Browsing by Author "Gorur, Abdul Kadir"

Now showing 1 - 3 of 3

Citation - WoS: 1
Citation - Scopus: 4
A Comparative Evaluation of Popular Search Engines on Finding Turkish Documents for A Specific Time Period
(Univ Osijek, Tech Fac, 2017) Gorur, Abdul Kadir; Bitirim, Yiltan; 107251; 06.01. Bilgisayar Mühendisliği; 06. Mühendislik Fakültesi; 01. Çankaya Üniversitesi
This study evaluates the popular search engines, Google, Yahoo, Bing, and Ask, on finding Turkish documents by comparing their current performances with their performances measured six years ago. Furthermore, the study reveals the current information retrieval effectiveness of the search engines. First of all, the Turkish queries were run on the search engines separately. Each retrieved document was classified and precision ratios were calculated at various cut-off points for each query and engine pair. Afterwards, these ratios were compared with the six years ago ratios for the evaluations. Besides the descriptive statistics, Mann-Whitney U and Kruskal-Wallis H statistical tests were used in order to find out statistically significant differences. All search engines, except Google, have better performance today. Bing has the most increased performance compared to six years ago. Nowadays: Yahoo has the highest mean precision ratios at various cut-off points; all search engines have their highest mean precision ratios at cut-off point 5; dead links were encountered in Google, Bing, and Ask; and repeated documents were encountered in Google and Yahoo.
Citation - WoS: 9
Citation - Scopus: 18
Multi-Label Classification of Text Documents Using Deep Learning
(Ieee, 2020) Mohammed, Hamza Haruna; Dogdu, Erdogan; Gorur, Abdul Kadir; Choupani, Roya; 06.01. Bilgisayar Mühendisliği; 06. Mühendislik Fakültesi; 01. Çankaya Üniversitesi
Recently, studies in the field of Natural Language Processing and its related applications continue to mount up. Machine learning is proven to be predominantly data-driven in the sense that generic model building methods are used and then tailored to specific application domains. Needless to say, this has proven to be a very effective approach in modeling the complicated data dependencies we frequently experience in practice, making very few assumptions, and allowing the information to talk for themselves. Examples of these applications can be found in chemical process engineering, climate science, healthcare, and linguistic processing systems for natural languages, to name a few. Text classification is one of the important machine learning tasks that is used in many digital applications today; such as in document filtering, search engines, document management systems, and many more. Text classification is the process of categorizing of text documents into a given set of labels. Furthermore, multi-label text classification is the task of categorization of text documents into one or more labels simultaneously. Over the years, many methods for classifying text documents have been proposed, including the popularly known bag of words (BoW) method, support vector machine (SVM), tree induction, and label-vector embedding, to mention a few. These kinds of tools can be used in many digital applications, such as document filtering, search engines, document management systems, etc. Lately, deep learning-based approaches are getting more attention, especially in extreme multi-label text classification case. Deep learning has proven to be one of the major solutions to many machine learning applications, especially those involving high-dimensional and unstructured data. However, it is of paramount importance in many applications to be able to reason accurately about the uncertainties associated with the predictions of the models. In this paper, we explore and compare the recent deep learning-based methods for multi-label text classification. We investigate two scenarios. First, multi-label classification model with ordinary embedding layer, and second with Glove, word2vec, and FastText as pre-trained embedding corpus for the given models. We evaluated these different neural network model performances in terms of multi-label evaluation metrics for the two approaches, and compare the results with the previous studies.
Self-Supervised Learning With BYOL for Non-Alcoholic Fatty Liver Disease Diagnosis Using Ultrasound Imaging
(Springer London Ltd, 2025) Buktash, Ali; Gorur, Abdul Kadir; 06.01. Bilgisayar Mühendisliği; 06. Mühendislik Fakültesi; 01. Çankaya Üniversitesi
Purpose:The study aims to evaluate the effectiveness of Bootstrap Your Own Latent (BYOL), a self-supervised learning method for diagnosing NAFLD from ultrasound images using limited labeled data, which represents a novel approach in this domain. Self-supervised learning provides an alternative approach to traditional supervised learning by learning useful representations from unlabeled data, thereby reducing the time and cost required by radiologists to annotate images.Methods:The pre-trained ResNet-50 and ResNet-101 on the labeled ImageNet dataset were used for BYOL pre-training on ultrasound images without relying on labels. The training was conducted using default and custom augmentation, as well as balanced and imbalanced class distribution protocols. The model was then evaluated using linear and fine-tuning protocols with varying percentages of labeled data. The model was trained using three shuffled subsets, each trained 10 times. The custom augmentation set was derived by testing various augmentation settings using 100% and 1% of the labels to enhance feature learning.Results:BYOL with ResNet-101 and using the proposed custom augmentation set achieved average accuracies of 93.44%, 92.29%, and 88.49% using 100%, 10%, and 1% of the training labels across three shuffled datasets. In addition, our proposed method attained an average accuracy of 96.9% using patient-specific leave-one-out cross-validation (LOOCV).Conclusion:BYOL, with the proposed custom augmentation set, can learn effective image representations without relying on a large amount of labeled data, thereby enhancing scalability since unlabeled images are easier to acquire. It surpasses BYOL with default augmentation and training under supervised learning, especially with a low-labeled data regime.