Wednesday, October 21, 2020

 
These articles have been peer-reviewed and accepted for publication in JICT, but are pending final changes, are not yet published and may not appear here in their final order of publication until they are assigned to issues. Therefore, the content conforms to our standards but the presentation (e.g. typesetting and proof-reading) is not necessarily up to the JICT standard. Additionally, titles, authors, abstracts and keywords may change before publication.
 

 
SARCASM DETECTION IN PERSIAN
1Zahra Bokaee Nezhad & 2Mohammad Ali Deihimi
1Department of Computer Engineering, Zand University, Iran
2Department of Electronics Engineering, Bahonar University, Iran 
zbokaee@gmail.com; m.a.deihimi@gmail.com
 
 
ABSTRACT
 
Sarcasm is a form of communication where the individual states the opposite of what is implied. Hence, detecting a sarcastic tone is somewhat complicated due to its ambiguous nature. On the other hand, identification of sarcasm is vital to various Natural Language Processing (NLP) tasks such as sentiment analysis and text summarization. However, research on sarcasm detection in Persian is very limited. Therefore, we investigate sarcasm detection technique on Persian tweets by combining deep learning-based and machine learning-based approaches. We propose four sets of features that cover different types of sarcasm. They are deep polarity feature, sentiment feature, part of speech feature and punctuation feature. We use these features to classify tweets as sarcastic and non-sarcastic. In this study, the deep polarity feature is proposed by conducting sentiment analysis using deep neural network architecture. In addition, to extract sentiment feature we decide to create a Persian sentiment dictionary which consists of four sentiment categories. We also provide a new Persian proverb dictionary using in the preparation step to enhance the accuracy of the proposed model. The performance of our model is analyzed using several standard machine learning algorithms. The results of the experiment show our method outperforms the baseline method and reaches an accuracy of 80.82%. We also study the importance of each set of proposed features and evaluate its added value to the classification.
 
Keywords: Sarcasm Detection, Natural Language Processing, Machine Learning, Sentiment Analysis, Classification.
  

 
PARAMETRIC FLATTEN-T SWISH: AN ADAPTIVE NON-LINEAR ACTIVATION FUNCTION FOR DEEP LEARNING
1Hock Hung Chieng, 1Noorhaniza Wahid & 2Pauline Ong
1Faculty of Information Technology and Computer Science, Universiti Tun Hussein Onn Malaysia, Malaysia
2Faculty of Mechanical and Manufacturing Engineering, Universiti Tun Hussein Onn Malaysia, Malaysia
hi160029@siswa.uthm.edu.my; nhaniza@uthm.edu.my; ongp@uthm.edu.my
 
 
ABSTRACT
 
Activation function is a key component in deep learning that performs non-linear mappings between the inputs and outputs. Rectified Linear Unit (ReLU) has been the most popular activation function across the deep learning community. However, ReLU contains several shortcomings that can result in inefficient training of the deep neural networks, these are: 1) the negative cancellation property of ReLU tends to treat negative inputs as unimportant information for the learning, hence resulting in a performance degradation; 2) the inherent predefined nature of ReLU is unlikely to promote additional flexibility, expressivity, and robustness to the networks; 3) the mean activation of ReLU is highly positive and leading to bias shift effect in network layers; 4) the multilinear structure of ReLU restricts the non-linear approximation power of the networks. To tackle these shortcomings, this paper introduces a Parametric Flatten-T Swish (PFTS) as an alternative to ReLU. By taking ReLU as a baseline method, the experiments shown PFTS improved classification accuracy on SVHN dataset by 0.31%, 0.98%, 2.16%, 17.72%, 1.35%, 0.97%, 39.99% and 71.83% on DNN-3A, DNN-3B, DNN-4, DNN-5A, DNN-5B, DNN-5C, DNN-6 and DNN-7, respectively. Besides, PFTS also achieved the highest mean rank among the comparison methods. The proposed PFTS manifests higher non-linear approximation power during training and thereby improve the predictive performance of the networks.
 
Keywords: activation function, deep learning, Flatten-T Swish, non-linearity, ReLU.
 

 
DYNAMIC PROBABILITY SELECTION FOR FLOWER POLLINATION ALGORITHM BASED ON METROPOLIS-HASTINGS LIKE CRITERIA

1Kamal Zuhairi Zamli, 1,2Fakhrud Din, 1Abdullah B. Nasser, 3Nazirah Ramli, and 3Noraini Mohamed
1Faculty of Computer Systems & Software Engineering, Universiti Malaysia Pahang, Malaysia
2 Department of Computer Science & IT, University of Malakand, Pakistan
3Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA Pahang, Malaysia

kamalz@ump.edu.my; fakhruddin@uom.edu.pk; abdullahnasser83@gmail.com.my
nazirahr, noraini_mohamed@uitm.edu.my

 

ABSTRACT

Flower pollination algorithm (FPA) is a relatively new meta-heuristic algorithm which adopts its metaphor from the proliferation role of flowers in plants. Having only one parameter control (i.e. the switch probability (pa) to choose from the global search (i.e. exploration) and local search (i.e. exploitation) is the main strength of FPA as compared to other meta-heuristic algorithms. However, FPA still suffers from variability of its performance as there is no one size fits all value for pa, depending on the characteristics of the optimization function. This paper proposes Flower Pollination Algorithm Metropolis-Hastings (FPA-MH) based on the adoption of Metropolis-Hastings criteria adopted from Simulated Annealing (SA) algorithm to enable dynamic selection of the pa probability. Adopting the problem of t-way test suite generation as our case study, the comparative evaluation with the original FPA, FPA-MH gives promising results owing to its dynamic and adaptive selection of search operators based on the need of the current search.
 
Keywords: Dynamic Probability Selection, Flower Pollination Algorithm, Optimization, t-way testing, Data mining.
 

 
AN INTELLIGENT SOFTWARE DEFINED NETWORKS CONTROLLER COMPONENT TO DETECT AND MITIGATE DENIAL OF SERVICE ATTACKS
Huseyin Polat & Onur Polat
Department of Computer Engineering, Gazi University, Turkey
 

ABSTRACT

Despite many advantages of Software Defined Networks (SDN) such as manageability, scalability and performance, it has inherent security threats. In particular, Denial of Service (DoS) attacks are major threats to SDN. The controller's processing and communication abilities are overwhelmed by DoS attacks. The capacity of the flow tables in the switching device is exhausted due to excess flows created by the controller because of malicious packets. DoS attacks on the controller cause the network performance to drop to a critical level. In this paper, a new SDN controller component was proposed to detect and mitigate DoS attacks in SDN controller. POX layer three controller component was used for underlying a testbed for PacketIn messages. Any packet from host is incremented to measure the rate of packet according to its device identification (i.e. dpid) and its input port number. Considering the rate of packets received by controller and threshold set, malicious can be detected and mitigated easily. A developed controller component was tested in Mininet simulation environment with an hping3 tool to build artificial DoS attacks. Using our enhanced controller component, DoS packets were prevented from accessing the controller and thus data plane (switching devices) was prevented from being filled with unwanted flows.

Keywords: Security, DoS Attack, Decision Making, Software Defined Networks, POX Controller.
 

 
VALIDATION ASSESSMENTS ON RESAMPLING METHOD IN IMBALANCED BINARY CLASSIFICATION FOR LINEAR DISCRIMINANT ANALYSIS

1Jamaluddin Hakiim & 2Nor Idayu Mahat
1Department of Mathematics, Universiti Putra Malaysia, Malaysia
2Centre for Testing, Measurement and Appraisal, Universiti Utara Malaysia, Malaysia

ahmadhakiimjamaluddin@gmail.com; noridayu@uum.edu.my

 

ABSTRACT

The curse of class imbalance affects the performance of many conventional classification algorithms including linear discriminant analysis (LDA). Data pre-processing approach through some resampling methods such as random oversampling (ROS) and random undersampling (RUS) is one of the treatments to alleviate such curse. There have been studies which attempted to address the effect of a resampling method on the performance of LDA. However, some based on different performance measures as well as validation strategies contradicted with each other. This manuscript attempts to shed more lights on the effect of a resampling method (ROS or RUS) on the performance of LDA based on true positive rate (TPR) and true negative rate (TNR) through five validation strategies i.e. leave-one-out cross-validation (loocv), kfcv, repeated k-fold cross-validation (rkfcv), naive bootstrap (B) and .632+ bootstrap (B632). 100 two-group bivariate normally distributed simulated and four real data sets with severe class imbalance ratio have been utilised. The analysis on the location and dispersion statistics of the performance measures further enlightened on (i) the effect of a resampling method on the performance of LDA, and (ii) the enhancement in the learning fairness of LDA on objects regardless of sample sizes hence reducing the effect of the curse of class.

KeywordsLinear discriminant analysis, pre-processing, resampling method, class imbalance, binary classification.
 
 

Universiti Utara Malaysia Press
 Universiti Utara Malaysia, 06010 UUM Sintok
Kedah Darul Aman, MALAYSIA
Phone: +604-928 4816, Fax : +604-928 4792

Creative Commons License

All articles published in Journal of Information and Communication Technology (JICT) are licensed under a Creative Commons Attribution 4.0 International License.