Context-aware recommender system for multi-user smart home

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 13, No. 3, June 2023, pp. 3192~3203
ISSN: 2088-8708, DOI: 10.11591/ijece.v13i3.pp3192-3203  3192
Journal homepage: http://ijece.iaescore.com
Context-aware recommender system for multi-user smart home
Shymaa Sobhy, Eman M. Mohamed, Arabi Keshk, Mahmoud Hussein
Department of Computer Science, Menoufia University, Menofia, Egypt
Article Info ABSTRACT
Article history:
Received Aug 4, 2022
Revised Sep 16, 2022
Accepted Oct 1, 2022
Smart home is one of the most important applications of the internet of
things (IoT). Smart home makes life simpler, easier to control, saves energy
based on user’s behavior and interaction with the home appliances. Many
existing approaches have designed a smart home system using data mining
algorithms. However, these approaches do not consider multiusers that exist
in the same location and time (which needs a complex control). They also
use centralized mining algorithm, then the system’s efficiency is reduced
when datasets increase. Therefore, in this paper, we firstly build a
context-aware recommender system that considers multi-user’s preferences
and solves their conflicts by using unsupervised algorithms to deliver useful
recommendation services. Secondly, we improve smart home’s responsive
using parallel computing. The results reveal that the proposed method is
better than existing approaches.
Keywords:
Data mining
Internet of things
Parallel computing
Recommender system
Smart homes
Unsupervised algorithms This is an open access article under the CC BY-SA license.
Corresponding Author:
Shymaa Sobhy
Department of Computer Science, Faculty of Computers and Information, Menofia University
Menofia, Egypt
Email: shaymaa.abdelaal@ci.menofia.edu.eg
1. INTRODUCTION
Internet of things (IoT) world is a network of connected devices for exchanging information data
and to control devices [1]. Devices with sensors and actuators combine objects on the Internet through
human-to-tool interactions. A main application of IoT is the smart home that is powered with various IoT
devices connected together [2]. The user can control the connected devices with smart home applications on
any electronic devices such as smartphones. The usage behavior of the IoT devices in the smart home is
important for collecting recurrent patterns relevant to the user. It can be viewed as a sequence of activities
and interactions with various devices to reflect the user’s intentions, actions and interactions patterns that
need analysis [3], [4].
Recommender systems can exploit knowledge collected from the users’ usage and offers some
helping recommendations to them [5]. They give more comfort, help older and disabled individuals, and
facilitate user’s life. They have methods that facilitate users ’ choices based on their preferences [6], [7].
Smart home recommendation applications become available to analysis the actions that matches user’s
interaction profile. They provide recommend services that user may need to perform. The goal is to help user
in controlling their life in best and comfortable manner.
Recently, a number of recommender system that could be delivered to user in their smart life has
been introduced. In [8] Quijano-Sánchez et al. introduce a system takes data of IoT devices to derive the
most frequent patterns and recommends the other possible active patterns. The patterns are most related to
the users and their usage of the IoT devices. In [9] Amatriain et al. introduce recommend actions that help
elderly people by providing smart agents that assist them. In [10] Thakur and Han introduce a recommender
system that address frequent choices of automation available for the user in his environment. In [11] Rasch
introduce a recommender system offers services to user based on the interaction profiles offered by the real

Int J Elec & Comp Eng ISSN: 2088-8708 
Context-aware recommender system for multi-user smart home (Shymaa Sobhy)
3193
existing sensors. In [12] Belghini et al. introduce a recommender system with supervised learning algorithms
to predict user action in the current contextual profile and provide recommendations. In [13] Gupta develop a
predictive model for location-aware optimizing resources in smart homes user that enhance his comfort. In
[14] Roy et al. introduce a method for context-aware services that adapt all corners of living part of home. In
[4] Nawara and Kashef introduce a recommender system help user when requires recommendations of
nearby places based on the ranking function. In [15] Miraoui introduce recommendations in a smart kitchen
by construct contextual profiles provide useful recipe options for the user.
The related works mentioned before provide useful services like helping elderly people by providing
smart agents, location-aware resources optimization in smart homes, predict next action and construct
interaction profiles that help user in his smart kitchen or living room or any part of home. However, they
have limitations when multiple users exist in same time in the same location. It results in complex scenarios
with conflicts between users in using different devices in home and how to deliver useful services for the
proper user related their priority on using devices. Another limitation is the use of traditional centralized data
mining algorithms in learning preferences. Such centralized algorithms are efficient in small datasets but
when datasets increased their efficiency is reduced.
This paper proposes a method to improve smart homes performance to be more interactive with
multiusers. It builds a context-aware recommender system that learns useful multi-user’s preferences by
using unsupervised algorithms and delivers useful recommendation services. It starts by simulating complex
multi user’s environment contexts in a virtual smart home tool to capture sensors data and construct
multiusers history of interactions is in the datasets. Then, it preprocesses multiusers datasets to manage
conflicts occurred by removing any redundancy. Finally, it constructs contextual information such as
location, time, and type of the day that are extracted from the previously collected data, and merge it with
multiusers preferences to provide useful context aware recommend for multiuser home devices. This paper
also proposes an approach to improve smart homes performance to be more responsive by using parallel
computing that solve challenges of traditional data mining algorithms in term of time and memory. These
algorithms used decentralized in learning process to find the most frequent activities and association rules of
multi users in smart home. The experimental results of the proposed method yielded good results in
providing useful recommendation for multiuser smart homes without conflicts between users complex
situations. The smart homes also become more responsive by reducing execution time.
In brief, the contributions of this paper are: i) making smart homes more interactive by build
context-aware recommendation system that provides useful recommendation for multiuser smart homes; and
ii) make smart homes more responsive using parallel computing that accelerate performance of learning
algorithms. The remainder of this paper is ordered as. The method is introduced in section 2. Section 3
provides the results and discussion. Section 4 presents the conclusion.
2. METHOD
This paper introduces a method that enables smart homes to be more responsive and interactive with
users. It recommends services for the user according to their current situation such as: turning off lights when
user is sleeping, turning on television when user exist in couch on the living room, turning on kitchen light
when user opens the kitchen door. In the following, the related work to the proposed method introduced then
the proposed method in brief is explained followed by method steps explanation in detail.
2.1. Related work to the proposed method
The recommender system starts by the training phase in which smart home learns a model of the
user activities. After the training phase, the system analyzes user’s situation and generates personalized
recommendations [16]. The system recommends services that help users in the future actions. There are
many algorithms for learning user’s actions like deep learning, machine learning, collaborative filtering,
content-based, knowledge-based and context-aware approach (that define the context related to the
application by different users) [17]. Context-aware recommendations create better user experiences to
improve performance in carrying out activities based on contexts such as location, identity, and actions of
users and physical sensors on devices with applications [18], [19]. They are aware of actions, their attributes
and possible activities that a user may perform by analyzing the user’s behavior and nature of doing various
actions [20], [21]. Data mining algorithms are classification; recommendation systems, clustering, and
association rule mining are used frequently to save useful or relevant information from massive datasets [22].
In the following, we review some of the existing techniques for smart home’s recommender system.
An action recommender system for IoT smart homes has been introduced [8]. The system takes data
of IoT devices as input then applies a pattern-mining algorithm to derive the most frequent and active
recurrent patterns of single user. Finally, it recommends the other possible actions from any active patterns.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 13, No. 3, June 2023: 3192-3203
3194
The resulted active patterns are most related to the users and their usage of the IoT devices making the
recommendations personalized. The system uses association rule mining for mining out recurrent patterns.
A framework was developed for generating a database that capture all possible user actions in any
activity and tries to find the activity goal [9]. They present an activity performance supervised recommender
system to recommend actions related to various activities of single user. They help elderly people by
providing environments with smart agents that can effectively assist them in user interactions and provides
useful services and gives recommend help in their current action.
The method proposed in [10] identified challenges that are useful to user of the smart homes like
difficult interfaces, configuration problems. It solved the challenges by proposing a recommender system that
addresses frequent choices of automation available for the user in the current situation. In [11], Rasch offer
the most important actions to user based on the interaction profile provided by the real sensors. They
recommend to use unsupervised learning such as association analysis to mine the rules from a frequent item
set by setting a minimum frequency threshold on the recurring items found.
A framework for finding useful user preferences and delivers relevant context-aware
recommendations is designed [12]. It transforms data into useful format by applying unsupervised algorithms
to find frequently occurring activities and to learn useful rules over those activities. Then, it constructs
contextual profile from the raw data and adding it to the rules mined. Finally, it applies supervised learning
algorithms to predict user action in the current contextual profile and provide recommendations.
A model that specifies location-aware optimizing resources in smart homes has been introduced
[13]. It offers efficient learning of user location’s context. Efficient prediction helps in controlling devices
with operations of automated services along user next location. Simulation results of the model on virtual
smart home indicates high prediction success and gives high optimization of energy consumption, operations,
and time of users to enhance their comfort.
A framework for context-aware services that simulate design of home living part has been proposed
[14]. In this framework, algorithms of machine learning such as naïve Bayes, fuzzy logic and case-based
reasoning techniques are used. Results show they improve the smart home user life and optimize their life.
A location-based context aware recommender method that finds a ranking function to offer top-k
services to the user has been introduced [4]. The contextual information is constructed by users as generated
rules using a rule-based language (RuleML). When user requires recommendations of the nearby places,
contextual data are constructed and rules is extracted. The top-k recommendations of nearby places based on
the ranking function are presented to the user.
The method proposed in [15] defines the context as sources to build a recommender model to offer
recommendations in a smart kitchen. It uses recorded interaction data to predict and recommend services to
the user. The recommendation model has three processes: food kind’s taxonomy, recipe taxonomy, and
condition feature data. The first process is to decide the food sections exist in refrigerator and cupboard in the
smart kitchen. Second, the model finds only those recipes which can be provided from the existing kinds in
the smart kitchen out of all the possible recipes. Finally, contextual profiles provide useful recipe options for
the user.
Table 1 provides a summary of related works. The summary includes the used mining algorithm,
IoT application, collect data is real or from smart home’s simulation, contextual data are considered or not,
parallel computing or centralized approach is used, and multi users in smart environment are handled or not.
The table shows that related work uses data unsupervised or supervised algorithms for learning and
predictions in real or simulated smart home environment. But, few of them add contextual information.
The related works focus on building a model for recommender system that help user in his smart
environment. However, none of the models deals with a high number of users and handles cases where lot of
activities is interleaved (i.e., complex activity definitions for multiusers in smart home). Also, they use
centralized data mining algorithms that find frequent user actions and gives useful recommendations for next
actions. Such algorithms are not efficient in larger context.
Table 1. The analysis of related works
RefNum Mining algorithm Application Focus Context information Multiusers Parallel
[8] Unsupervised learning smart home Real No No No
[9] Supervised learning smart home Real No No No
[10] Unsupervised learning smart home Real No No No
[11] Unsupervised learning smart home Real Yes No No
[12] Association prediction smart homes Simulation Yes No No
[13] Prediction algorithm smart homes Visualization No No No
[14] Machine learning living room Visualization No No No
[4] Ranking function smart homes Real Yes No No
[15] Machine learning smart kitchen Visualization Yes No No

3195
2.2. The processes of the proposed method
The processes of the proposed method are shown in Figure 1. It starts with capturing sensors data.
Then, history of interaction formalization is recorded in the datasets and data is pre-processed to address the
complex situation of multi users. After that, conflicts are managed; frequent items and association rules are
learned with Apriori algorithm in parallel. Finally, contextual profile like location and time of the day are
added to the rules to provide useful services for smart applications.
Figure 1. Process of the proposed method
2.2.1. Capture sensors data
Serving the user in an effective manner in his smart environment requires datasets for recording the
interaction captured from a smart home. Figure 2 shows the techniques for generating datasets for smart
home for data analysis. It shows two main methods: real and simulation. When building real smart home,
many challenges are faced like i) having a robust and continuous collecting techniques for the sensors data
and ii) finding the best annotation techniques for user actions. Simulation tool handle these limitations of real
datasets by dataset generation with robust algorithms to capture sensor data. The simulation tools design can
be done using three ways: model-based, interactive and hybrid approaches. The model-based uses statistical
models to generate datasets. The interactive method relies on real-time using an avatar navigates by
simulated person. The hybrid combines model-based and interactive approaches. The targets from simulation
tool are visualization, dataset generation and context-awareness.
Figure 2. Techniques for datasets generation
Table 2 compares smart home simulation tools. They are compared by simulation technique, free
access, scripting availability in simulation, multi users or single user and time related considering. As shown
in the table, most of the existing tools problems are not open source and none of them handle multiuser.

 ISSN: 2088-8708
3196
In our proposed method, personal data are collected in a virtual environment as real datasets are
limited in numbers. The dataset has been collected by an open source simulator called OpenSHS that attempt
to create a realistic dataset to facilitate research in smart home design [3]. The users simulate various actions
of their day and simulator records the state of sensors placed in different rooms of the home. The OpenSHS
tool design is a bedroom, kitchen, living, hallway, bathroom, and office. The smart home is installed with
twenty-nine binary sensors places at different objects such as television, lights which offers the state of the
object on which is placed. List of sensors name, type and description in OpenSHS datasets are showed in
Table 3. The simulator unfortunately is designed for single user as it has one dataset for all interactions of
that user. Thus, we simulate the interaction for multi users separately. Then, the collected datasets are edited
to handle location of each user in the time of that context and inject some conflicts between all users (i.e., the
users exist at the same time and location). Finally, we convert all datasets to indicate the actions of all users
(see the details in the preprocessing step).
Table 2. Smart home simulation tools comparison
RefNum Simulation tool Simulation technique Open source scripting Multiusers Time related
[23] Home I/O interactive NO NO NO YES
[3] OpenSHS Hybrid YES NO NO YES
[4] SIMACT Model based YES YES NO NO
[24] IE Sim interactive NO YES NO YES
[25] SE SiM Model based YES YES NO NO
[26] Persim 3D Model based NO NO NO NO
Table 3. Design of OpenSHS simulator
Sensor_Num Sensor name Sensor type Sensor description
1 Wardrobe binary bedroom wardrobe sensor
2 tv binary living TV sensor
3 oven binary kitchen oven sensor
4 officeLight binary office ceiling light sensor
5 officeDoorLock binary office door lock sensor
6 officeDoor binary office door sensor
7 officeCarp binary office carpet sensor
8 office binary office room office sensor
9 mainDoorLock binary main door lock office
10 mainDoor binary main door sensor
11 livingLight binary living ceiling light sensor
12 livingCarp binary living carpet sensor
13 kitchenLight binary kitchen ceiling light sensor
14 kitchenDoorLock binary kitchen door lock sensor
15 kitchenDoor binary kitchen door sensor
16 kitchenCarp binary kitchen carpet sensor
17 hallwayLight binary hallway ceiling light sensor
18 fridge binary kitchen fridge sensor
19 couch binary living couch sensor
20 bedroomLight binary bedroom ceiling light sensor
21 bedroomDoorLock binary bedroom door lock sensor
22 bedroomDoor binary bedroom door sensor
23 bedroomCarp binary bedroom carpet sensor
24 bedTableLamp binary bathroom bed Table lamp
25 bed binary bathroom bed sensor
26 bathroomLight binary bathroom ceiling light sensor
27 bathroomDoorLock binary bathroom door lock sensor
28 bathroomDoor binary bathroom door sensor
29 bathroomCarp binary bathroom carpet sensor
30 Activity String current user activity sensor
31 Timestamp String Timestamp every second
2.2.2. History of interaction formalization
In this step, the history of interactions is constructed and saved on databases. The proposed
approach uses OpenSHS tool to save all users interactions by sensor names to datasets that has a timestamp
column of datetime. Figure 3 present the design phase at the top and simulation phase at the bottom with time
stamp. The datasets also has an activity column which divides the activity of the user such as eat, sleep, walk
and others. OpenSHS datasets generation is hybrid model that creates rich realistic datasets depends on
replication algorithm [3]. Two elements were used, the researcher and the participant. Most of the work done
by OpenSHS is done by researchers. Participants are anyone who voluntarily simulates their activities.

3197
Three main phases: the design phase, the simulation phase, and the aggregation phase. In design
Phase researchers create a virtual environment, import smart devices, assign actions to tags, and develop
context. After developing the smart home model, the researcher creates a context for the simulation. Context
is the specific time range that researchers are interested in modeling, for example, situations in the morning,
afternoon, and evening. In simulation phase the researcher to specify which context to simulate. Each context
has the default state of the sensor and position of the participant. Then the participants started to simulate
their Activities in this situation. During the simulation, the sensor output and the status of various devices are
recorded and saved in a temporary data set. The modeling or simulation phase aims to capture the details of
the actual interaction between participants. In aggregation phase the researcher can aggregate the
participants’ generated sample activities such as create the final recorded event. The result of the modeling
phase forms a series of typical actions for each context.
The aggregation stage aims to provide a solution for generating large data sets in a short modeling
time. The replication algorithm used to reproduce the results of the simulation phase by extracting enough
samples for each expected context. This feature allows OpenSHS to combine the advantages of both
approaches, a hybrid approach. OpenSHS depend on game engine of on Blender with Python editors. Blender
is available for all three major operating systems. Blender uses OpenGL for its Game Engine which is also, a
cross-platform 3D technology available for the major operating systems. The Blender Game Engine: The
physics engine facilitates the simulation of different types of real sensors and devices. For example, Blender
has a (Near) sensor that only triggers when the user-controlled 3D avatar is physically close to other objects
in the scene. All logic and interactions between the avatar and the virtual environment are developed with it.
In addition, all OpenSHS modules are programmed in Python [27].
Figure 3. The design phase and simulation phase
2.2.3. Preprocessing datasets for multiusers environment
The proposed approach preprocesses OpenSHS row data to work with multiusers without conflicts
between users in the same time and location. Table 4 shows parts of datasets contents after preprocessing
with devices off. For example, first user actions on devices like television in living and kitchen light in
kitchen can be either on or off. Datasets represented in binary format as 1 means devices state is open and 0
means that devices is off. Also, on locations like living or kitchen 1 mean user are exit on that location and 0
mean that he is not exist. The steps of dataset preprocessing are: i) edit sensor names for all users’ datasets
with user’s id; ii) insert location column based on five sensors called kitchen, bathroom, bedroom, office,
living, and hallway. This captured from the sensors states located in the carpets of every part of the house,
representing presence of the users in that part; and iii) convert sensor states that are saved in binary format of

 ISSN: 2088-8708
3198
zeros and ones to represent the user’s actions on smart home devices like kitchen light is on and kitchen light
is off by add more columns to some sensors in dataset that represent the off action of devices.
Table 4. The datasets of first user preprocess with devices off and on and location
u1
tv
off
u1
tv
on
u1
oven
off
u1
oven
on
u1office
Light on
u1office
Light off
u1kitchen
Light on
u1kitchen
Light off
u1
Office
u1
Living
u1
Kitchen
u1
Bedroom
u1
Bathroom
1 0 0 0 1 0 0 1 0 0 1 0 0
0 0 0 0 0 0 1 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 1 0 0 0 0 0 1 0 1 1 0
0 0 0 0 0 0 0 0 1 0 1 1 0
0 0 0 0 0 0 0 0 1 0 1 1 0
0 0 0 0 0 0 0 0 1 0 1 1 0
2.2.4. Find frequent events in parallel
The purpose of data mining techniques is to extract a valuable information and knowledge from
huge amounts of data stored in databases. It is the process of finding out formerly unknown, useful and
valuable patterns from a large amount of data stored in a database. Data mining examines databases for
identifying hidden patterns and valuable information that sometimes experts may not observe as it occurs
outside their expectations [13]. The discovered patterns are accessible to the user and could be stored as new
information in the information database. Classical Apriori algorithm is for learning data mining association
rule and discovers association among different set of data [14].
Searching for most occurring itemsets is one of the most targets for data mining fields [28]. Apriori
algorithm is the most efficiently used technique for specifying most occurring itemsets from a transactional
dataset [29], [30]. However, the Apriori algorithm searches the dataset a lot of times for specifying candidate
itemsets. But, when the dataset becomes very big, both space used and computational cost can be more
expensive. Memory and CPU resources are unfortunately very limited, which make execution of the
algorithm inefficient with low performance. Thus, the proposed method use multi-processing to execute
Apriori algorithm in parallel [31], [32] with the following steps of parallel mining to get frequent events for
each user datasets: i) define number of parallel process to run as N process, ii) divide the input file into
different N parts for parallel processing, iii) start function that calculated the frequent dataset parallel, and
iv) create the pool object to start mapping the input to different process object and join parts.
2.2.5. Learning association rules in parallel
To find rules that specify relation between frequent items with association rules in parallel as
frequent items list size is huge [33]–[35]. In the following steps the algorithm use multi-processing to execute
association algorithm. This is performed in parallel are used for each resulted frequent items from previous
step by Apriori algorithm. We present here the steps of association rule parallel learning: i) splitting the input
frequent itemsets resulted from Apriori algorithm into a list of subset data frames; ii) map the list of subset
data frames with pool map function that take frequent patterns as input and the min threshold to calculate the
rules and return rules; and iii) join resulted rules lists in one list and print rules with antecedents and
consequences.
Figure 4 shows a flowchart for Apriori and association rule algorithms for single data set, while the
parallelization of these algorithms is shown in Figure 5. The Apriori algorithm set a minimum value for
support and confidence. It then reads each transaction and finds rules for the items that have certain default
existence support and have a minimum value for co-occurrence with other items confidence. Later on, it
extracts all the subsets having higher value of support than minimum threshold, and select all the rules from
the subsets with confidence value higher than minimum threshold. This is done for all datasets in parallel by
multiprocessing as explained before.
2.2.6. Manage conflicts between resulted rules
Merging all users’ rules in one data frame results in conflicts between users exists in same locations
at the same time. The proposed method handles it by the following steps: i) search merged rules to specify
locations for the different users exist on same time; ii) separate the rules into smaller ones based on locations
founded in step one; iii) for all resulted sets of rules convert it from many antecedents and many consequents
to one to one; iv) remove resulted duplication in rules; v) set for all rules different priority for active devices
that indicates who can access it if multi users exist at the same time; and vi) select user with high priority on
each active device on the specified location.

3199
Figure 4. Flowchart for Apriori and association rule algorithms for single dataset
Figure 5. Flowchart for parallel computing of Apriori and association rule algorithms

 ISSN: 2088-8708
3200
2.2.7. Merge multiusers rules and contextual profile
Many related works ignore contextual profile construction. But it is very important to make the
simulation aware of environment and offer information useful to the users that help him in his life. The
proposed method uses OpenSHS row data and analysis to construct context profiles. Along with the activity
column already placed in OpenSHS row dataset, three contexts proposed: i) time of the day as hours of the
day are classified into six parts of the day, ii) type of the day specifying all days on the week by separated it
to a weekday or weekend, iii) user location calculated by the sensors states located in the carpets of every
part of the home rooms.
2.2.8. Recommending useful context-aware services for smart applications
After adding contextual information to row data and get frequent rules with consequent and
antecedent. They are collected in one file indicates each data set columns with resulted consequent. This file
used to automate services like IF (U1ovenON) Then (U1kitchenlighton). Table 5 show subset of resulted
rules antecedents and consequents for kitchen location after running parallel Apriori and association rules of
multi users of tree different users and managing conflict between then as found that user1 and user2 found in
the same time in kitchen so we give priority to user on kitchen light and oven. Then, the consequent column
adds to the original data to be used in service automation and prediction of supervised algorithms.
Table 5. The resulted rule for kitchen location
RuleNum Antecedents Consequents
0 u1ovenon u1kitchenLighton
1 u1Kitchen u1kitchenLighton
2 u1Kitchen u1kitchenLightoff
3 u1ovenoff u1kitchenLightoff
6 u3mainDoorLockon u3kitchenDooron
7 u3Kitchen u3kitchenDooron
8 u3kitchenDooron u3mainDoorLockon
9 u3Kitchen u3mainDoorLockon
10 u3mainDoorLockon u3fridgeoff
11 u3Kitchen u3fridgeoff
12 u3mainDoorLockon u3bedroomDooron
13 u3Kitchen u3bedroomDooron
14 u3bedroomDooron u3mainDoorLockon
20 u3kitchenDooron u3fridgeoff
22 u3bedroomDooron u3kitchenDooron
24 u3kitchenDooron u3bedroomDooron
26 u3bedroomDooron u3fridgeoff
51 u3fridgeoff u3mainDoorLockon
76 u3fridgeoff u3kitchenDooron
114 u3fridgeoff u3bedroomDooron
3. RESULTS AND DISCUSSION
To evaluate the proposed method, the experimental setting is Intel core i5 CPU and 8.00 GB RAM
on 64-bit operating system to build a context-aware recommendation system for multiuser smart homes in
simulation environment tool called OpenSHS (built by Python and blender). The anaconda jupyter notebook
Python are used to implement mining algorithms and Python multi-processing was used to improve
performance with parallel execution. This section firstly introduces datasets resulted for multiusers.
Secondly, experimental results for parallel computing of Apriori and association rules are introduced.
3.1. Datasets for multiusers
Table 6 show results of applying the proposed method of converting simulation row data into
multiuser data, and remove any duplication or conflict if two or more user exist in the same location. Three
users in home and dataset records them in four different locations so conflict is found between users as
Table 6 shows. The proposed method separates the rules of three users and for each location it specifies
user’s priority on active devices. Table 6 includes 168 frequent items and 1,811 rules for user one, and for
user two there are 104 frequent items and 526 rules, and 76 frequent items and 486 rules for user three. The
total merged rules for all users are 2,823 rules.
Table 7 Show results of conflict removing for merged rules for all users that equal 2,823 rules.
When they are separated into four different locations, the rules are 130 for office, 641 for living, 152 for
kitchen, and 900 for bedroom with total 1,823 the other rules appears when one user exists. The rules are then
cleaned by removing duplication. The rules become 17 for office, 66 for living, 29 for kitchen, and 177 for

3201
bedroom with total 289 rules. Finally, after specifying which user can access devices by setting priority, the
final rules are 12 for office, 42 for living, 20 for kitchen, 80 for bedroom (i.e., the total rules are 154) as
shown in Table 7 so results indicates reducing rules from 1,823 to 154 rule without conflicts.
Table 6. The merged rules for three users
NUM users U1 U2 U3
Frequent items 168 104 76
Association rules 1811 526 486
Merged rules for all users=2823
Table 7. The conflict removing for merged rules
Location office Location living Location kitchen Location bedroom Total
Merged rules number 130 641 152 900 1823
Rules number after duplication remove 17 66 29 177 289
Rules number after setting users priority 12 42 20 80 154
3.2. Parallel and sequential execution time of Apriori and association rules
This section shows results for the execution time for Apriori and association rules of the three
different datasets for different context with three users. The execution time is 45 ms in the parallel case while
it is 142 ms in sequential for dataset one. Second dataset requires 34.2 ms during the parallel execution, while
94 ms in the sequential case. Third dataset has in total 26 ms in parallel case and 73 ms during sequential
execution. Figure 6 comparing three different users datasets for parallel and sequential execution time in (a)
Apriori algorithm and (b) association rules.
(a)
(b)
Figure 6. Comparing three different users datasets for parallel and sequential execution time in (a) Apriori
algorithm and (b) association rules

 ISSN: 2088-8708
3202
4. CONCLUSION
This paper proposed method to improves smart home performance to be more responsive to his
users. It built a model that extracts useful user-preferences and delivers context-aware recommendations. The
proposed method makes smart home more interactive/responsive with users by the use of parallel
computation of association rule algorithm. The proposed method transforms OpenSHS simulator data into
suitable format and detects frequently occurring events. Then, it extracts and combines the context profile of
interaction to associate it with resulted rules. The proposed method introduces method for learning multi-user
preferences with multi device that allows smart home users to efficiently use the appliances. The proposed
method finds frequent items and association rules for multiple datasets for all users. It extracts meaningful
rules without conflict between the multi-users in the same time and location in the smart home. Experiments
on a massive dataset demonstrated outstanding applicability of this algorithm. As a future work, we plan to
improve a security model for smart home that is remotely controlled via web application. This can be done
through authorization by access control and considering contextual information like location, time of request,
and history of user action.
REFERENCES
[1] P. Sethi and S. R. Sarangi, “Internet of things: architectures, protocols, and applications,” Journal of Electrical and Computer
Engineering, pp. 1–25, 2017, doi: 10.1155/2017/9324035.
[2] M. Domb, “Smart home systems based on internet of things,” in IoT and Smart Home Automation [Working Title], IntechOpen,
2019.
[3] N. Alshammari, T. Alshammari, M. Sedky, J. Champion, and C. Bauer, “OpenSHS: open smart home simulator,” Sensors,
vol. 17, no. 5, May 2017, doi: 10.3390/s17051003.
[4] D. Nawara and R. Kashef, “IoT-based recommendation systems-an overview,” in 2020 IEEE International IOT, Electronics and
Mechatronics Conference (IEMTRONICS), Sep. 2020, pp. 1–7, doi: 10.1109/IEMTRONICS51293.2020.9216391.
[5] K. Rasch, “Smart assistants for smart homes,” KTH Royal Institute of Technology, 2013.
[6] Y. Himeur et al., “A survey of recommender systems for energy efficiency in buildings: Principles, challenges and prospects,”
Information Fusion, vol. 72, pp. 1–21, Aug. 2021, doi: 10.1016/j.inffus.2021.02.002.
[7] L. Yao, X. Wang, Q. Z. Sheng, S. Dustdar, and S. Zhang, “Recommendations on the internet of things: requirements, challenges,
and directions,” IEEE Internet Computing, vol. 23, no. 3, pp. 46–54, May 2019, doi: 10.1109/MIC.2019.2909607.
[8] L. Quijano-Sánchez, I. Cantador, M. E. Cortés-Cediel, and O. Gil, “Recommender systems for smart cities,” Information Systems,
vol. 92, Sep. 2020, doi: 10.1016/j.is.2020.101545.
[9] X. Amatriain, N. Oliver, and J. M. Pujol, “Data mining methods for recommender systems,” in Recommender systems handbook,
Springer, 2011, pp. 39–71.
[10] N. Thakur and C. Y. Han, “A context-driven complex activity framework for smart home,” in 2018 IEEE 9th Annual Information
Technology, Electronics and Mobile Communication Conference (IEMCON), Nov. 2018, pp. 801–806, doi:
10.1109/IEMCON.2018.8615079.
[11] K. Rasch, “An unsupervised recommender system for smart homes,” Journal of Ambient Intelligence and Smart Environments,
vol. 6, no. 1, pp. 21–37, 2014, doi: 10.3233/AIS-130242.
[12] N. Belghini, N. Gouttaya, W. Bouab, and A. Sayouti, “Pervasive recommender system for smart home environment,”
International Journal of Applied Information Systems, vol. 10, no. 9, pp. 1–7, May 2016, doi: 10.5120/ijais2016451528.
[13] S. Gupta, “A context-aware personalized recommender system for automation in IoT based smart home environment,” Dublin
Business School, 2020.
[14] A. Roy, S. K. Das Bhaumik, A. Bhattacharya, K. Basu, D. J. Cook, and S. K. Das, “Location aware resource management in
smart homes,” in Proceedings of the First IEEE International Conference on Pervasive Computing and Communications, 2003.
(PerCom 2003)., 2003, pp. 481–488, doi: 10.1109/PERCOM.2003.1192773.
[15] M. Miraoui, S. El-etriby, C. Tadj, and A. Zaid Abid, “A hybrid modular context-aware services adaptation for a smart living
room,” Intelligent Automation and Soft Computing, vol. 24, no. 2, pp. 299–308, Jun. 2018, doi: 10.1080/10798587.2017.1281565.
[16] Pratibha and P. D. Kaur, “A context-aware recommender engine for smart kitchen,” in Smart Innovations in Communication and
Computational Sciences, Springer Singapore, 2019, pp. 161–170.
[17] P. Mishra, S. K. Gudla, A. D. ShanBhag, and J. Bose, “Alternate action recommender system using recurrent patterns of smart
home users,” in 2020 IEEE 17th
Annual Consumer Communications and Networking Conference (CCNC), Jan. 2020, pp. 1–6,
doi: 10.1109/CCNC46108.2020.9045128.
[18] H. F. Abdulkarem, G. Y. Abozaid, and M. I. Soliman, “Context-aware recommender system frameworks, techniques, and
applications: a survey,” in 2019 International Conference on Innovative Trends in Computer Engineering (ITCE), Feb. 2019,
pp. 180–185, doi: 10.1109/ITCE.2019.8646564.
[19] M. Altulyan, L. Yao, X. Wang, C. Huang, S. S. Kanhere, and Q. Z. Sheng, “Recommender systems for the internet of things: a
survey,” 2020, arXiv:2007.06758.
[20] A. Alotaibi and C. Perera, “Smart home human activity simulation tool for openhab-based research,” Tech. Rep. ACM, 2019.
[21] S. Faieq, A. Front, R. Saidi, H. El Ghazi, and M. D. Rahmani, “A context-aware recommendation-based system for service
composition in smart environments,” Service Oriented Computing and Applications, vol. 13, no. 4, pp. 341–355, Dec. 2019,
doi: 10.1007/s11761-019-00277-7.
[22] A.-K. Moon, H.-H. Kim, J.-Y. Park, and Y.-I. Choi, “Context based user profile for personalization in ubiquitous computing
environments,” The Journal of Korean Institute of Communications and Information Sciences, vol. 34, no. 58, pp. 542–551, 2009.
[23] I. B. B. Gottfried and H. Aghajan, “Learning about preferences and common behaviours of the user in an intelligent
environment,” Behaviour Monitoring and Interpretation-BMI: Smart Environments, vol. 3, 2009.
[24] B. Riera and B. Vigário, “HOME I/O and FACTORY I/O: a virtual house and a virtual plant for control education,” IFAC-
PapersOnLine, vol. 50, no. 1, pp. 9144–9149, Jul. 2017, doi: 10.1016/j.ifacol.2017.08.1719.
[25] J. Synnott, L. Chen, C. Nugent, and G. Moore, “IE Sim-A flexible tool for the simulation of data generated within intelligent
environments,” in Lecture Notes in Computer Science, Springer Berlin Heidelberg, 2012, pp. 373–378.

3203
[26] B. Ho, D. Vogts, and J. Wesson, “A smart home simulation tool to support the recognition of activities of daily living,” in
Proceedings of the South African Institute of Computer Scientists and Information Technologists 2019 on ZZZ - SAICSIT ’19,
2019, pp. 1–10, doi: 10.1145/3351108.3351132.
[27] J. W. Lee, S. Cho, S. Liu, K. Cho, and S. Helal, “Persim 3D: context-driven simulation and modeling of human activities in smart
spaces,” IEEE Transactions on Automation Science and Engineering, vol. 12, no. 4, pp. 1243–1256, Oct. 2015, doi:
10.1109/TASE.2015.2467353.
[28] J. Synnott, C. Nugent, and P. Jeffers, “Simulation of smart home activity datasets,” Sensors, vol. 15, no. 6, pp. 14162–14179, Jun.
2015, doi: 10.3390/s150614162.
[29] X. Yuan, “An improved Apriori algorithm for mining association rules,” 2017, doi: 10.1063/1.4977361.
[30] J. Dongre, G. L. Prajapati, and S. V Tokekar, “The role of Apriori algorithm for finding the association rules in data mining,” in
2014 International Conference on Issues and Challenges in Intelligent Computing Techniques (ICICT), Feb. 2014, pp. 657–660,
doi: 10.1109/ICICICT.2014.6781357.
[31] P. Stanišić and S. Tomović, “Apriori multiple algorithm for mining association rules,” Information Technology and Control, vol.
37, no. 4, 2008.
[32] S. Biswas, N. Biswas, and K. C. Mondal, “Parallel and distributed association rule mining algorithms: a recent survey,”
Information Management and Computer Science, vol. 2, no. 1, pp. 15–24, Sep. 2019, doi: 10.26480/imcs.01.2019.15.24.
[33] R. Agrawal and R. Srikant, “Fast algorithms for mining association rules,” in Procedings 20th
International Conference Very
Large Data Bases, vol. 1215, 1994, pp. 487–499.
[34] R. Agrawal, T. Imieliński, and A. Swami, “Mining association rules between sets of items in large databases,” in Proceedings of
The 1993 ACM SIGMOD international Conference on Management of Data-SIGMOD ’93, 1993, pp. 207–216,
doi: 10.1145/170035.170072.
[35] A. Savasere, E. R. Omiecinski, and S. B. Navathe, “An efficient algorithm for mining association rules in large databases,” in
Proceedings of the 21th
International Conference on Very Large Data Bases, 1995, pp. 432–444.
BIOGRAPHIES OF AUTHOR
Shymaa Sobhy is a Ph.D. student at Menoufia University Faculty of Computers
and Information, Egypt. She works Teaching Assistant at Computer Science Department,
Menoufia University, Egypt. She received the B.S. Degree in Computer Science Faculty of
Computers and Information, Menoufia University in May 2012 and an M.S. Degree in
Computer Science from Menoufia University, Faculty of Computers and Information in 2018.
Her main research interest is in software engineering, data mining, database, internet of things,
data privacy, security, web application, artificial intelligence and recommender system. She
can be contacted at email: sshh_yy_mmaa@yahoo.com.
Eman M. Mohamed received her B.Sc. in Computer Science from Menoufia
University, Faculty of Computers, and Information in 2008. She received her M.Sc. in
Computer Science from Menoufia University, Faculty of Computers, and Information in 2012.
She received his Ph.D. in computer science from Menoufia University, Faculty of Computers,
and Information in 2019. She works a Teacher at Computer Science Department, Menoufia
University, Egypt. Her research interest includes cloud computing, big data, distributed
system, database, bioinformatics, data privacy, and security. She can be contacted at email:
eman.mohamed@ci.menofia.edu.eg.
Arabi Keshk received the B.Sc. in Electronic Engineering in Computer Science
and Engineering from Menoufia University, Faculty of Electronic Engineering in 1987. He
received the M.Sc in Electronic Engineering in Computer Science and Engineering from
Menoufia University, Faculty of Electronic Engineering in 1995. He received his Ph.D. in
Electronic Engineering from Osaka University, Japan in 2001. He works as Associate
Professor Head of Computer Science Department 2009. He works as dean of Faculty of
Computers and Information, Menoufia University at 2016. He works as the present of Delta
Technological University in 2022. His research interest includes cloud computing, big data
analytics, distributed systems, digital systems testing, software testing and data security. He
can be contacted at email: arabikeshk@yahoo.com.
Mahmoud Hussein received his B.Sc. in Computer Science from Menoufia
University, Faculty of Computers and Information in 2006. He received his M.Sc. in
Computer Science from Menoufia University, Faculty of Computers and Information in 2009.
He received his Ph.D. in Software Engineering from Swinburne University of Technology,
Faculty of Information and Communications Technology in 2013. He works as Teaching
Assistant at Menofia University in 2007. He works Assistant Lecturer at Menofia University
in 2009. He does his Postdoctoral in Swinburne University of Technology. He works his
research interest includes software engineering, data mining, machine learning, data privacy,
and security. He can be contacted at email: mahmoud.hussein@ci.menofia.edu.eg.

Context-aware recommender system for multi-user smart home

More Related Content

Similar to Context-aware recommender system for multi-user smart home (20)

More from IJECEIAES (20)

Recently uploaded (20)

Context-aware recommender system for multi-user smart home