature study to be carried out. The results are often presented visually to facilitate their interpretation [15, 16].
We, therefore, selected and studied 71 papers published in the period from 1 January 2007 to 31 December 2019. The objective of this SMS is to make classification and contributions counting of selected papers. Particularly, we need to have a clear vision about (1) the most frequently used ML techniques in the development of intelligent mobile apps, (2) the application domains that have been covered in the literature, (3) where the literature has been published, and (4) which kind of mobile user data are collected and analyzed.
The rest of this paper is organized as follows. Section 2 presents the SMS methodology and different rules on which we are basing our study to perform this work. The results of the study are then described and discussed in Section 3, while Section 4 presents some implications for researchers and practitioners. The limitations of the study are then presented in Section 5. Finally, Section 6 gives conclusions and future works.
2. Research Methodology
To have an overview of the state of research in a specific topic and to decide on the axes where to dig, a literature study is requested. However, researchers in the fields of Software Engineering (SE) often do not follow systematic approaches in carrying out their literature studies. Recently, several guidelines have been proposed to structure a literature study in the field of SE by applying either a Systematic Mapping Study (SMS) or a Systematic Literature Review (SLR) [15–20]. As described in [15, 16] an SMS offers a superficial overview of a particular topic by providing a count and classification of research works published in this topic. This is often done visually using graphics. An SMS will help in structuring the research topic in question and will also allow better conduct of subsequent research work as SLR [15, 16]. Compared to SMS and as described in [17–20] SLR allows further literature study by performing an in-depth analysis of identified works. Therefore, an SLR is an SMS that includes some additional steps; for instance, it reviews the adopted methodology in each work and assesses the obtained results [15, 16].
Figure 1 shows the SMS process as it was described in [15, 16]. The process consists of five steps: (1) defining the Research Questions (RQ) of the study; (2) extracting keywords and defining the search string of the study; (3) extracting the relevant researches through scientific digital databases; (4) selection of relevant papers for the study; (6) data extraction from selected papers.
Description of the systematic mapping study process. The process starts with the definition of research questions and ends with the data extraction operation.
The RQ defines the purpose of the study that aims at structuring the body of knowledge related to a specific topic. In this direction, each research question must have a clear objective that is related to the subject of the study and must specify explicitly which data will be extracted from the selected papers [15, 16].
To identify keywords and formulate search strings from the research questions, SMS uses the PICO (Population, Interventions, Comparison, and Outcomes) model, which consists of four key components: population, intervention, comparison, and outcomes. Population refers to a specific SE role, type or application area. Intervention refers to software technology or methodology that addresses specific issues in SE. Comparison identifies the technologies, techniques, tools, methods, or strategies to be extracted and compared. Finally, outcomes should be related to the factors of the importance of the intervention to practitioners. The terms used to describe each component are then used to define the search string of the study [15, 16].
The defined search strings will be used to extract the relevant researches covering the subject of the study; this is mainly done through Scientific Digital Databases (SDDB). However, the search string must be adapted to the search roles in each SDDB [15, 16].
Inclusion and exclusion criteria must be used to retain only the researches that match the subject of the study; these criteria will be applied to the titles and abstracts of extracted researches. Nevertheless, we can have some researches that need a full review of the content before deciding about their inclusion or exclusion [15, 16].
Finally, the retained researches must be exploited to extract the data that responds to the specified research questions [15, 16].
2.1. Research Questions
The aim of this study is to have an overview of published works about the use of ML techniques in the realization of intelligent mobile apps. For that, we define six RQ that are presented in Table 1 and that cover the scope of developing intelligent mobile apps. Each research question is accompanied by an explanation that presents the rationale behind its adoption by the authors.
Definition and description of the research questions of the study.
ID
Research question
Rationale
RQ1
In which years, sources, and publication channels papers were published?
To identify where studies concerning this field of research can be found and whether there are specific publication channels; it also indicates when effort regarding this research area was made
RQ2
Which research types are adopted in selected papers?
To provide an overview of different types of research outlined in the literature regarding the application of ML in intelligent mobile apps development
RQ3
Which application fields are targeted in selected papers?
To determine in which application fields ML techniques were carried out and researchers were interested
RQ4
Which contexts are targeted in selected papers?
To determine in which context different studies were performed
RQ5
Which kinds of data are collected from mobile devices?
To identify which data are interesting researches to perform data analysis with ML techniques
RQ6
Which ML models, tasks, and techniques are used to analyze mobile data?
To determine which kind of ML techniques researchers were often interested in
2.2. Search Strings
To find the keywords of the study we follow the PICO model presented in [16]. As shown in Table 2, for each component we have a number of sentences that describe the subject of study. To extract keywords from these sentences we used roles proposed in [16] that classify keywords in many sets; each one is concerned with one component of PICO model. Table 3 presents the extracted keywords from PICO components.
Description of our PICO components that matches the intelligent mobile interaction.
Component
Description
Population
Intelligent mobile interaction studies
Interventions
(i) User-oriented mobile application (interface) design and development(ii) User experience oriented mobile application design and development(iii) Adaptive, reactive, or responsive mobile application design and development
Comparison
We compare studies on intelligent mobile interaction area by identifying and comparing the different approaches and techniques used to collect mobile user data and behavior, to analyze and learn from mobile user data and behaviors, or to predict user desires
Outcomes
Maximize user satisfaction in interaction with mobile applications
The sets of the extracted keywords from the PICO components.
Set
Component
Keywords
1
Population
Intelligent, mobile, interaction
2
Interventions
User, application, interface, design, development, user experience, adaptive, reactive, responsive
3
Comparison
Approaches, techniques, mobile user, data, behavior, analyze, learn, predict, desire
4
Outcomes
User, satisfaction, interaction, application
After defining the keywords sets, we reorganize them to six groups as presented in Table 4, with each group containing keywords that either are synonyms, present different forms of the same word, or are terms that have similar or related semantic meaning within the domain of intelligent mobile applications [18]. The definitive search string is obtained by concatenating words from groups as follows: (design OR development) AND (mobile OR user OR “mobile user”) AND (intelligent OR adaptive OR reactive OR responsive) AND (application OR interface) AND (analyze OR learn OR predict) AND (interaction OR “user experience” OR data OR behavior OR desire OR satisfaction). The Boolean OR is used to assemble terms in the same group and the Boolean AND is used to join groups of terms.
Grouping of identified terms in six groups depending on whether they are similar or they are often used together.
Terms
G1
G2
G3
G4
G5
G6
Intelligent
∗
Mobile
∗
Interaction
∗
User
∗
Application
∗
Interface
∗
Design
∗
Development
∗
User experience
∗
Adaptive
∗
Reactive
∗
Responsive
∗
Mobile user
∗
Data
∗
Behavior
∗
Analyze
∗
Learn
∗
Predict
∗
Desire
∗
Satisfaction
∗
2.3. Candidate Papers
The defined search string was used to search primary studies from four digital libraries: IEEE Xplore, ACM Digital Library, ScienceDirect, and Springer Link, since they were the most commonly used to publish SE studies [21].
2.4. Paper Selection
This section identifies the inclusion/exclusion (IC/EC) criteria we used to assess the relevance of the primary studies retrieved by applying the search string to the four digital libraries. IC/EC criteria were applied to the title and abstract of each paper. Doubtfully, we use the full-text to decide upon a paper whether it will be included or excluded. As presented in Table 5, we identified three inclusion criteria and four exclusion criteria:
Inclusion and exclusion criteria.
Category
Criteria
Inclusion
Studies presenting methods and techniques to develop intelligent mobile applications that support intelligent interactions with users
Studies presenting methods and techniques to analyze mobile user data
Studies published between 2007 and 2018
Exclusion
Studies not presented in English
Studies not accessible in full-text
Books and gray literature
Studies that are duplicates
2.5. Data Extraction
To collect all relevant data from the selected studies and in order to provide answers to the different RQs defined in Table 1 we used the template presented in Table 6. It provides a description of different data items that will be extracted from the selected papers. Each data item is provided with its name, with its data type, and also with the research questions to which they refer.
Data extraction template.
Data item
Name
Value
RQ
1
ID
Integer
—
2
Authors
Set of names of the authors
RQ1
3
Title
Title of the paper
RQ1
4
Publication channel
Kind of publication channel
RQ1
5
Publication source
Name of publication source
RQ1
6
Year
Calendar year
RQ1
7
Research type
Which research strategy was followed
RQ2
8
Study field
In which business fields the research was applied
RQ3
9
Study context
In which context the research was conducted
RQ4
10
Data collection strategy
Which kinds of data were collected from mobile users
RQ5
11
ML model
Which machine learning approach was adopted
RQ6
12
ML task
Which data mining tasks were used by selected papers
RQ6
13
ML technique
Which data mining techniques were used by selected papers
RQ6
Publication Channel refers to a regularly published journal or to an academic event such as conference, workshop, symposium, or seminar.
Publication Source refers to the effective name of the journal or the academic events that have published each selected paper.
Research type refers to a classification of selected studies in relation to the stage of completeness of the realized work. Petersen et al. in [15, 16] and Wieringa et al. in [22] define five types of research studies as follows: Evaluation Research (ER), Solution proposal (SP), Validation Research (VR), Philosophical Papers (PP), and Opinion Papers (OP). SP paper presents a novel technique or an improvement of an existing technique and argues for its relevance without a validation. VR makes an In-depth analysis of a note implemented solution proposal. ER provides an empirical study of a problem or an implementation of a technique in practice. PP presents a new way of looking at things or a new conceptual framework. Finally, an OP gives an opinion about what is wrong or good about something, how we should do something [15, 16].
Study Field refers to the application business field of the study such as e-Learning, e-Commerce, e-Government, and e-Health. A study that does not specify the application domain is classified as Generic.
Study Context refers to the context in which the study was carried out. In [17], Kitchman and Charters have defined four contexts: Academic, Organization, Industrial, and Government.
ML Model refers to the model of learning that is used to extract knowledge from data. In this work, we consider five types of ML models: Supervised Learning (SL), Unsupervised Learning (UL), Semi-Supervised Learning (SSL), Active Learning (AL), and Reinforcement Learning (RL) [23, 24]. SL algorithms learn from training data sets that provide examples about relations between data inputs and the target outputs and generalize results as a model that can predict outputs from new inputs [23, 24]. UL algorithms attempt to identify similarities between inputs, which help in their categorization [23, 24]. Learning algorithm attempts to describe data instead of SL algorithms that attempt to make a prediction of outputs. SSL algorithms are a subcategory of SL algorithms that are used when unlabeled data are easily available and it is difficult or expensive to have labeled data. The learning algorithm attempts to use labeled and unlabeled data when searching for a model (with less human intervention). It recursively attempts to find a model using labeled data and apply it on unlabeled data, then only detected labels with high accuracy are retained and added to the labeled data set [25]. AL algorithms are a subcategory of SL that are used in situations where there are huge data sets in which there are few labeled data for training. In such case label prediction is difficult or expensive to obtain. The learning algorithm attempts to add new labeled tuples to training data set by querying periodically a user for labels until having acceptable data set size that allows making supervised learning [23, 26]. RL algorithms are used in case of dynamic environments in which an agent tries to find the suitable action to execute in response to a specific situation or event. The agent learns behavior through his errors by trying different possibilities until he finds a suitable action to perform [23, 27].
ML techniques refer to the algorithms that are used to perform the learning task. These algorithms can be categorized into many ML Tasks for each of the ML models [23, 24]. For instance, SL model contains classification task and regression Task, and UL model contains association task and clustering task. SSL and AL models are considered as variants of the SL model; consequently, they use the same ML tasks [23, 24]. Figure 2 presents an inventory of ML techniques most used in practice and grouped by ML models and ML tasks.
Data analysis techniques grouped by ML models and tasks.
3. Results and Discussion
This section presents the findings related to this systematic map. Firstly, we introduce an overview of the result of the selection process; and secondly, all the results for each research question are presented.
3.1. Overview of the Selected Studies
Our search in the fourth digital libraries provided 9238 candidate papers. However, 9167 papers were excluded after applying the exclusion criteria, thus leading to the identification of 71 articles regarding the use of ML techniques in intelligent mobile app design and development.
Table 7 shows the number of selected papers in each step of the selection process. In fact, the duplicate papers were first discarded and only one study was considered. Besides, the papers reporting the same study were also excluded and only the most recent one was included. Then, all papers not written in English or not accessible in full-text were excluded. Finally, many papers required a full-text review before deciding about their inclusion or exclusion.
The number of retained papers after each step of our selection process.
Database
Returned studies
Year filter (2007–2018)
Title/Abstract review
Full-text review
Add manually
Retained studies
IEEE
1694
1356
36
19
5
24
ACM
3223
2475
36
16
5
21
SCDirect
560
500
16
2
3
5
Springer
3761
3229
47
16
5
21
The number of selected papers is respectable and reflects the importance of the research topic addressed in this study. Also, this number will allow conducting the study with acceptable data size that will give a credible overview of the targeted research topic. The complete list of selected studies with their relevant data is provided in Table 8.
Data extracted from the selected papers.
ID
RQ1
RQ2
RQ3
RQ4
RQ5
RQ6
Ref
Publication channel
Year
Research type
Study context
Study field
Data collection strategy
ML model
ML tasks
ML techniques
1
[28]
Conference
2019
VS
Academic
e-Learning
Context
SL
Classification
ANN
2
[29]
Journal
2019
VS
Academic
e-Health
Context, interactions, profile
SL
Classification
ANN
3
[30]
Conference
2019
SP
Academic
e-Health
Context, interactions, profile
SL
Classification
ANN
4
[31]
Conference
2019
ES
Recommendation
Context
UL
Clustering
K-means
5
[32]
Conference
2019
VS
Academic
e-Learning
Interactions
UL
ANN
6
[33]
Conference
2019
ES
Academic
Recommendation
Interactions
RL
Q-learning.
7
[34]
Journal
2019
ES
Academic
e-Learning
Context, interactions
SL + UL
Clustering, classification
K-means, SVM
8
[35]
Journal
2019
SP
Academic
e-Learning
Context, interactions
UL
Association, clustering
AR, K-means
9
[36]
Journal
2019
ES
Academic
e-Coaching
Context
SL
Classification
MM
10
[37]
Conference
2019
SP
Academic
e-Commerce
Context, interactions
SL
Classification
RBC
11
[38]
Journal
2019
ES
Academic
e-Learning
Context, interactions
SL
Classification
ANN, SVM
12
[39]
Conference
2018
VS
Government
Generic
Interactions, preferences, profile
RL
—
Q-learning
13
[40]
Conference
2018
ES
Academic
e-Learning
Interactions
SL
Classification
ANN
14
[41]
Conference
2018
SP
Government
Entertainment
Interactions, preferences
SL
Classification
RF
15
[42]
Journal
2018
VS
Organization
e-Health
Context
SL
Classification
RBC
16
[43]
Conference
2018
ES
Organization
Recommendation
Interactions
SL
Regression
KNN, NBC
17
[44]
Journal
2018
ES
Industrial
e-Health
Context, profile
SL
Classification
RBC
18
[45]
Conference
2017
SP
Organization
e-Health
Context
SL
Classification
DT
19
[46]
Conference
2017
ES
Academic
Gamming
Context
SL
Classification
RBC
20
[47]
Conference
2017
SP
Academic
e-Coaching
Feedbacks
SL
Classification, regression
DT, LR
21
[48]
Conference
2017
ES
Academic
fs management
Interactions
MCDM
22
[49]
Conference
2017
SP
Academic
Generic
Context, feedbacks, profile
SL
Regression
SVM
23
[50]
Journal
2017
PP
Organization
Generic
Context
SL
Classification
24
[51]
Journal
2017
ES
Organization
e-Coaching
Context, interactions
SL
Classification, clustering
DT, HAC
25
[52]
Journal
2017
ES
Organization
e-Coaching
Interactions
26
[53]
Conference
2016
SP
Organization
Generic
Context, interactions
27
[54]
Conference
2016
SP
Organization
e-Coaching
Context
SL
Classification
SVM
28
[55]
Conference
2016
ES
Organization
Recommendation
Context, interactions
SL
Association
AR
29
[56]
Symposium
2016
SP
Academic
Generic
Context, interactions
SL
Classification, clustering
ANN, K-means
30
[57]
Conference
2016
SP
Academic
Generic
Interactions
31
[58]
Journal
2016
ES
Organization
e-Learning
Context, profile
32
[59]
Conference
2015
ES
Organization
Notification management
Context
SL
Classification
NBC, RF
33
[60]
Conference
2015
SP
Academic
Gamming
Emotion
Classification
34
[61]
Conference
2015
ES
Academic
Natural language processing
Context, interactions
SL
Regression
LR
35
[62]
Conference
2015
VS
Academic
Recommendation
Context
SL
Classification
SVM
36
[63]
Journal
2015
ES
Academic
Gamming
Interactions
SL
Clustering
K-means
37
[64]
Journal
2015
ES
Academic
e-Learning
Context
SL
Classification
RBC
38
[65]
Journal
2014
ES
Government
Entertainment
Context, preferences
SL
Classification
Tp, CBFM
39
[66]
Journal
2014
VS
Academic
e-Health
Context
SL
Classification
RBC
40
[67]
Conference
2014
PP
Academic
Data visualization
Context
41
[68]
Conference
2014
ES
Organization
Entertainment
Context, interactions
SL
Classification
DT, KNN
42
[69]
Journal
2014
ES
Organization
e-Health
Context
SL
Classification
RBC
43
[70]
Journal
2014
PP
Government
Generic
Context
SL
Classification
RBC
44
[71]
Workshop
2014
ES
Academic
e-Commerce
Preferences, feedbacks
AL
Regression
UR
45
[72]
Conference
2013
VS
Academic
Generic
Context, interactions
SL
Classification
RBC
46
[73]
Conference
2013
ES
Academic
Generic
Interactions
SL
Clustering
K-means
47
[74]
Conference
2013
SP
Academic
e-Learning
Context, interactions
SL
Association
AR
48
[75]
Conference
2013
VS
Government
Gamming
Interactions
SL
Classification
ANN
49
[76]
Journal
2013
ES
Government
Generic
Preferences
SL
Classification
KNN
50
[77]
Journal
2013
ES
Academic
e-Health
Context
SL
Classification
RBC
51
[78]
Conference
2012
ES
Government
e-Health
Context
SL
Regression
FBA
52
[79]
Journal
2012
SP
Academic
e-Learning
Profile, interactions
SL
Classification
ANN, FL
53
[80]
Conference
2012
SP
Government
e-Learning
Context, preferences, profile
SL
Classification
DT
54
[81]
Conference
2012
VS
Academic
e-Learning
Preferences, profile
SL
Clustering
K-means
55
[82]
Journal
2012
VS
Organization
Smart home
Feedback, interactions
RL
Q-learning
56
[83]
Journal
2012
SP
Organization
e-Travel
Context, preferences
SL
Classification
SVM
57
[84]
Journal
2011
VS
Academic
Recommendation
Context, interactions
SL
Classification
NBC
58
[85]
Conference
2011
VS
Academic
Gamming
Interactions
SL
Classification
CN
59
[86]
Conference
2011
VS
Academic
e-Learning
Context, profile
SL
Classification
ANN, FL
60
[87]
Conference
2011
ES
Academic
e-Learning
Profile
61
[88]
Journal
2011
ES
Government
e-Learning
Preferences
SL + UL
Clustering, classification
DT
62
[89]
Conference
2010
ES
Academic
e-Travel
Context, interactions
UL
Clustering
FBA
63
[90]
Conference
2010
SP
Academic
Profile
SL + UL
Clustering, classification
ANN, K-means
64
[91]
Conference
2010
SP
Government
e-Health
Context
SL
Classification
ANN, FL
65
[92]
Journal
2009
ES
Government
e-Learning
Context, interactions, profile
SL
Classification, regression
RBC, MM
66
[93]
Conference
2009
ES
Academic
Generic
Context
SL
Classification
NBC
67
[94]
Journal
2009
PP
Academic
e-Learning
Context, profile
SL
Classification
68
[95]
Journal
2009
VS
Government
Generic
Interactions
UL
Clustering
FBA
69
[96]
Conference
2008
VS
Organization
E-commerce
Interactions
SL
Association
FBA
70
[97]
Journal
2008
VS
Industrial
Generic
Interactions
SL
Regression
SPA
71
[98]
Conference
2007
VS
Academic
Recommendation
Profile
SL
Classification
NBC, FL, MM
3.2. RQ1: In Which Years, Sources, and Publication Channels Papers Were Published?
Figure 3 presents the variation in the number of selected papers over the years between 2007 and 2019. It shows that the average annual growth rate of publications is 30% for each year. Nearly, two-thirds (66.2%) of the studies were published in the second half of the observed period between 2013 and 2019. Moreover, 11 papers were published in 2019, which is a significant number if it is compared to other years. Consequently, it is very clear that researchers are becoming more and more interested in this research area. This trend is justified by many other works. For instance, Garousi et al. have conducted a research about trends in SE; their work showed that mobile subject is the second hottest research topic in SE area [99]. Also, Karanatsiou et al. have affirmed that mobile app development is counted among the most frequent research topics in SE. Finally, Zhu et al. affirm that the improvement of mobile performances in terms of processing power and data storage capacity in addition to the advances in cloud computing has a role in the support machine learning algorithms by the mobile devices. They also highlight new challenges in exploiting massive data distributed over a large number of edge devices including smartphones [100].
Distribution of selected papers over the period 2007–2018.
Figure 3 shows also that almost every year the number of papers published in conferences is more important than the other channels. Moreover, Figure 4 shows that 54% of selected papers were published in conferences, while 40% were published in journals. Finally, only 4% were published in symposium and 2% were published in workshops. The high percentage of conference papers is explained by the fact that researchers often report their primary results at first in conferences, then synthesize them in a journal article after validation [101, 102]. Therefore, researches on the topic of intelligent mobile applications have not reached yet the required level of maturity and they still have several axes where to fetch. Also, the percentage of papers published in journals shows that there are some validated results that can be exploited in future works.
The percentage of each publication channel in selected papers.
Table 9 presents the publication sources where there were published at least two of the selected studies. Thus, the most frequent publication source is “Lecture Notes in Computer Science” with a percentage of 16.9%, succeeded by “ACM International Joint Conference on Pervasive and Ubiquitous Computing” with 4.23%. Finally, we find “IEEE Transactions on Learning Technologies,” “International Conference on Advanced Learning Technologies,” “Personal and Ubiquitous Computing,” “Procedia Computer Science,” and “Universal Access in the Information Society” with 2.81% for each of them. All these publication sources are considered to have a high-ranking, thing that reflects the high quality of most of the selected studies.
Most frequent publication sources in the selected studies.
Title
Type
Libraries
Ranking
Number
ACM international joint conference on pervasive and ubiquitous computing
Conference
ACM
A∗
3
IEEE transactions on learning technologies
Journal
IEEE
Q1
2
International conference on advanced learning technologies
Conference
IEEE
B
2
Lecture notes in computer science
Journal
Springer
Q2
12
Personal and ubiquitous computing
Journal
Springer
Q2
2
Procedia computer science
Journal
Elsevier
-
2
Universal access in the information society
Journal
Springer
Q2
2
3.3. RQ2: Which Research Types Are Adopted in Selected Papers?
Figure 5 shows that ER (Evaluation Research) is the most adopted research type with a percentage of 44%, succeeded by SP (Solution Proposal) with a percentage of 26% and VR (Validation Research) with a percentage of 24%. Finally, PP (philosophical paper) is the less adopted research type with 6%.
The percentage of each research type in selected papers.
ER are more dominant, and this means that several studies have proposed final solutions that are evaluated and experimented in practice. These studies can be reviewed in-depth in future works as SLR. Like ER, VR are also pertinent and can be considered in future SLR studies. Other research types must be ignored in subsequent work [14]. The codominance of SP, VR, and PP to the detriment of ER reflects that the majority of proposed solutions are not implemented or experimented in real context. SP works propose implemented solutions that are not yet experimented. However, VR propose new implemented and experienced solutions; nevertheless, experiments are realized as simulations, prototyping, or experiments in a lab. Finally, PP works present a new conceptual framework solution that is not yet either implemented or experienced [16, 17]. ER works are already experimented and validated; that is why they are published equally in journals and conferences as shown in Figure 6. Figure 6 shows also that SP and VR are often published in conferences. This is because these works are firstly reported as primary results in conferences, then they will be synthesized and reported in journals after validation [101, 102].
The percentage of research types in journals and conferences.
3.4. RQ3: Which Application Fields Are Targeted in Selected Papers?
Figure 7 shows that every year the total number of papers that focus on a specific field is greater than those that are generic. Moreover, only 22% of selected papers have proposed generic solutions that can be applied on several fields and 78% are oriented to a specific field. Therefore, we note that researchers are focusing more and more on specific fields. Figure 8 shows the variation in application fields adopted in selected papers. It shows that the most targeted application fields are e-Learning with 24% and e-Health with 17%.
Distribution of selected papers over the period 2007–2018 for each application field category.
The number of selected studies by application field.
The dominance of e-Learning and e-Health mobile applications can be explained by the fact that they have a massive social and economic impact on consumers and enterprises than other fields. They create better living conditions [103]. Mobile applications for e-Learning provide the possibility to access high-quality educational resources and educational content that is adapted to users’ needs and learning speed. Mobile applications for e-Health represent a creative solution to bring health care services to more people especially in emerging countries [103].
3.5. RQ4: Which Contexts Are Targeted in Selected Papers?
Figure 9 shows that 47% of selected studies were conducted in an academic context, 35% were funded by organizations, 17% were adopted by government institutes, and only 1% of selected studies were conducted in an industrial context. Figure 10 shows that during the last 4 years all studies were conducted either in an academic context or in an organization context, except two studies that were conducted in government context. Figure 10 shows also that studies that were conducted in an organization context begin to be more frequent than academic studies. Moreover, in the last for years, works funded by organizations are becoming more dominant.
The percentage of each study context in selected papers.
Distribution of selected papers over the period 2007–2018 for each research context.
The growing number of projects funded by organizations can be explained by the fact that organizations (companies, enterprises) are becoming aware of the importance of mobile technologies as innovative solutions for the realization of economic and human development, so they fund many research and development projects in the area of mobile technologies [103]. Industrials are more interested in hardware and infrastructure aspects instead of the software engineering aspect, which explains the few number of research studies funded by industrials. Finally, regarding the number of works funded by governments, it can be explained by the fact that governments spending on adoption and usage of mobile technologies do not reach the expected economic and social aspirations [103].
3.6. RQ5: Which Kinds of Data Are Collected from Mobile Devices?
Figure 11 shows that context data are the most collected from mobile users with a percentage of 42%, succeeded by interaction data with 33%, profile data with 13%, preference data with 8%, and feedback data with 4%. But it must be emphasized that, in the same work, several types of data may be collected from the mobile user.
The percentage of each collected data type in selected papers.
Context and interaction data are the most used because they provide dynamic data about the mobile user. They can help in taking decisions that are most suitable to the current context of the user in conjunction with his previous actions. Context data give information about the current state of the user (identity, health state, mental state, physical state, psychological state, etc.) and about its environment (location, time, current activity, etc.). However, interaction data gives information about the previous user actions (apps consultation history, notification history, usage log, clicks, etc.) that can help in predicting future user actions. Profile and preference data often provide static information about the user and help to have a primary idea about the user without considering context and interaction data that are not available at the beginning of use of mobile apps. Finally, feedback data are not very important in conducting interactions with users because usually they are considered in application maintenance.
3.7. RQ6: Which Machine Learning Models, Data Mining Tasks, and Techniques Are Used to Analyze Mobile Data?
Figure 12 shows that supervised learning is the most used machine-learning model in selected papers with a percentage of 73%; however, 11% have used unsupervised learning, 4% have used reinforcement learning, and 1% has used active learning. Finally, 11% of papers have not used any machine-learning model. Figure 13 shows that every year, SL is the most dominant model in selected studies. Also, it shows that there was no usage of UL techniques and there were very modest attempts to introduce UL, RL, and AL techniques in intelligent mobile apps development.
The percentage of each machine learning model in selected papers.
Evolution of Machine Learning Models usage in selected papers between 2007 and 2018.
As the majority of selected studies have used supervised learning models, Figure 14 shows that classification is the most used data mining tasks with a percentage of 66%, succeeded by clustering with a percentage of 16% and regression with a percentage of 13%. Finally, the association task comes in the last range with 5%.
The percentage of adopted data analysis tasks in selected papers.
Table 10 presents data mining techniques that are used more than one time in selected papers. Thus, the most frequent technique is ANN that was used in 12 studies, followed by RBC technique that was used in 11 studies, K-means that was used in 8 studies, DT, SVM that were used in 6 studies for each of them, NBC that was used in 5 studies, FL and FBA that were used in 4 studies for each of them, KNN, MM, Q-learning, and AR that were used in 3 studies for each of them, and finally RF and LR that were used in 2 studies for each of them.
Data mining techniques that are used by more than one of the selected papers.
Techniques
Number of papers
ANN (artificial neural network)
12
RBC (rules-based classifier)
11
K-means
8
DT (decision tree)
6
SVM (support vector machine)
6
NBC (naïve Bayesian classifier)
5
FL (fuzzy logic)
4
FBA (frequency-based algorithm)
4
KNN (K-nearest neighbour)
3
MM (Markov models)
3
Q-learning
3
AR (association rules)
3
RF (random forest)
2
LR (logistic regression)
2
We conclude that the use of SL models (and specifically classification tasks) is a natural phenomenon because several works attempt in the majority of cases to classify users in order to know how to interact with them according to their profiles, preferences, contexts, or actions. In addition, SL algorithms are known by their simplicity because they attempt to learn from the training dataset to find model that makes conjunction between inputs and desired outputs; also, both the input and desired output data are known in advance; therefore, manual interventions to correct results are minimal [23, 24]. Finally, many SL services are now available on the cloud to facilitate the integration ML algorithm in mobile apps [27].
In many cases, UL tasks were used in the data preparation stage to make data exploration or data dimension reduction. Often, UL tasks are used before resorting to SL algorithms for prediction tasks; also, it needs more interventions than SL tasks to correct obtained results [23, 24].
RL algorithms were used in mobiles to make automation of some tasks by finding a suitable action for a specific situation or event. They were used especially for energy consumption prediction, memory allocation prediction, and mobile app usage prediction (predict users actions in interaction with mobile apps). They were less used because of their complexity by comparing them with SL and UL [23, 27].
AL tasks are the less used in mobile apps, because of their use in situations where there are huge data sets in which there are few labeled data for training. In such case label prediction is difficult or expensive to obtain [23, 26].
4. Implications for Researchers and Practitioners
This section presents some implications and recommendations for researchers and practitioners that are deduced from the analysis of data extracted from selected papers.
4.1. RQ1
Despite the important number of returned papers by the defined search string, there were a significant number of excluded papers at each level of the selection process. This is mainly due to the fact that many research axes share the same vocabulary with mobile apps development as mobile networking, vehicle human interface interaction, and robot programming. So, researchers must define clearly the purpose of their works with terms that belong to the software engineering vocabulary. Also, many excluded papers are not described enough and do not provide sufficient details that can help in the facilitation of the selection process. So, researchers must make more effort in synthesizing their works by ensuring an acceptable level of scientific credibility that will help in evaluating correctly the quality of the work.
4.2. RQ2 and RQ3
Researchers and developers working on the development of intelligent mobile applications are focusing more and more on specific fields as e-Learning and e-Health, and this is at the expense of generic solutions. This orientation can be explained by the fact that each field has its specificities and constraints and that it is more difficult to validate more generic solutions than those that focus on specific fields. Thus, the majority of generic solutions in selected papers are not validated. So, it is recommended that future works take also this orientation of targeting a specific application field.
4.3. RQ4
The majority of selected studies are done in academic and organizational context; therefore researchers and developers still have to exert more efforts to make validation and standardization of proposed solutions, things that can encourage industrial and governments to adopt and finance more research in the area of intelligent mobile application.
4.4. RQ5 and RQ6
For the majority of selected studies, choosing a ML technique over another is not always justified. Furthermore, many works use techniques that are easy to implement but not necessarily efficient. For instance, RBC that is the second most used technique is known by its ease of implementation and interpretation but it is not the best in terms of quality [23, 24]. So, it is strongly recommended for researchers to make more effort in making proof of their choice in terms of ML techniques. On the other hand, all studies do not express clearly the nature of the logic adopted in applications. For instance, a purely user-oriented application changes its logic in conjunction with each user separately from others. Cornerwise, lightweight user-oriented application changes its logic to make global adaptations without distinction between users. So, the implemented intelligence in the two cases is not the same. Firstly, the data size is not the same; secondly, the nature of collected data is not the same (dynamic and real-time data in the first case and static and non-real-time data in the second case); thirdly the ML tasks and techniques are not the same; and finally, the degree of real-time adaptation is not the same. So, it is recommended for researchers to determine the nature of adopted logic in their solutions to have a clear idea about their contributions and to have a correct evolution of their works.
5. Limitations of the Study
Many research axes share the same vocabulary with mobile apps development as mobile networking, vehicle human interface interaction, and robot programming. So, it was very difficult to make correct keywords based on the inclusion/exclusion of papers returned by the defined search string. Therefore, it is possible that many papers have been excluded due to the lack of credible keywords or precise description. Also, given the huge number of returned papers by indexed databases, we restricted our search to only the most credible databases [19].
6. Conclusion and Future Studies
ML provides promising techniques to extract knowledge from a large dataset; the objective is to obtain patterns and models that will help in developing intelligent applications that will learn from data collected about user context, profile, and interactions. Several studies have attempted to integrate ML in the mobile field in order to provide intelligent mobile applications. Nevertheless, we do not have an overview of which techniques are used and how they are exploited in mobile apps design and development.
The aim of this paper was to define the state of research on the topic of ML techniques applied in mobile apps design and development. For this purpose, we have performed a systematic mapping study that aims primarily to respond to the research questions presented in Table 1. A total of 60 studies were selected and analyzed according to the following criteria: year, sources and publication channel, research type and methods, kind of collected data, and finally adopted ML models, tasks, and techniques.
The obtained results show that the average annual growth rate of papers publication is 25%. These papers were published in different journals and appeared at several conferences regarding computer science and software engineering fields. Evaluation search is the more dominant research type. e-Learning and e-Health are the most targeted application fields. The majority of studies are conducted in an academic context. Context and interaction data are the most collected data from the mobile user. Three-thirds of selected papers have used supervised learning models and more specifically classification tasks. Rules-Based Classifier is the most used ML technique.
Most of the studies are conducted in an academic context, a thing that reflects that either the topic is not attractive for governments and industrials or it is too difficult to apply the obtained result in real context. So, we recommend exerting more effort to make standardization and proposition tools that support ML technique application in mobile context.
Most of the studies depended closely on the application field in the sense that each domain has its own specificity in terms of the type of collected data and also in terms of its analysis objectives. Also, generic solutions cannot be evaluated if they are not applied on a specific field. Therefore, future studies must be oriented to a specific application field.
In the majority of cases, SL techniques are used to classify users in function of their context or/and profile. This is used to adapt the application behavior or interface to the user. But for most of the selected studies, the choice of a ML technique is not always justified. Therefore, researchers must make more attempts to introduce other ML methods and assess their performance in mobile context and specifically those that are more adapted to dynamic environments as mobile. For instance, RL techniques are known for their ability to learn from user’s actions and that can automatically make real-time decisions that attempt to maximize user satisfaction and all that without any training dataset.
Our future studies will be consecrated to the realization of a systematic literature review (SLR) that will make an in-depth analysis of all selected studies in addition to the studies published recently on the topic of intelligent mobile application design and development.
Conflicts of Interest
The authors declare that they have no conflicts of interest regarding the publication of this paper.
ClementJ.Combined global apple app store and google play app downloads from 1st quarter 2015 to 2nd quarter 2019 (in billions)2019https://www.statista.com/statistics/604343/number-of-apple-app-store-and-google-play-app-downloads-worldwide/ReinselD.GantzJ.RydningJ.2017Cupertino, CA, USASeagateDehlingerJ.DixonJ.Mobile application software engineering: challenges and research directionsProceedings of the 2011 Workshop on Mobile Software EngineeringOctober 2011Santa Monica, CA, USA2932WassermanT.Software engineering issues for mobile application developmentProceedings of the FSE/SDP Workshop on Future of Software Engineering ResearchNovember 2010Santa Fe, NM, USAKönig-RiesB.Challenges in mobile application development2009512697110.1524/itit.2009.9055HammershøjA.SapuppoA.TadayoniR.Challenges for mobile application developmentProceedings of the 2010 14th International Conference on Intelligence in Next Generation Networks2010Berlin, Germany18KiselevaJ.WilliamsK.HassanA.Predicting user satisfaction with intelligent assistantsProceedings of the 39th International ACM SIGIR Conference on Research and Development in Information Retrieval2016ACM4554KiselevaJ.WilliamsK.JiangJ.Understanding user satisfaction with intelligent assistantsProceedings of the 2016 ACM on Conference on Human Information Interaction and RetrievalJuly 2016Pisa, ItalyACM121130HanS. H.KimK. J.YunM. H.KimJ.Identifying mobile phone design features critical to user satisfaction2004141152910.1002/hfm.100512-s2.0-0346503027TaruteA.NikouS.GatautisR.Mobile application driven consumer engagement201734414515610.1016/j.tele.2017.01.0062-s2.0-85010685414IstepanianR. S. H.Al-AnziT.m-Health 2.0: new perspectives on mobile health, machine learning and big data analytics2018151344010.1016/j.ymeth.2018.05.0152-s2.0-85048786854DiazJ. C. T.MoroA. I.CarriónP. V. T.Mobile learning: perspectives20151213849PousttchiK.TilsonD.LyytinenK.HufenbachY.Introduction to the special issue on mobile commerce: mobile commerce research yesterday, today, tomorrow—what remains to be done?201519412010.1080/10864415.2015.10293512-s2.0-84944312229DengY.Deep learning on mobile devices: a review10993Proceedings of the Mobile Multimedia/Image Processing, Security, and Applications 2019April 2019Baltimore, MD, USAPetersenK.FeldtR.MujtabaS.MattssonM.Systematic mapping studies in software engineering12Proceedings of the 12th International Conference on Evaluation and Assessment in Software Engineering (EASE)June 2008Bari, Italy110PetersenK.VakkalankaS.KuzniarzL.Guidelines for conducting systematic mapping studies in software engineering: an update20156411810.1016/j.infsof.2015.03.0072-s2.0-84929464206KitchenhamB.ChartersS.2007Keele, UKKeele UniversityKofod-PetersenA.2012Cophenhagen, DenmarkAlexandra InstituteKitchenhamB.BreretonO. P.BudgenD.TurnerM.BaileyJ.LinkmanS.Systematic literature reviews in software engineering—a systematic literature review200951171510.1016/j.infsof.2008.09.0092-s2.0-56649086628KitchenhamB.200433Keele, UKKeele UniversityZhangH.BabarM. A.TellP.Identifying relevant studies in software engineering201153662563710.1016/j.infsof.2010.12.0102-s2.0-79953708792WieringaR.MaidenN.MeadN.RollandC.Requirements engineering paper classification and evaluation criteria: a proposal and a discussion200611110210710.1007/s00766-005-0021-62-s2.0-31044444123MarslandS.2011Boca Raton, FL, USAChapman and Hall/CRCHanJ.PeiJ.KamberM.2011Amsterdam, NetherlandsElsevierZhuX.JerryX.2005Madison, WisconsinUniversity of Wisconsin-Madison Department of Computer SciencesSettlesB.2009Madison, WisconsinUniversity of Wisconsin-Madison Department of Computer SciencesKaelblingL. P.LittmanM. L.MooreA. W.Reinforcement learning: a survey1996423728510.1613/jair.301CaoH.An intelligent speech interaction model for mobile teachingProceedings of the 2019 International Conference on Intelligent Transportation, Big Data and Smart City, ICITBS 20192019170173TheiligM.KorbelJ. J.MayerG.HoffmannC.ZarnekowR.Employing environmental data and machine learning to improve mobile health receptivity2019717982317984110.1109/access.2019.2958474YangZ.ReddyV. J.KesidiR.JinF.Addict free—a smart and connected relapse intervention mobile appProceedings of the 2019 ACM International Conference Proceeding SeriesAugust 2019Vienna, Austria202205GheraibiaM. Y.Gouin-VallerandC.Intelligent mobile-based recommender system framework for smart freight transportProceedings of the 2019 ACM International Conference Proceeding Series2019219222WuY.ZhangJ.DongQ.The use of SDAE in noisy English mispronunciation detection and diagnosis towards application in mobile learningProceedings of the 2019 ACM International Conference Proceeding Series2019176180ShenZ.YangK.DuW.ZhaoX.ZouJ.DeepAPP: a deep reinforcement learning framework for mobile application usage predictionSenSys 2019—Proceedings of the 17th Conference on Embedded Networked Sensor SystemsNovember 2019New York, NY, USA.153165LiuQ.HuangJ.WuL.ZhuK.BaS.CBET: design and evaluation of a domain-specific chatbot for mobile learning2019119Vélez-LangsO.Caicedo-CastroI.Recommender systems for an enhanced mobile e-learning201911786Berlin, GermanySpringerLockJ. C.TramontanoA. G.GhidoniS.BellottoN.ActiVis: mobile object detection and active guidance for people with visual impairments201911752Berlin, GermanySpringerAlikhademiK.AI-based technical approach for designing mobile decision aids2019103316316910.1007/978-3-030-23528-4_232-s2.0-85069670666CeliktutanO.DemirisY.Inferring human knowledgeability from eye gaze in mobile learning environments201911134Berlin, GermanySpringer193209HoB.-J.BalajiB.KoseogluM.SrivastavaM.Nurture: notifying users at the right time using reinforcement learningProceedings of the ACM International Joint Conference on Pervasive and Ubiquitous Computing2018Seattle, WA, USAMuradD.WangR.TurnbullD.WangY.SlionsProceedings of the ACM Multimedia Conference on Multimedia Conference2018Seoul, KoreaConstantinidesM.DowellJ.A framework for interaction-driven user modeling of mobile news reading behaviourProceedings of the Conference on User Modeling, Adaptation and Personalization 20182018Halifax, CanadaEspositoM.MinutoloA.MegnaR.ForastiereM.MagliuloM.De PietroG.A smart mobile, self-configuring, context-aware architecture for personal health monitoring20186713615610.1016/j.engappai.2017.09.0192-s2.0-85032855138PuC.WuZ.ChenH.XuK.CaoJ.A sequential recommendation for mobile apps: what will user click next app?Proceedings of the IEEE International Conference on Web Services 2018July 2018San Francisco, CA, USAHussainJ.HassanA. U.BilalH. S. M.Model-based adaptive user interface based on context and user experience evaluation201812111610.1007/s12193-018-0258-22-s2.0-85041911953FallahzadehR.MinorB.EvangelistaL. S.CookD. J.GhasemzadehH.Mobile sensing to improve medication adherenceProceedings of the 2017 ACM/IEEE International Conference on Information Processing in Sensor NetworksApril 2017Pittsburgh, PA, USAFrancilletteY.GouaichA.AbroukL.Adaptive gameplay for mobile gamingProceedings of the IEEE Conference on Computational Intelligence and GamesAugust 2017New York, NY, USAHeiY. T.KitN. C.NgV.An anti-drug mobile application with smart alerting for parentsProceedings of the International Conference on Applied System InnovationMay 2017Sapporo, JapanKabassiK.VirvouM.AlepisE.Reasoning about users actions in a mobile environment using a combination of HPR with MAUTProceedings of the International Conference on Information, Intelligence, Systems & Applications (IISA)August 2017Larnaca, CyprusAbusairM.User- and analysis-driven context aware software development in mobile computingProceedings of the 2017 Joint Meeting on Foundations of Software EngineeringSeptember 2017Paderborn, GermanyQinX.TanC. W.ClemmensenT.Context-awareness and mobile HCI: implications, challenges and opportunities2017Berlin, GermanySpringerSpanakisG.WeissG.BohB.LemmensL.RoefsA.Machine learning techniques in eating behavior e-coaching: balancing between generalization and personalization201721464565910.1007/s00779-017-1022-42-s2.0-85020463919GonçalvesV. P.de Almeida NerisV. P.SeraphiniS.Providing adaptive smartphone interfaces targeted at elderly people: an approach that takes into account diversity among the elderly201716112914910.1007/s10209-015-0429-92-s2.0-84944629165YangQ.ZimmermanJ.SteinfeldA.TomasicA.Planning adaptive mobile experiences when wireframingProceedings of the 2016 ACM Conference on Designing Interactive SystemsJune 2016Brisbane, AustraliaBarataF.KowatschT.TinschertP.FillerA.Personal MobileCoachProceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous ComputingSeptember 2016Heidelberg, GermanyHashemiM.HerbertJ.A next application prediction service using the BaranC frameworkProceedings of the Annual Computer Software and Applications Conference (COMPSAC)June 2016Atlanta, GA, USADekaB.HuangZ.KumarR.ERICA: interaction mining mobile appsProceedings of the Annual Symposium on User Interface Software and TechnologyOctober 2016Tokyo, JapanRaheelS.Improving the user experience using an intelligent adaptive user interface in mobile applicationsProceedings of the IEEE International Multidisciplinary Conference on Engineering Technology (IMCET)November 2016Beirut, Lebanon6468AbechM.da CostaC. A.BarbosaJ. L. V.RigoS. J.da Rosa RighiR.A model for learning objects adaptation in light of mobile and context-aware computing201620216718410.1007/s00779-016-0902-32-s2.0-84957039665MehrotraA.MusolesiM.HendleyR.PejovicV.Designing content-driven intelligent notification mechanisms for mobile applicationsProceedings of the ACM International Joint Conference on Pervasive and Ubiquitous ComputingSeptember 2015Osaka, JapanKatmadaA.ChatzakisM.ApostolidisH.MavridisA.PanagiotisS.An adaptive serious neuro-game using a mobile version of a bio-feedback deviceProceedings of the International Conference on Interactive Mobile Communication Technologies and LearningNovember 2015Thessaloniki, GreeceChenY.-N.SunM.RudnickyA. I.GershmanA.Leveraging behavioral patterns of mobile applications for personalized spoken language understandingProceedings of the 2015 International Conference on Multimodal InteractionNovember 2015Seattle, WA, USAKimS.HongE.ParkB.ParkH.A user customized service provider framework based on machine learningProceedings of the International Conference on Ubiquitous and Future NetworksJuly 2015Sapporo, JapanTorokL.PelegrinoM.TrevisanD. G.CluaE.MontenegroA.A mobile game controller adapted to the gameplay and user’s behavior using machine learning2015Berlin, GermanySpringerMoralesR.IglerB.BöhmS.ChitchaipokaP.Context-aware mobile language learning201556828710.1016/j.procs.2015.07.1982-s2.0-84939145749ZhuH.ChenE.XiongH.YuK.CaoH.TianJ.Mining mobile user preferences for personalized context-aware recommendation20145412710.1145/25325152-s2.0-84919608533MizouniR.MatarM. A.Al MahmoudZ.AlzahmiS.SalahA.A framework for context-aware self-adaptive mobile applications SPL201441167549756410.1016/j.eswa.2014.05.0492-s2.0-84904166766JovanovicM.StarcevicD.JovanovicZ.Bridging user context and design models to build adaptive user interfacesProceedings of the International Conference on Human-Centered Software EngineeringSeptember 2014Paderborn, GermanySpringer10.1007/978-3-662-44811-3_3SchedlM.BreitschopfG.IonescuB.Mobile music geniusProceedings of the 2014 International Conference on Multimedia Retrieval2014Dallas, TX, USAAlnanihR.OrmandjievaO.RadhakrishnanT.Empirical evaluation of intelligent mobile user interfaces in healthcare2014Berlin, GermanySpringerZimmermannG.VanderheidenG. C.StrobbeC.Towards deep adaptivity—a framework for the development of fully context-sensitive user interfaces2014Berlin, GermanySpringerLamcheB.Active learning strategies for exploratory mobile recommender systemsProceedings of the 2014 Workshop on Context-Awareness in Retrieval and RecommendationJanuary 2014Amsterdam, NetherlandsJainR.BoseJ.ArifT.Contextual adaptive user interface for android devicesProceedings of the 2013 Annual IEEE India ConferenceDecember 2013Mumbai, IndiaNivethikaM.VithiyaI.AnntharshikaS.DeegallaS.Personalized and adaptive user interface framework for mobile applicationProceedings of the 2013 International Conference on Advances in Computing, Communications and InformaticsAugust 2013Mysore, IndiaJianW.WangQ.ZhangX.FuX.ZhengX.Intelligent information processing and data mining in the application of mobile learningProceedings of the 2013 International Conference on Intelligent Networks and Intelligent SystemsNovember 2013Shenyang, ChinaBudihartoW.RachmawatiR. N.RickyM. Y.RumondorP. C. B.The psychological aspects and implementation of adaptive games for mobile applicationProceedings of the International Joint Conference on Awareness Science and Technology and Ubi-Media ComputingNovember 2013Aizu-Wakamatsu, JapanVildjiounaiteE.SchreiberD.KyllönenV.StänderM.NiskanenI.MäntyjärviJ.Prediction of interface preferences with a classifier selection approach20137432134910.1007/s12193-013-0127-y2-s2.0-84891714393AlnanihR.OrmandjievaO.RadhakrishnanT.Context-based and rule-based adaptation of mobile user interfaces in mHealth20132139039710.1016/j.procs.2013.09.0512-s2.0-84896933726ZhangJ.TangH.ChenD.ZhangQ.deStress: mobile and remote stress monitoring, alleviation, and management platformProceedings of the IEEE Global Telecommunications ConferenceDecember 2012Anaheim, CA, USAAl-HmouzA.ShenJ.Al-HmouzR.YanJ.Modeling and simulation of an adaptive neuro-fuzzy inference system (ANFIS) for mobile learning20125322623710.1109/tlt.2011.362-s2.0-84866111296GómezS.ZervasP.SampsonD. G.FabregatR.Delivering adaptive and context-aware educational scenarios via mobile devicesProceedings of the International Conference on Advanced Learning TechnologiesJuly 2012Rome, ItalyVirvouM.AlepisE.TroussasC.A mobile expert system for tutoring multiple languages using machine learningProceedings of the International Conference on E-Learning and E-Technologies in EducationSeptember 2012Lodz, PolandGilM.PelechanoV.Exploiting user feedback for adapting mobile interaction obtrusiveness2012Berlin, GermanySpringerLathiaN.CapraL.MagliocchettiD.Personalizing mobile travel information services2012481195120410.1016/j.sbspro.2012.06.1095LeeH.ChoiY. S.KimY.An adaptive user interface based on spatiotemporal structure learning201149611812410.1109/mcom.2011.57839962-s2.0-79958717915HarrisonB.RobertsD. L.Using sequential observations to model and predict player behaviorProceedings of the International Conference on Foundations of Digital Games—FDG ’11July 2011Bordeaux, France9198Al-HmouzA.ShenJ.YanJ.Al-HmouzR.Modeling mobile learning system using ANFISProceedings of the International Conference on Advanced Learning TechnologiesJuly 2011Athens, GA, USARazekM. A.BardesiH. J.Towards adaptive mobile learning systemProceedings of the International Conference on Hybrid Intelligent Systems (HIS)December 2011Malacca, MalaysiaSuJ. M.TsengS. S.LinH. Y.ChenC. H.A personalized learning content adaptation mechanism to meet diverse user needs in mobile learning environments2011211-254910.1007/s11257-010-9094-02-s2.0-79955813238WessonJ. L. J. J. L.SinghA.van TonderB.Van TonderB.van TonderB.Can adaptive interfaces improve the usability of mobile applications?2010Berlin, GermanySpringerPaireekrengW.WongK. W.Mobile content personalisation using intelligent user profile approachProceedings of the International Conference on Knowledge Discovery and Data Mining2010Washington, DC, USAKrejcarO.Adaptivity types in mobile user adaptive system framework2010Berlin, GermanySpringerMartinE.CarroR. M.Supporting the development of mobile adaptive learning environments: a case study200921233610.1109/tlt.2008.242-s2.0-77749264050KimY.ChoS.A recommendation agent for mobile phone users using bayesian behavior predictionProceedings of the International Conference on Mobile Ubiquitous Computing, Systems, Services and Technologies2009Sliema, MaltaAl-HmouzA.ShenJ.YanJ.A machine learning based framework for adaptive mobile learning2009Berlin, GermanySpringerGlavinicV.LjubicS.KukecM.On efficiency of adaptation algorithms for mobile interfaces navigation2009Berlin, GermanySpringerNurmiP.ForsblomA.FloréenP.PeltonenP.SaarikkoP.Predictive text input in a mobile shopping assistantProceedings of the International Conference on Intelligent User Interfaces2008Gran Canaria, SpainHartmannM.SchreiberD.Proactively adapting interfaces to individual users for mobile devices2008Berlin, GermanySpringerBonto-KaneM. V. A.Use of formal computational models for designing intelligent mobile device interfacesProceedings of the International Conference on Human Computer Interaction with Mobile Devices and Services2007Barcelona, SpainGarousiV.MäntyläM. V.Citations, research topics and active countries in software engineering: a bibliometrics study201619567710.1016/j.cosrev.2015.12.0022-s2.0-84959387510ZhuG.LiuD.DuY.YouC.ZhangJ.HuangK.Toward an intelligent edge: wireless communication meets machine learning2020581192510.1109/mcom.001.1900103ShawM.Writing good software engineering research papersProceedings of the 25th International Conference on Software Engineering2003Portland, OregonIEEE726736HappellB.From conference presentation to journal publication: a guide200815210.7748/nr2008.01.15.2.40.c63282-s2.0-41449093337BezerraJ.BockW.CandelonF.2015Boston, MA, USABoston Consulting Group