Microsoft Word - ETASR_V12_N5_pp9186-9191
Engineering, Technology & Applied Science Research Vol. 12, No. 5, 2022, 9186-9191 9186
www.etasr.com Rohilla et al.: An Empirical Framework for Recommendation-based Location Services Using …
An Empirical Framework for Recommendation-based
Location Services Using Deep Learning
Vinita Rohilla
Department of Computer Science and Engineering, School of
Engineering and Technology, Sharda University, Greater
Noida, India and
Department of CSE, Maharaja Surajmal Institute of
Technology, New Delhi, India
vinita_rohilla@msit.in
Mandeep Kaur
Department of Computer Science and Engineering
School of Engineering and Technology
Sharda University
Greater Noida, India
mandeep.kaur@sharda.ac.in
Sudeshna Chakraborty
Department of Computer Science and Technology
Lloyd Institute of Engineering and Technology
Greater Noida, India
sudeshna2529@gmail.com
Received: 8 June 2022 | Revised: 8 July 2022 | Accepted: 10 July 2022
Abstract-The large amount of possible online services throws a
significant load on the users' service selection decision-making
procedure. Α number of intelligent suggestion systems have been
created in order to lower the excessive decision-making expense.
Taking this into consideration, aν RLSD (Recommendation-based
Location Services using Deep Learning) model is proposed in this
paper. Alongside robustness, this research considers the
geographic interface between the client and the service. The
suggested model blends a Multi-Layer-Perceptron (MLP) with a
similarity Adaptive Corrector (AC), which is meant to detect
high-dimensional and non-linear connections, as well as the
location correlations amongst client and services. This not only
improves recommendation results but also considerably reduces
difficulties due to data sparseness. As a result, the proposed
RLSD has strong flexibility and is extensible when it comes to
leveraging context data like location.
Keywords-location services; recommendation system; deep
learning; similarity corrector
I. INTRODUCTION
Numerous corporations and organizations have started to
wrap their created business apps into more and more
widespread multiple online services [1-4]. Many lightweight
recommendation methods have also been created to reduce the
expenses of consumer decision-making. Generally, by
evaluating available client data, a recommender system can
deduce a person's approximated tastes and preferences and then
offer reliable suggestions [5-9]. It efficiently captures the
relevant information and preferences of service clients.
Location is an important component in decision
recommendations since the Quality of Service (QoS) is
sometimes significantly dependent on client or service location.
Moreover, location-aware QoS data frequently contain
sensitive user information, such as a user's most recently
requested service set, the amount of time it takes a user to
summon a services located in a nation, etc. Service-Oriented
Architecture (SOA) is commonly employed in decentralized
computing environments [10], including cloud computing and
mobile computing. Location-based services are more
successful to provide recommendations based on user's
interests [12, 13].
Collaborative Filtering (CF) is a popular way for creating a
recommendation system. In recommendation systems, the
current user-based CF mechanism seeks for a comparable user
based on their prior behavior. The ideas must be according to
the user's interest. For example, a user's visited locations are
denoted by pins of various forms. The 4 customers check in at
9 distinct locations: 'Exhibition,' 'Hotel,' 'Campground,'
'Shopping Centre,' 'Cineplex,' 'Opera Cinema,' 'Cafe,' 'Library,'
and 'College.' The CF approach often creates a similarity
measure between various users and offers suggestions based on
the maximum value. Let's create a suggestion for customer A1.
We could observe that user A1s activities are more comparable
to users A2s activities because both went into 'Cineplex,'
'Opera Cinema,' and 'Cafe.' Based on the activities of user A2,
the CF technique now recommends 'Library' to user A1.
Unfortunately, this is not the greatest fit for user A1 when user
A1's visiting information indicates a preference in
'Entertainment' venues, like 'Gallery.' To address this
constraint, we will construct an interested profile for every
customer so that suggestions may be made as needed.
The following are the major contributions to this study:
• An RLSD framework is developed, which blends an MLP
with a similarity Adaptive Corrector in a new way.
Corresponding author: Vinita Rohilla
Engineering, Technology & Applied Science Research Vol. 12, No. 5, 2022, 9186-9191 9187
www.etasr.com Rohilla et al.: An Empirical Framework for Recommendation-based Location Services Using …
• The performance of the suggested strategy was assessed
and compared with 9 advanced options.
• The proposed technique not only improves
recommendation performance, but it also significantly
reduces difficulties due to data sparseness.
II. LITERATURE SURVEY
Authors in [14] focused mostly on the cold-start dilemma
for service advice depending on user and service locations. To
resolve this concern, they offered a service recommendation
approach based on CF and geo location, which takes into
account the characteristics of services at the edges, movement,
and customer requests over time. To begin, they used a
multidimensional weighted approach to synthesize the service
quality for every dimension in different temporal chunks. In
respect of recommendations accuracy, the empirical findings
suggest that their multidimensional inverse similarity
recommendation method (MDITCF) built on time-aware CF
surpasses the Inverted CF recommendation method. Authors in
[15] suggested a novel deep CF approach to service
recommendations, dubbed Location-aware Deep CF (LDCF).
These main advancements included in this approach are: 1) the
location attributes are located into high-dimensional dense
imbed matrices. MLP encapsulates the HD and non - linear
features. Similarity ACs are integrated in the output nodes to
first rectify the forecasting QoS. With this, LDCF not only
understands the high-dimension and non-linear correlation
among the user and services, but it can also dramatically reduce
data sparseness. Extensive studies on a real-world Web service
[15] database show that LDCF's recommendations efficiency
clearly exceeds 9 state-of-the-art services recommendations
approaches. Authors in [16] used location information in the
factorization machine for QoS predictions, while
contextualized information helps the CF similarity calculation.
These computations understand the lower dimension and linear
characteristics of customers and service providers. Authors in
[17] suggested a tailored location-aware CF approach for web
service recommendations. Picking comparable neighbors for
the user based service the suggested technique takes advantage
of both customer location and online services. The technique
also incorporated an improved similarity metric for user and
web services through accounting for their customized
contribution. The test findings demonstrate that this
methodology greatly increases QoS predictive performance and
computing efficiency when compared to earlier CF-based
systems.
III. PROPOSED METHODOLOGY
The major problem in recommending services is estimating
QoS. CF is the most extensively utilized QoS predicting
technology currently available. Conventional CF techniques,
on the other hand, suffer from two flaws: 1) The similarity
modeling approach used by conventional CF-based techniques
can only gain knowledge of low-dimension and linear
characteristics or features from previous networks of user and
service and 2) the commonly used data sparseness dilemma in
reality has a massive effect on recommendation systems'
efficiency. This section provides a detailed description of the
proposed model and methodology. The section briefly
describes the loss function, the optimizer, and the algorithm
used in this work.
A. Deep Neural Networks
A Deep Neural Network (DNN) is one type of Neural
Network with more than 2 layers: an input layer, at least one
hidden layer, and an output layer. In a process known as
"feature hierarchy," every layer conducts certain kinds of
sorting and ordering. Handling with unorganized or
unstructured input is among the most important applications of
highly powerful neural network models [18].
Fig. 1. Recommendation based location services based on deep learning
architecture.
The RLSD model is a multiple-layered feed forward neural
network with 3 main layers. An Input Layer is used to create
the input variables needed from the Intermediate Layer and the
similarities needed for the Output Nodes. The Intermediate
Layer is employed for centralized training to ensure getting
high-dimension and non-linear characteristics.
1) Input Layer
The Input layer's primary function is to regulate the initial
Input. The Input contains user identification, location-based
info, service identifiers, and service geo locations in Keras1's
Embedding Layer, that may be thought as a special fully-
connected layer lacking bias terms. Essentially, the Embedded
Layer uses one-hot encoding on the inputs to build a '0' vector
with a given dimension, while the vector's i-th point is set to 1
[19]. The category characteristics are translated to high-
dimension fully connected embedding vectors using this
procedure. The mapping procedure is depicted in (1) - (4):
��� = ��(��� � + ��) (1)
��� = ��(��� �� + ��) (2)
��� = ��(��� � + ��) (3)
��� = ��(��� �� + ��) (4)
where � and � are the user's identifier and the service’s
identifier, correspondingly, �� and �� are the initial inputs for
the user’s and service’s location, �� and �� are user's and
service's embedding weighted matrices, �� is a bias term
Engineering, Technology & Applied Science Research Vol. 12, No. 5, 2022, 9186-9191 9188
www.etasr.com Rohilla et al.: An Empirical Framework for Recommendation-based Location Services Using …
initialized to 0, �� is the embedding layer's activation function,
which is considered as the identity function for this work, ���
and ��� are the k-dimensional user_id and location embedding
vector correspondingly, and U and S are the k-dimensional
service_id and location embedding vector correspondingly.
These identifying feature vectors representations are
integrated with the appropriate location features set to generate
the user feature map and service feature map. Next, we append
both feature maps to generate the intermediate layers input
vector sequence. The equation is written in the following form:
� = ɸ(��� , ��� ) = �
���
���
� (5)
S = ɸ(��� , ��� ) = �
���
���
� (6)
X = ɸ(U, V) = ���� (7)
where Φ is the mergence operation, U and S are user and
service embedding vectors, and X is the input vector.
An Adaptive Corrector (AC) that computes similarities
among user-location embedding and service-location
embedding is present. The AC employs a similarity approach,
including location similarities among consumers and services
into the neural network's forward propagation phase. As a
result, the AC bridges gaps across deep learning with CF. The
AC's operation outcome is immediately transferred to the
Output Nodes, bypassing the Intermediate Layer. The AC is
adaptable and may be used for all other similarity computations
like cosine and Euclidean similarity. Equations (8) and (9)
describe the procedure:
!" = cosider(�� , ��) =
�� . ��
$|��|$ ||��||
(8)
!" = cosider(�� , ��) = &�� . �� (9)
where !" is the AC similarity output. Cosine and Euclidean
similarities are both present in this. When no specific comment
is made, (8) is applied to determine the cosine similarity
outcome of the AC.
2) Intermediate Layer
The Intermediate Layer is responsible for processing the
input vectors from the Input Nodes in order to capture non-
linear information. The high-dimensional non-linear correlation
among consumers and services is learned using a densely
integrated MLP architecture in this work. First but foremost,
we must select the non-linear activation. The activation
function SeLU (Scaled Exponential Linear Unit) presents
numerous benefits. It allows building a map featuring qualities
that led to SNNs (Self-normalized Neural Networks).
Secondly, in order for the DNN to train and learn many
attributes, the network structure must adhere to the standard
tower architecture [20]. Lastly, to avoid over fitting, we employ
L2 regularization on weights. In the Intermediate Layer, the
forward propagating of the input vector sequence is described
by:
'(
)*+ = �((,(� - + �() (10)
'(
)*+
= ��(,�� �('./�
)*+ + ��), I = 3, 4,…,N -1 (11)
'012 = '3−1
012
(12)
where '6/�
)*+
is the ith layer of the Intermediate Layer’s output,
�� the Intermediate Layer bias term, and ')*+ the Intermediate
Layer's output.
3) Output Layer
The Output Layer is largely responsible for producing the
overall forecast results. The FRLS paradigm divides users and
services into two paths. We immediately combine the outcomes
of these two paths, as in [20, 21]. We mix the similarity
outcome ' !" with the output vector ')*+ to create a new
output vector sequence o. Finally, FRLS makes the final
projections using a single-layered neural network. Gaussian
distribution is used to initialize the parameters of this layer, as
illustrated in (13)-(14):
o = ɸ(' !" , ')*+ ) = � 7897:;<� (13)
�= �,� = �6(,6� ' + �6 ) (14)
where �>�,� is the predicted QoS values of user u raising the s
service, and �6 is the ordinary identity function that signifies
the activation function of the last layer.
We get a typical matrix factorization network base
Gaussian Matrix Factorization (GMF) distribution on one side
and an MLP network on the other.
B. Multi-Layer Perceptron Networks
An MLP network is simply a formal moniker for the most
fundamental type of DNN. An MLP network is made up of 5
dense hidden layers and the SeLu activation function. We
began by embedding the user and items, then flattened and
appended them. Afterwards, we added the first dropout layer,
followed by 4 hidden layers with 64 neurons, 32 neurons, 16
neurons, 8 neurons, and finally one densely integrated neuron.
Then a dropout and batch normalization layer was added
between hidden layers 1 and 2 and 2 and 3. Lastly, we got our
final output layer, which is nothing more than a one fully-
connected neuron. We utilized log cosh as the error function
and Nadam as optimizer.
C. Matrix Factorization Networks
A Gaussian MF networks is essentially a conventional
point-wise MF that uses Stochastic Gradient Decent (Adam
with this context) to approximate the factoring of a matrix
(user×item) into two matrices, i.e. user×latent feature and latent
feature×item.
D. Combining the two Networks: NeuMF
Finally, the Matrix factorization and MLP network are
merged into the NeuMF network. Essentially, this merely
appends the results of two networks. The two networks are then
fused, and a single output node is added to the newly integrated
model. From this, the configuration is like above, with a binary
log cosh loss and the Nadam optimizer.
Engineering, Technology & Applied Science Research Vol. 12, No. 5, 2022, 9186-9191 9189
www.etasr.com Rohilla et al.: An Empirical Framework for Recommendation-based Location Services Using …
To improve the accuracy of representing probability
distributions, many variants have been proposed. An alternative
approach for improvement is through the reconstruction loss:
f(t: a) =
�
?log(cosh(at)) =
�
?log
@ ABC @ DAB
( (15)
Nadam optimizer can be written as:
(16)
Nesterov is used by Nadam optimizer to modify the
gradients 1 step forward by substituting the prior m_vec (mt-1)
in (16) with the current m_vec (mt):
(17)
Algorithm 1: RLSD (Recommendation based location
services using the Deep Learning Algorithm.
Required: user_service request matrix (Rm), density of matrix
(Dm), DNN topology design (t), learning rates (lr), decayed
ratios (Dr), no. of iterations (Ni).
Output: Weighted matrix and the bias ��, ��, ,�, …, ,6 , ��,
�(, …, �6 .
Strategy:
step 1. Scarce Rm based on Dm
step 2. Make training inputs entries for EFG?.6 and test input
entries EF@�F
step 3. Create input attributes/features � , ��, �, ��;
step 4. Construct the DNN using t and (10)-(12)
step 5. Set �� , �� , ,� , …, ,6 according to the Gaussian
distribution;
step 6. Set ��, �(, …, �6 to 0;
step 7. For each epoch = 1 to Ni do
step 8. For each user and service in Rtrain do
step 9. Create embedding vectors sequence using (1)-(4);
step 10. Create input vector sequence using (5)-(7)
step 11. Create the AC output from either (8) or (9)
step 12. Make prediction, �>�,� using (13)-(14)
step 13. end for
step 14. Give lr and Dr to Nadam optimizer
step 15. Modify parameters of the model using the Nadam
minimization (15)
step 16. for each user and service in Rtest do
step 17. Use (20)-(21) to measure the performance of the
model
step 18. end for
step 19. end for
IV. RESULTS AND DISCUSSION
This section discusses the experimentation results for
location-based services on the WS-Dream dataset using Python
programming. To assess the effectiveness of the proposed
method, two evaluation measures are used.
A. Dataset
We ran tests upon the WS-Dream dataset which is a
substantial Web services dataset gathered and managed in [22],
which consists of 19,74,675 QoS values from 339 users on
5,825 services. This dataset includes information on users' and
services' locations. QoS data are presented in this work as a
user-service matrix, with the row indicator representing the
user identifier and the column indicator representing the
services identifier. This work utilized response time and
throughput as inputs to the RLSD.
B. Pre-processing
This research makes use of two spatially relevant dataset
features/attributes: Country Name (CN) and Autonomous
System Number (ASN). Users were spread throughout 30
nations and 136 ASNs, whereas services were spread across
990 ASNs across 73 countries, according to database figures.
For CN, we utilized Sklearn3's category encoding to convert
the classed characteristics into numerical embeddings, such that
each classified attribute is expressed by a nation code. The data
can be viewed like after pre-processing:
� = (�_�I, �JKL, �ML) (18)
K = (K_�I, KJKL, KML) (19)
where U and S are the user input embedding vector and service
input embedding vector correspondingly, U_ID is the user
identifier, UASN is an autonomous system numeric user code,
UCN is the nationwide user code, and S_ID, SASN, and SCN
are the same as above.
C. Parameter Setting
For CF methods (UPCC, LACF, IPCC, RegionKNN,
etc.),the best-k value is set to 10, the learning rate is fixed at
1e
-3
, the total of implicit feature factors (size) is fixed to 10, the
number of total iterations is fixed to 300, the regularization
variables are fixed to 0.1, and the randomized factor for the
rarefy matrix is fixed to 7. Deep learning techniques (NCF,
RLSD), can be developed using the Keras API, wherein the
Gaussian distribution function is utilized (average is 0, stdev is
0.01) to set the model hyper parameters and (15) to modify the
hyper parameters. Then we configured the batch size to 256,
learning rate to 1e
-4
, 4 MLPs, and Nadam optimizer.
D. Evaluation Metrics
The recommendation success of the specified approach is
measured using two fundamental statistical evaluation metrics:
Mean Absolute Error (MAE) and Root Mean Squared Error
(RMSE):
MAE =
∑ |O�,��,� / O⏞�,�|
Q (20)
RMSE =
�
Q
∑ (��,� − �⏞�,� )(�,� (21)
where ��,� is the QoS initial value for the user u raising the
service s, �⏞�,� is the QoS prediction value u raising the service
s, and N is the overall QoS value.
Tables I and II show the obtained results by the algorithms
performed in [21], with which the proposed model of this paper
is compared to. The values in the Tables show that the
proposed model is better in MAE and RMSE for response time
(RT) and TP (throughput).
Engineering, Technology & Applied Science Research Vol. 12, No. 5, 2022, 9186-9191 9190
www.etasr.com Rohilla et al.: An Empirical Framework for Recommendation-based Location Services Using …
TABLE I. RESPONSE TIME EXPERIMENTAL RESULTS
Approach
Density
0.5 0.10 0.15 0.20 0.25
MAE RMSE MAE RMSE MAE RMSE MAE RMSE MAE RMSE
UMEAN 0.876 0.853 0.873 1.857 0.874 1.857 0.873 1.858 0.874 1.858
IMEAN 0.703 0.567 0.686 1.542 0.684 1.533 0.681 1.529 0.680 1.525
UPCC 0.634 0.377 0.553 1.311 0.511 1.258 0.483 1.220 0.467 1.189
IPCC 0.633 1.397 0.591 1.341 0.507 1.258 0.454 1.208 0.431 1.175
UIPCC 0.624 1.386 0.579 1.328 0.498 1.247 0.448 1.197 0.425 1.165
RegionKNN 0.594 1.641 0.577 1.637 0.569 1.627 0.569 1.617 0.562 1.619
LACF 0.682 1.500 0.650 1.468 0.610 1.416 0.582 1.381 0.562 1.357
PMF 0.568 1.537 0.487 1.321 0.451 1.221 0.430 1.171 0.416 1.139
NCF 0.440 1.325 0.385 1.283 0.372 1.253 0.362 1.205 0.349 1.138
LDCF 0.246 0.7398 0.2787 0.895 0.2291 0.7573 0.2370 0.8004 0.2585 0.8819
RLSD (proposed) 0.1765 0.4977 0.1699 0.530 0.1892 0.6155 0.1831 0.5977 0.1710 0.5643
TABLE II. THROUGHPUT EXPERIMENTAL RESULTS
Approach
Density
0.5 0.10 0.15 0.20 0.25
MAE RMSE MAE RMSE MAE RMSE MAE RMSE MAE RMSE
UMEAN 54.333 110.296 53.947 110.345 53.971 110.201 53.906 110.190 53.862 110.194
IMEAN 27.342 65.844 26.962 64.843 26.757 64.266 26.669 64.069 26.595 63.873
UPCC 27.559 60.757 22.687 54.598 20.525 50.906 19.243 48.834 18.253 47.135
IPCC 27.102 62.665 26.270 60.479 25.487 57.561 23.726 54.564 22.286 52.293
UIPCC 27.070 60.510 22.440 54.506 20.256 50.585 18.888 48.238 17.863 46.392
RegionKNN 26.857 69.614 25.352 68.015 24.947 67.365 24.687 66.923 24.746 66.831
LACF 27.419 65.770 24.847 62.057 22.943 58.816 21.562 56.507 20.587 54.785
PMF 18.943 57.020 16.004 47.933 14.668 43.642 13.988 41.652 13.398 40.025
NCF 15.468 49.703 13.616 46.034 12.284 42.317 11.833 41.263 11.312 39.534
LDCF 17.629 57.746 14.711 50.730 13.952 48.871 13.199 47.323 12.332 45.142
RLSD (proposed) 19.517 61.822 16.154 53.344 14.958 50.681 14.388 49.393 14.332 49.195
Figures 2 and 3 show the loss values obtained by both
models in response time and throughput respectively. The
lesser the loss value, the more accurate the simulation (unless
the model was over fitted to the training set). The loss value
can be computed during training and testing and is used to
determine how good the model performs for the two different
sets. Loss, like accuracy, is also not a proportion. It is the
summarization of the errors incurred in training and test sets
with each sample. It is clearly visible that the proposed RLSD
model performed better than the existing methods by reducing
the loss values in both response time and throughput.
Fig. 2. Loss values of response time.
Fig. 3. Loss values of throughput.
V. CONCLUSION
A recommendation system is the backbone for retrieving
and processing information that aims to forecast user interests.
This study offers a novel deep learning-based approach called
RLSD to provide recommendation services to predict QoS
which incorporates a similarity AC to improve the prediction
values. In the future, we plan to analyze and solve additional
privacy concerns in location service recommendations, as well
as improve the performance, using a parallel and distributed
recommendations system to manage the required massive data.
This model may also improve the results of [15, 23-29].
REFERENCES
[1] Y. Wang, Z. Cai, Z.-H. Zhan, Y.-J. Gong, and X. Tong, "An
Optimization and Auction-Based Incentive Mechanism to Maximize
Social Welfare for Mobile Crowdsourcing," IEEE Transactions on
Computational Social Systems, vol. 6, no. 3, pp. 414–429, Jun. 2019,
https://doi.org/10.1109/TCSS.2019.2907059.
[2] Y. Zhang, Y. Zhou, F. Wang, Z. Sun, and Q. He, "Service
recommendation based on quotient space granularity analysis and
covering algorithm on Spark," Knowledge-Based Systems, vol. 147, pp.
25–35, May 2018, https://doi.org/10.1016/j.knosys.2018.02.014.
[3] S. Zhang, X. Li, Z. Tan, T. Peng, and G. Wang, "A caching and spatial
K-anonymity driven privacy enhancement scheme in continuous
location-based services," Future Generation Computer Systems, vol. 94,
pp. 40–50, May 2019, https://doi.org/10.1016/j.future.2018.10.053.
[4] S. Zhang, G. Wang, and Q. Liu, "A Dual Privacy Preserving Scheme in
Continuous Location-Based Services," in IEEE Trustcom/BigDataSE/
ICESS, Sydney, NSW, Australia, Aug. 2017, pp. 402–408,
https://doi.org/10.1109/Trustcom/BigDataSE/ICESS.2017.264.
[5] L. Qi, R. Wang, C. Hu, S. Li, Q. He, and X. Xu, "Time-aware distributed
service recommendation with privacy-preservation," Information
Engineering, Technology & Applied Science Research Vol. 12, No. 5, 2022, 9186-9191 9191
www.etasr.com Rohilla et al.: An Empirical Framework for Recommendation-based Location Services Using …
Sciences, vol. 480, pp. 354–364, Apr. 2019, https://doi.org/10.1016/j.ins.
2018.11.030.
[6] W. Dou, X. Zhang, J. Liu, and J. Chen, "HireSome-II: Towards Privacy-
Aware Cross-Cloud Service Composition for Big Data Applications,"
IEEE Transactions on Parallel and Distributed Systems, vol. 26, no. 2,
pp. 455–466, Oct. 2015, https://doi.org/10.1109/TPDS.2013.246.
[7] Y. Xu, L. Qi, W. Dou, and J. Yu, "Privacy-Preserving and Scalable
Service Recommendation Based on SimHash in a Distributed Cloud
Environment," Complexity, vol. 2017, Dec. 2017, Art. no. e3437854,
https://doi.org/10.1155/2017/3437854.
[8] A. Ramlatchan, M. Yang, Q. Liu, M. Li, J. Wang, and Y. Li, "A survey
of matrix completion methods for recommendation systems," Big Data
Mining and Analytics, vol. 1, no. 4, pp. 308–323, Dec. 2018,
https://doi.org/10.26599/BDMA.2018.9020008.
[9] Y. Zhang et al., "Covering-Based Web Service Quality Prediction via
Neighborhood-Aware Matrix Factorization," IEEE Transactions on
Services Computing, vol. 14, no. 5, pp. 1333–1344, Sep. 2021,
https://doi.org/10.1109/TSC.2019.2891517.
[10] I. M. Delamer and J. L. M. Lastra, "Service-Oriented Architecture for
Distributed Publish/Subscribe Middleware in Electronics Production,"
IEEE Transactions on Industrial Informatics, vol. 2, no. 4, pp. 281–294,
Aug. 2006, https://doi.org/10.1109/TII.2006.885188.
[11] B. AlBanna, M. Sakr, S. Moussa, and I. Moawad, "Interest Aware
Location-Based Recommender System Using Geo-Tagged Social
Media," ISPRS International Journal of Geo-Information, vol. 5, no. 12,
Dec. 2016, Art. no. 245, https://doi.org/10.3390/ijgi5120245.
[12] N. Kumar, A. Hashmi, M. Gupta, and A. Kundu, "Automatic Diagnosis
of Covid-19 Related Pneumonia from CXR and CT-Scan Images,"
Engineering, Technology & Applied Science Research, vol. 12, no. 1,
pp. 7993–7997, Feb. 2022, https://doi.org/10.48084/etasr.4613.
[13] A. B. S. Salamh and H. I. Akyuz, "A Novel Feature Extraction
Descriptor for Face Recognition," Engineering, Technology & Applied
Science Research, vol. 12, no. 1, pp. 8033–8038, Feb. 2022,
https://doi.org/10.48084/etasr.4624.
[14] M. Yu, G. Fan, H. Yu, and L. Chen, "Location-based and Time-aware
Service Recommendation in Mobile Edge Computing," International
Journal of Parallel Programming, vol. 49, no. 5, pp. 715–731, Oct.
2021, https://doi.org/10.1007/s10766-021-00702-5.
[15] N. Kumar and D. Aggarwal, "LEARNING-based Focused WEB
Crawler," IETE Journal of Research, pp. 1–9, Feb. 2021, https://doi.org/
10.1080/03772063.2021.1885312.
[16] Y. Yang, Z. Zheng, X. Niu, M. Tang, Y. Lu, and X. Liao, "A Location-
Based Factorization Machine Model for Web Service QoS Prediction,"
IEEE Transactions on Services Computing, vol. 14, no. 5, pp. 1264–
1277, Sep. 2021, https://doi.org/10.1109/TSC.2018.2876532.
[17] J. Liu, M. Tang, Z. Zheng, X. Liu, and S. Lyu, "Location-aware and
personalized collaborative filtering for web service recommendation,"
IEEE Transactions on Services Computing, vol. 9, no. 5, pp. 686–699,
Sep. 2016, https://doi.org/10.1109/TSC.2015.2433251.
[18] U. Khan, K. Khan, F. Hassan, A. Siddiqui, and M. Afaq, "Towards
Achieving Machine Comprehension Using Deep Learning on Non-GPU
Machines," Engineering, Technology & Applied Science Research, vol.
9, no. 4, pp. 4423–4427, Aug. 2019, https://doi.org/10.48084/etasr.2734.
[19] L. Zhang, T. Luo, F. Zhang, and Y. Wu, "A Recommendation Model
Based on Deep Neural Network," IEEE Access, vol. 6, pp. 9454–9463,
2018, https://doi.org/10.1109/ACCESS.2018.2789866.
[20] X. He, L. Liao, H. Zhang, L. Nie, X. Hu, and T.-S. Chua, "Neural
Collaborative Filtering," in 26th International Conference on World
Wide Web, Perth, Australia, Apr. 2017, pp. 173–182, https://doi.org/
10.1145/3038912.3052569.
[21] H.-J. Xue, X.-Y. Dai, J. Zhang, S. Huang, and J. Chen, "Deep matrix
factorization models for recommender systems," in 26th International
Joint Conference on Artificial Intelligence, Melbourne, Australia, Aug.
2017, pp. 3203–3209.
[22] Z. Zheng, H. Ma, M. R. Lyu, and I. King, "QoS-Aware Web Service
Recommendation by Collaborative Filtering," IEEE Transactions on
Services Computing, vol. 4, no. 2, pp. 140–152, Apr. 2011,
https://doi.org/10.1109/TSC.2010.52.
[23] N. Kumar, M. Gupta, D. Sharma, and I. Ofori, "Technical Job
Recommendation System Using APIs and Web Crawling,"
Computational Intelligence and Neuroscience, vol. 2022, Jun. 2022, Art.
no. e7797548, https://doi.org/10.1155/2022/7797548.
[24] N. Kumar, "Segmentation based Twitter Opinion Mining using
Ensemble Learning," International Journal on Future Revolution in
Computer Science & Communication Engineering, vol. 3, no. 9, pp. 1–9,
Sep. 2017.
[25] N. Kumar and P. Dahiya, "Weighted similarity page rank: an
improvement in WPR and WSR," International Journal of Computer
Engineering and Applications, vol. 11, no. 8, pp. 1–11, 2017.
[26] J. Mor, N. Kumar, and D. Rai, "Effective presentation of results using
ranking & clustering in meta search engine," COMPUSOFT, An
international journal of advanced computer technology, vol. 7, no. 12,
Art. no. 2957, 2018.
[27] J. Mor, D. D. Rai, and D. N. Kumar, "An XML based Web Crawler with
Page Revisit Policy and Updation in Local Repository of Search
Engine," International Journal of Engineering & Technology, vol. 7, no.
3, pp. 1119–1123, Jun. 2018, https://doi.org/10.14419/ijet.v7i3.12924.
[28] N. Kumar and P. Singh, "Meta Search Engine with Semantic Analysis
and Query Processing," International Journal of Computational
Intelligence Research, vol. 13, no. 8, pp. 2005–2013, 2017,
https://doi.org/10.37622/IJCIR/13.8.2017.2005-2013.
[29] N. Kumar, "Document Clustering Approach for Meta Search Engine,"
IOP Conference Series: Materials Science and Engineering, vol. 225,
Dec. 2017, Art. no. 012291, https://doi.org/10.1088/1757-899X/225/
1/012291.