Article
Computer Science, Artificial Intelligence
Samia Boukir
Summary: This study investigates the use of margin and diversity, two key concepts in ensemble learning, to develop a versatile uncertainty-driven ensemble classifier under the scarcity of labeled data. New semi-supervised definitions are proposed for both margin and diversity, and new robust ensemble metrics are introduced to strengthen the semi-supervised classification scheme. The relevance of these new criteria is examined in change detection experiments, and the underlying fusion rule significantly improves the change detection performance.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Danyang Li, Zhuhong Zhang, Guihua Wen
Summary: Ensemble pruning improves system performance and reduces storage requirements in integration systems. Most approaches evaluate the competence and relationships of classifiers by analyzing their predictions to remove low-quality or redundant classifiers. However, finding the best way to represent classifiers and create ensemble diversity remains a research problem. To address this, we propose a new classifier selection method called CRCEEP, which incorporates two new classifier representation learning methods and a clustering ensemble method. Extensive experiments on UCI datasets demonstrate the effectiveness of CRCEEP and the importance of classifier representation.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Youwei Wang, Jiangchun Liu, Lizhou Feng
Summary: Ensemble learning is widely used in text classification field to construct strong classifiers. A text length considered adaptive Bagging ensemble learning algorithm (TC_Bagging) is proposed to improve text classification accuracy. It compares different deep learning methods in processing long and short texts, constructs optimal base classifier groups, and uses an adaptive threshold group based random sampling method to train text sample subsets of different lengths. The algorithm combines the smooth inverse frequency based text vector generation algorithm with the traditional weighted voting classifier ensemble method to achieve better classification performance than baseline methods.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Lin Mu, Peiquan Jin, Yiwen Zhang, Hong Zhong, Jie Zhao
Summary: This paper proposes a self-supervised learning method to recognize synonyms in short texts on social networks. The method involves generating pseudo-labels using a clustering algorithm and obtaining feature representations using a deep-learning model. Experimental results demonstrate the effectiveness of the proposed method.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Tianshu Yang, Nicolas Pasquier, Frederic Precioso
Summary: A novel semi-supervised consensus clustering algorithm is proposed in this article, which utilizes closed pattern mining technique to generate a recommended consensus solution without the need for inputting the number of generated clusters k, and can improve the quality of clustering results.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Information Systems
Ammar Mohammed, Rania Kora
Summary: Deep learning-based models have outperformed classical machine learning models in various text classification tasks in the past decade. However, finding the most suitable deep learning classifier remains a challenge. This study proposes a new meta-learning ensemble method that combines baseline deep learning models using 2-tiers of meta-classifiers to improve classification performance. Experimental results demonstrate that the proposed method significantly enhances the accuracy of baseline deep models and outperforms state-of-the-art ensemble methods.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Automation & Control Systems
Baohua Shen, Juan Jiang, Feng Qian, Daoguo Li, Yanming Ye, Gholamreza Ahmadi
Summary: This paper proposes an AHC-based ensemble semi-supervised clustering algorithm (SSEHCCI) to improve performance by combining the results of several output partitions and utilizing constraints information for semi-supervised clustering. Experimental results show that SSEHCCI outperforms existing semi-supervised algorithms on some UCI datasets.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Murat Dener, Sercan Gulburun
Summary: A model combining supervised and unsupervised learning algorithms is proposed to predict malware using clustering to enhance the performance of supervised classifiers. The model achieves high accuracy and f1 scores on the BODMAS, EMBER 2018, and Kaggle datasets. The tiered positioning of classifiers significantly reduces prediction time.
CMC-COMPUTERS MATERIALS & CONTINUA
(2023)
Article
Computer Science, Artificial Intelligence
Jovan Chavoshinejad, Seyed Amjad Seyedi, Fardin Akhlaghian Tab, Navid Salahian
Summary: Semi-supervised nonnegative matrix factorization combines the strengths of matrix factorization in learning part-based representation and can achieve high learning performance with limited labeled data and a large amount of unlabeled data. Recent research focuses on utilizing self-supervised learning to enhance semi-supervised learning. This paper proposes an effective Self-Supervised Semi-Supervised Nonnegative Matrix Factorization (S4NMF) model that directly extracts a consensus result from ensembled NMFs with similarity and dissimilarity regularizations. Experimental results on standard benchmark datasets demonstrate the effectiveness of the proposed model in semi-supervised clustering.
PATTERN RECOGNITION
(2023)
Article
Chemistry, Analytical
Whanhee Cho, Yongsuk Choi
Summary: This paper introduces a semi-supervised learning approach for text classification called LMGAN, based on generative adversarial networks (GAN). LMGAN utilizes BERT and GAN-BERT, multiple generators, and hidden layer outputs to enrich the distribution of fake data, addressing the limitation of early GAN-based methods in generating high-quality fake data. Experimental results demonstrate that LMGAN achieves better performance even with limited amounts of labeled data.
Article
Computer Science, Artificial Intelligence
Tianyi Luo, Yang Liu
Summary: The wisdom of the crowd and ensemble learning have shown that the majority voting answer is usually more accurate. This paper introduces a new method, called Machine Truth Serum, that leverages machine learning algorithms to reveal when the minority has the true answer, leading to improved classification performance.
Article
Computer Science, Artificial Intelligence
Tuanfei Zhu, Cheng Luo, Zhihong Zhang, Jing Li, Siqi Ren, Yifu Zeng
Summary: This paper introduces a structure-preserving Oversampling method for high-dimensional imbalanced time series classification, OHIT, and integrates it into boosting framework to form a new ensemble algorithm OHITBoost. Extensive experiments on several publicly available time-series datasets demonstrate their effectiveness.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Information Systems
Ruby Rani, D. K. Lobiyal
Summary: This paper attempts to construct corpus specific stopwords lists for Hindi text documents using statistical and knowledge-based methods, and proposes an evaluation method to examine their behavior using text mining models.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Environmental Sciences
Hamid Jafarzadeh, Masoud Mahdianpari, Eric Gill, Fariba Mohammadimanesh, Saeid Homayouni
Summary: The study investigates the capability of different ensemble learning algorithms for satellite image classification, with XGBoost showing superior performance in multispectral, hyperspectral, and Polarimetric Synthetic Aperture Radar (PolSAR) data classification.
Article
Computer Science, Information Systems
Xiang Ge, Xuexiang Yu, Xu Yang
Summary: This paper proposes a novel hyperspectral image classification method that combines dynamic semi-supervised multiple-kernel collaborative representation ensemble selection with superpixel (SP) consistency constraints. The method divides the image into different SP blocks and treats each block as an independent classification task. It selects high-confidence samples from unlabeled data and assigns pseudo-labels to expand the training sample set, and uses a multiple-kernel collaborative representation classifier as the base classifier to improve classification performance.
Article
Computer Science, Interdisciplinary Applications
Aytug Onan
Summary: The study evaluated the predictive performance of conventional supervised learning methods, ensemble learning methods, and deep learning methods, as well as the efficiency of text representation and word-embedding schemes in sentiment analysis on MOOC evaluations. Analysis of a corpus containing 66,000 MOOC reviews indicated that deep learning-based architectures outperformed other methods for sentiment analysis on educational data mining. The highest predictive performance was achieved by long short-term memory networks combined with GloVe word-embedding scheme-based representation, with a classification accuracy of 95.80%.
COMPUTER APPLICATIONS IN ENGINEERING EDUCATION
(2021)
Article
Computer Science, Interdisciplinary Applications
Aytug Onan, Mansur Alp Tocoglu
Summary: The study aims to use weighted word embeddings and clustering techniques to cluster MOOC discussion forum posts and identify question topics. By evaluating four word-embedding schemes, four weighting functions, and four clustering algorithms, it is found that weighted word-embedding schemes combined with clustering algorithms outperform conventional schemes.
COMPUTER APPLICATIONS IN ENGINEERING EDUCATION
(2021)
Article
Computer Science, Software Engineering
Aytug Onan
Summary: A deep learning architecture combining TF-IDF-weighted Glove word embedding with CNN-LSTM architecture outperforms conventional deep learning methods in sentiment analysis of product reviews from Twitter.
CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE
(2021)
Article
Engineering, Multidisciplinary
Hasan Bulut, Aytug Onan, Serdar Korukoglu
SADHANA-ACADEMY PROCEEDINGS IN ENGINEERING SCIENCES
(2020)
Article
Computer Science, Information Systems
Aytug Onan
Summary: This paper proposes a bidirectional convolutional recurrent neural network architecture for sentiment analysis, which utilizes bidirectional LSTM and GRU layers to extract past and future contexts, and employs a group-wise enhancement mechanism to strengthen important features and weaken less important ones. Experimental results demonstrate that this architecture achieves state-of-the-art performance in sentiment analysis.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Gul Cihan Habek, Mansur Alp Tocoglu, Aytug Onan
Summary: With the growth of the cryptocurrency trading market, sentiment analysis of cryptocurrency comments has become crucial. A novel deep neural network architecture was introduced for sentiment classification, showing an accuracy of 93.77% in experimental results.
APPLIED ARTIFICIAL INTELLIGENCE
(2022)
Article
Engineering, Electrical & Electronic
Mahmut Agrali, Volkan Kilic, Aytug Onan, Esra Meltem Koc, Ali Murat Koc, Rasit Eren Buyuktoka, Turker Acar, Zehra Adibelli
Summary: The conventional approach for identifying GGO in medical imaging is CNN, which shows promising performance in COVID-19 detection. However, CNN has limitations in capturing the structured relationships of GGO. This paper proposes a novel framework called DeepChestNet that leverages structured relationships by performing segmentation and classification on lung, pulmonary lobe, and GGO, leading to enhanced detection and diagnosis of COVID-19.
INTERNATIONAL JOURNAL OF IMAGING SYSTEMS AND TECHNOLOGY
(2023)
Article
Computer Science, Information Systems
Aytug Onan
Summary: We propose a novel hierarchical graph-based text classification framework that leverages contextual node embedding and BERT-based dynamic fusion to capture complex relationships between nodes in the hierarchy. The framework consists of seven stages: Linguistic Feature Extraction, Hierarchical Node Construction, Contextual Node Embedding, Multi-Level Graph Learning, Dynamic Text Sequential Feature Interaction, Attention-Based Graph Learning, and Dynamic Fusion with BERT. Experimental results on benchmark datasets demonstrate significant improvements in classification accuracy compared to state-of-the-art methods.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2023)
Article
Computer Science, Information Systems
Aytug Onan
Summary: The process of creating high-quality labeled data is crucial but time-consuming. This paper proposes a text augmentation framework called SRL-ACO that leverages Semantic Role Labeling and Ant Colony Optimization techniques to enhance the accuracy of natural language processing models without requiring manual data annotation.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2023)
Proceedings Paper
Acoustics
Ozge Taylan Moral, Volkan Kilic, Aytug Onan, Wenwu Wang
Summary: Describing the semantic content of an image through natural language has attracted significant interest in computer vision and language processing. Existing image captioning approaches have limitations in generating accurate captions due to their inability to effectively use visual information. This paper proposes an improved method using multi-layer GRU to enhance the semantic coherence of captions, and experimental results demonstrate its superiority.
2022 30TH EUROPEAN SIGNAL PROCESSING CONFERENCE (EUSIPCO 2022)
(2022)
Proceedings Paper
Acoustics
Ozkan Cayli, Volkan Kilic, Aytug Onan, Wenwu Wang
Summary: Image captioning is the task of generating descriptive captions for visual content using natural language automatically. Recent advancements in deep neural networks have improved the generation of natural and semantic text in image captioning. However, maintaining gradient flow between neurons in consecutive layers becomes challenging with deeper networks. In this paper, the authors propose integrating an auxiliary classifier into the residual recurrent neural network to enhance caption generation by enabling gradient flow to reach bottom layers. Experiments on MSCOCO and VizWiz datasets demonstrate the superiority of the proposed approach over state-of-the-art methods in multiple performance metrics.
2022 30TH EUROPEAN SIGNAL PROCESSING CONFERENCE (EUSIPCO 2022)
(2022)
Proceedings Paper
Cell & Tissue Engineering
Duygu Degirmenci, Mehmet Akif Ozdemir, Onan Guren, Aytug Onan
Summary: This study aims to improve classification performance of EEG signals for MI tasks by extracting discriminative features with NMF from TFD obtained by WSST, achieving outstanding accuracy, kappa, and F1 score with various classifiers. WSST provides energy distributions with highly localization capability in TFD, offering a promising approach for MI task classification.
2022 MEDICAL TECHNOLOGIES CONGRESS (TIPTEKNO'22)
(2022)
Proceedings Paper
Engineering, Electrical & Electronic
Rumeysa Keskin, Ozge Taylan Moral, Volkan Kilic, Aytug Onan
Summary: The study introduces an automatic image captioning system for smartphones, utilizing advanced visual information and decoder structure to generate more meaningful image descriptions. The system performs well on the MSCOCO dataset and is integrated into a custom Android application, IMECA, for offline caption generation.
29TH IEEE CONFERENCE ON SIGNAL PROCESSING AND COMMUNICATIONS APPLICATIONS (SIU 2021)
(2021)
Article
Computer Science, Information Systems
Aytug Onan, Mansur A. L. P. Tocoglu
Summary: This research aims to present an effective sarcasm identification framework on social media data by utilizing neural language models and deep neural networks. The model includes a three-layer stacked bidirectional long short-term memory architecture and introduces an inverse gravity moment based term weighted word embedding model to preserve word-ordering information. The presented model achieves promising results with a classification accuracy of 95.30% for the sarcasm identification task.
Article
Computer Science, Artificial Intelligence
Aytug Onan, Mansur Alp Tocoglu
TURKISH JOURNAL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCES
(2020)
