Alumni

No	Nama	Email	judul Penelitian	Abstrack	Link Publish
1	ELFRIDA BA SIAHAAN	elfrida.siahaan@binus.ac.id	IMPROVING DETECTION SQL INJECTION ATTACKS USING REFERRER HEADER AND URI WEBLOG	Deteksi Injeksi SQL menyediakan kemampuan untuk memantau serangan Injeksi SQL pada situs web. Saat ini, para peneliti menggunakan Pembelajaran Mendalam untuk mendeteksi Injeksi SQL. Namun, deteksi ini memiliki keterbatasan, seperti False Positives (FP), False Negatives (FN) yang tinggi, dan Akurasi yang rendah karena deteksi Injeksi SQL hanya menggunakan data URI. Pada saat yang sama, serangan tidak hanya terjadi melalui URI tetapi juga Referrer. Oleh karena itu, penelitian ini bertujuan untuk menggunakan kombinasi URI dan Referrer untuk mendeteksi serangan dan membahas peningkatan kinerja model karena menambahkan Referrer. Langkah pertama penelitian ini adalah melakukan praproses dataset dan kemudian melakukan vektorisasi menggunakan Word2Vec. Metode Word2Vec dan CNN diusulkan menggunakan kombinasi URI dan Referrer, kemudian dibandingkan dengan CNN Word2Vec menggunakan URI. Hasil eksperimen menunjukkan bahwa metode yang diusulkan berkinerja lebih baik daripada metode lain dan mendapatkan akurasi lebih dari 99% dari payload dengan tingkat kesalahan yang rendah.	https://www.jatit.org/volumes/Vol101No18/29Vol101No18.pdf
2	ALFONSUS SUCAHYO HARIAJI	alfonsus.hariaji@binus.ac.id	ALGORITHMICALLY GENERATED MALICIOUS DOMAIN DETECTION USING N-GRAMS EMBEDDING AND ATTENTION-BASED BIDIRECTIONAL GATED RECURRENT UNIT	Botnets are one of the recent main cyber security threats. In order to avoid detection, botnets use Domain Generation Algorithm (DGA) to generate malicious domain names and maintain communication between infected bots and command and control server (C&C). Botnet malwares use various algorithm to generate domain names such as arithmetic, hashing, and wordlist/dictionary techniques. Recent traditional machine learning and deep learnin based detection methods need handcrafted domain name features which require more effort and advanced expertise and knowledge. This study aims to detect and classify DGA malicious domain without manually define and handcraft domain name features by only using the domain name. Ngrams method was used to create sequences of domain names and then vectorize the sequences using word embedding technique to create n-grams embedding model. After vectorization, Bidirectional Gated Recurrent Unit (BiGRU) was used for domain name classification and attention mechanism was used to improve classification performance by applying attention weight. The experiment results demonstrate the N-Grams Embedding and Attention-based BiGRU model proposed in this paper can detect and classify various type of DGA domains generated by arithmetic, hashing, and wordlist algorithm more effective compared to older algorithm such as CNN and LSTM for both DGA malicious domain detection and classification task. The use of attention mechanism can also improve the accuracy and performance of the DGA malicious domain detection model compared to models that do not use attention mechanism.	https://www.jatit.org/volumes/Vol101No18/31Vol101No18.pdf
3	Restu Herdian Herdian	restu.herdian001@binus.ac.id	The Implementation of hybrid methods in data mining for Predicting customer churn in the telecommunications sector	In recent years, the telecommunication industry is growth and become very competitive where has reached the point maintaining customer is very essential than acquiring new customer. And the two key factor for maintaining customer, the first is defining the segment of customer want to churn and the second is accuracy of predictive model. In this article we propose the hybrid model based on decision tree and artificial neural network (ANN) with the two stages of process to answer the problem of maintaining customer, the first is a segmentation phase with decision rules and the second is a prediction phase with artificial neural network (ANN). Our finding in benchmarked against the previous algorithms (decision tree and ANN) with the AUC metrics show the proposed model or hybrid achieves better accuracy and with the comprehensive information of what a drive customer churn	https://www.iocscience.org/ejournal/index.php/mantik/article/view/3706
4	Daniel Wilianto	daniel.wilianto@binus.ac.id	Automatic Short Answer Grading on High School’s E-Learning Using Semantic Similarity Methods	Grading students’ answers has always been a daunting task which takes a lot of teachers’ time. The aim of this study is to grade students’ answers automatically in a high school’s e-learning system. The grading process must be fast, and the result must be as close as possible to the teacher assigned grades. We collected a total of 840 answers from 40 students for this study, each already graded by their teachers. We used Python library sentencetransformers and three of its latest pre-trained machine learning models (all-mpnet-base-v2, alldistilroberta-v1, all-MiniLM-L6-v2) for sentence embeddings. Computer grades were calculated using Cosine Similarity. These grades were then compared with teacher assigned grades using both Mean Absolute Error and Root Mean Square Error. Our results showed that all-MiniLM-L6-v2 gave the most similar grades to teacher assigned grades and had the fastest processing time. Further study may include testing these models on more answers from more students, also fine tune these models using more school materials.	https://www.ceeol.com/search/article-detail?id=1103595
5	Fransisco Junius Amadeus	-	Extractive Text Summarization for Indonesian News Article Using Ant System Algorithm	The act of simplifying a text from its original source is known as text summarization. Instead of capturing the substance of the original content, an effective summary should be able to convey the information. Recent research on this form of extractive summarization has produced encouraging findings. A graphical model and a modified ant system method will be combined in this literature to provide a solution. The pheromone modification will decide which pertinent phrases will be selected to be a decent summary structure, while the modification process will focus on the point at which the graph construction will be built to represent an article. Additionally, a dataset (Indosum) including news stories that are often utilized in relevant research will be used in accordance to the summary in Indonesian. In addition, the ROUGE approach will be utilized as a tool for evaluation to rate the summary’s quality. Finally, this paper concludes with the challenges and future directions of text summarization.	https://www.jait.us/uploadfile/2023/JAIT-V14N2-295.pdf
6	William Harly	-	CNN-BERT for measuring agreement between argument in online discussion	Purpose With the rise of online discussion and argument mining, methods that are able to analyze arguments become increasingly important. A recent study proposed the usage of agreement between arguments to represent both stance polarity and intensity, two important aspects in analyzing arguments. However, this study primarily focused on finetuning bidirectional encoder representations from transformer (BERT) model. The purpose of this paper is to propose convolutional neural network (CNN)-BERT architecture to improve the previous method. Design/methodology/approach The used CNN-BERT architecture in this paper directly uses the generated hidden representation from BERT. This allows for better use of the pretrained BERT model and makes finetuning the pretrained BERT model optional. The authors then compared the CNN-BERT architecture with the method proposed in the previous study (BERT and Siamese-BERT). Findings Experiment results demonstrate that the proposed CNN-BERT is able to achieve a 71.87% accuracy in measuring agreement between arguments. Compared to the previous study that achieve an accuracy of 68.58%, the CNN-BERT architecture was able to increase the accuracy by 3.29%. The CNN-BERT architecture is also able to achieve a similar result even without further pretraining the BERT model. Originality/value The principal originality of this paper is the proposition of using CNN-BERT to better use the pretrained BERT model for measuring agreement between arguments. The proposed method is able to improve performance and also able to achieve a similar result without further training the BERT model. This allows separation of the BERT model from the CNN classifier, which significantly reduces the model size and allows the usage of the same pretrained BERT model for other problems that also did not need to finetune their BERT model.	https://www.emerald.com/insight/content/doi/10.1108/IJWIS-12-2021-0141/full/html
7	Rifqi Ramadhani Almassar	: rifqi.almassar@binus.ac.id	Detection of traffic congestion based on twitter using convolutional neural network model	Microblogging is a form of communication between users to socialize by describing the state of events in real-time. Twitter is a platform for microblogging. Indonesia is one of the countries with the largest Twitter users, people can share information about traffic jams. This research aims to detect traffic jams by extracting tweets in the form of vectors and then inserting them into the Convolution neural network (CNN) model and getting the best model from CNN+Word2Vec, CNN+FastText, and support vector machine (SVM). Data retrieval was conducted using the Rapidminer application. Then, the context of the tweets was checked so that there were 2777 data consisting of 1426 congestion road data and 1351 smooth road data. The data was taken from certain coordinate points in around Jakarta, Indonesia. Then, preprocessing and changes to vector form were carried out using the Word2Vec and FastText methods, then inserted into the CNN model. The results of CNN+Word2Vec and CNN+FastText were compared to the SVM method. The evaluation was done manually using the actual traffic conditions. The highest result obtained using test data by the CNN+FastText method are 86.33% while CNN+Word2Vec is 85.79% and SVM is 67.62%.	https://www.proquest.com/openview/bb67fbd4b4d710911fc5bd217c14fbcc/ 1?pq-origsite=gscholar&cbl=1686339
8	Suminar Ariwibowo	-	Hate Speech Text Classification Using Long Short-Term Memory (LSTM)	This paper aims to build hate speech text classification model by applying a combination of LSTM and FastText. The features of hate speech & non-hate speech, target hate speech, and categories of the hate speech. Dataset of those features taken from previous research by Okky Ibrohim. FastText word embeddings is used for formation of text vectors that will be used as input of the LSTM training model. The evaluation results obtained by getting the level of accuracy using confusion matrix. The accuracy value of text classification in this study is 83.52% on the classification of hate speech, 78.44% on the classification of target labels for hate speech, 82.75% on the classification of the label for category of hate speech.	https://ieeexplore.ieee.org/abstract/document/10034908/metrics#metrics
9	Dewa Bagus Gde Khrisna Jayanta Nugraha	-	Age Classification of Moviegoers Based on Facial Image Using Deep Learning	Jumlah penonton bioskop di Indonesia terus meningkat dari tahun ke tahun hingga tahun 2019. Namun, akibat pandemi COVID-19, sebagian besar bioskop di Indonesia tutup pada awal tahun 2020. Penonton bioskop semakin beralih ke platform digital untuk menonton film. Berdasarkan film yang ditayangkan, dapat dibagi menjadi tiga kategori, yaitu film anak-anak, film remaja, dan film dewasa. Diperlukan suatu sistem yang dapat secara otomatis mengklasifikasi wajah penonton berdasarkan kategori usianya. Dengan menggunakan Deep Learning, penelitian ini bertujuan untuk mengklasifikasi usia penonton berdasarkan foto wajah. Tahap pertama melibatkan pengumpulan data dari tiga dataset: All-Age-Face, FaceAge, dan FGNET, yang kemudian digabungkan dan diberi label ulang berdasarkan kelompok usia. Preprocessing dan pengujian hyperparameter juga dilakukan. Menemukan learning rate dan layer bottleneck terbaik adalah tujuan dari pengujian hyperparameter. Proses pelatihan menggunakan learning rete dan dua layer bottleneck terbaik dengan enam model, yaitu MobileNet, MobileNetV2, VGG16, VGG19, Xception, dan ResNet101V2. Global Average Pooling ditambahkan pada akhir layer di setiap model. Model MobileNet pada dua layer bottleneck menghasilkan nilai akurasi pengujian terbaik sebesar 85,44 persen dalam penelitian ini.	https://www.temjournal.com/content/113/TEMJournal August2022_1406_1415.pdf