Welcome to the Open Access System!
OpenAccess@IKU is the Academic Open Access System of Istanbul Kultur University. It was established in June 2014 to digitally store and open access the academic outputs of Istanbul Kultur University in international standards. OpenAccess@IKU includes academic outputs such as articles, presentations, thesis, books, book chapters, reports produced within the body of Istanbul Kultur University.
Recent Submissions
Psychological Assistant: Assessing The Emotional State of a Patient Using Triple Video Analysis Method
(Institute of Electrical and Electronics Engineers Inc., 2024) ALKAN, MUSTAFA; ELMASRY, WİSAM
The 'Psychological Assistant' presents a groundbreaking approach to remote emotion assessment by integrating video analysis techniques, computer vision, speech recognition, and Natural Language Processing (NLP). Leveraging pre-trained models such as Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), Artificial Neural Networks (ANNs), and advanced NLP algorithms, the system analyzes facial expressions, voice signals, and Text Classification to provide mental health practitioners with comprehensive insights during remote consultations. Through the use of methods such as averaging and combining over time, the system ensures a thorough emotion evaluation, promising high accuracy and reliability in mood identification. This innovative integration of NLP enhances the system's capability to understand and interpret textual cues, allowing for a more holistic assessment of patients' mental states. The technology holds the potential for early intervention, personalized treatment plans, and an elevated standard of care in remote mental health services, representing a significant advancement in digital healthcare solutions. © 2024 IEEE.
AI-based Multimodal Resume Ranking Web Application for Large Scale Job Recruitment
(Institute of Electrical and Electronics Engineers Inc., 2024) YAZICI, MEHMET BATUHAN; SABAZ, DAMLA; ELMASRY, WİSAM
This paper presents a resume-ranking web application that improves recruitment through advanced deep-learning techniques. The system uses the YOLOv9 model fine-tuned with our newly created custom dataset for segment detection on resumes of various structures, EasyOCR for text recognition, mBERT fine-tuned for text classification, and GLiNER for named entity recognition with regular expressions. These models and techniques efficiently extract, categorize, and match resume information with job descriptions. We created a custom dataset for our object detection training, and while we trained three models, YOLOv9 achieved the highest performance with a score of 0.84 mAP. Our hybrid matching approach provides highly accurate and relevant resume rankings using the embedding model, gte-large-en-v1.5, and cosine similarity for semantic matching with dense vectors with extracted keywords and BM25 for keyword relevance. The web application allows HR professionals to upload resumes seamlessly, define job descriptions, and view ranked results, providing a tailored solution to specific recruitment needs. Although we faced challenges such as text extraction accuracy and zero-shot NER limitations, our system demonstrated a solid overall performance. This paper demonstrates the potential of state-of-the-art deep learning models to enhance recruitment processes and provides a valuable tool for HR professionals to identify the most suitable candidates efficiently. © 2024 IEEE.
Transformative Approaches to Customer Sentiment Analysis and Customer Feedback Scoring in CRM Platforms
(Institute of Electrical and Electronics Engineers Inc., 2024) Cevik, Rabia; Celik, Ahmet Erkan; AKBULUT, AKHAN
This study introduces an innovative system designed to predict customer satisfaction scores through the integration of sentiment analysis of customer feedback alongside all related factors from a Customer Relationship Management (CRM) system. The system implements the latest transformer models like BERT and RoBERTa then assess customer sentiment using an ensemble learning voting mechanism for accurate sentiment classification, and adaptive customer satisfaction rating. The model generates baseline scores dynamically, based on factors like customer loyalty, and frequency of interactions with the firm, thus enhancing accuracy and relevance when assessing satisfaction. The system is also developed to utilize Turkish data optimizing usage in market shares for firms serving that user group. Empirical results indicate that the ensemble learning approach significantly improves the accuracy of sentiment analysis and the reliability of satisfaction quantification. This resource provides additional contribution to the CRM literature by providing a credible and scalable mechanism to assess customer satisfaction to potentially be implemented in practice across industries. Future work will focus on extending the system's scalability and enhancing its predictive capabilities across diverse sectors. © 2024 IEEE.
Edge Information Assisted Decoder for Business Process Anomaly Detection
(Institute of Electrical and Electronics Engineers Inc., 2024) Ayaz, Teoman Berkay; Ozcan, Alper; AKBULUT, AKHAN
Anomaly detection as a subject focuses on the identification of data point which significantly deviate from what is the norm or the standard of the dataset. This gives anomaly detection a wide range of applications where the detection of irregularities is often times of crucial importance such as Business Process Management (BPM). In this study we present a novel type of decoder referred to as 'Edge Information Assisted Decoder' (EIAD), working on graph data to incorporate edge indexes and attributes into the decoding to achieve improved anomaly detection. We tested a total of 8 encoder-decoder combinations to comparatively evaluate them and prove the effectiveness of the proposed method. The proposed method and the best encoder-decoder combination, the graph attention network (GAT) encoder and the edge-conditioned convolution (ECC) decoder yielded an increase of 0.31 in F1-score from 0.32 to 0.63 when compared to the baseline multi-layer perceptron (MLP) decoder model, both with the optimal optimizer. The empirical results show that the proposed approach has a potential to improve graph based anomaly detection. © 2024 IEEE.
Secure, Robust and Optimized Algorithm: Towards Enhancing Digital Image Watermarking
(Institute of Electrical and Electronics Engineers Inc., 2024) ULUTÜRK, CEYDA; AKDENİZ, FİDAN; VAROL, MELİKE; ELMASRY, WİSAM
This paper introduces a robust, secure, and optimized digital watermarking methodology that employs three distinct algorithms, each designed to enhance the security and robustness of embedding watermarks in digital images. The first algorithm ensures data integrity with CRC-32 checksums, compresses data using Gzip, encrypts with AES, and embeds watermarks through Least Significant Bits (LSB) coupled with the Fisher Yates Shuffle algorithm. The second algorithm adopts QR for data integrity, Zlib for data compression, DES for encryption, and Discrete Cosine Transform (DCT) for embedding. The third algorithm combines CRC-32 for data integrity and Gzip for compression with AES encryption and LSB embedding, enhancedby Particle Swarm Optimization (PSO) to optimize embedding parameters. The effectiveness of these algorithms is assessed using a comprehensive set of image quality metrics, including Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Normalized Cross-Correlation (NC), Average Difference (AD), Structural Content (SC), Maximum Difference (MD), Local Mean Square Error (LMSE), and Normalized Absolute Error(NAE). Furthermore, the resilience of these algorithms against common attacks is analyzed through Histogram Analysis, LSB Attack, and Chi-Square Analysis. This paper aims to improve digital image watermarking by integrating advanced encryption, compression, and optimization techniques, addressing crucial challenges in data protection and integrity. © 2024 IEEE.
