Bildiri Koleksiyonu | Bilgisayar Mühendisliği Bölümü

Bu koleksiyon için kalıcı URI

https://hdl.handle.net/11729/4898

Listeleniyor 1 - 20 / 131

Sentiment analysis for hotel reviews in Turkish by using LLMs
(Institute of Electrical and Electronics Engineers Inc., 2024) Özdemir, Ata Onur; Giritli, Efe Batur; Can, Yekta Said
The field of sentiment analysis plays a pivotal role in consumer decision-making and service quality improvement within the hospitality industry. This study explores the application of Large Language Models (LLMs) for sentiment analysis of Turkish hotel reviews, contributing to the understanding of customer feedback and satisfaction. We created a dataset of 5,000 reviews by translating an English corpus into Turkish, which was then utilized to evaluate the performance of a state-of-the-art Turkish language model, TURNA. The study demonstrates that LLMs, particularly TURNA, outperform traditional machine learning algorithms and other advanced models in sentiment classification tasks, achieving an accuracy of 99.4%. This research underscores the potential of LLMs to enhance the accuracy of sentiment analysis, offering valuable insights for the tourism and hospitality sectors. The findings contribute to the ongoing evolution of sentiment analysis methodologies and suggest that LLMs can significantly improve t he understanding a nd processing of customer feedback in Turkish hotel reviews.
Retinal disease classification using optical coherence tomography angiography images
(Institute of Electrical and Electronics Engineers Inc., 2024) Aydın, Ömer Faruk; Nazlı, Muhammet Serdar; Tek, Faik Boray; Turkan, Yasemin
Optical Coherence Tomography Angiography (OCTA) is a non-invasive imaging modality widely used for the detailed visualization of retinal microvasculature, which is crucial for diagnosing and monitoring various retinal diseases. However, manual interpretation of OCTA images is labor-intensive and prone to variability, highlighting the need for automated classification methods. This study presents an aproach that utilizes transfer learning to classify OCTA images into different retinal disease categories, including age-related macular degeneration (AMD) and diapethic retinopathy (DR). We used the OCTA-500 dataset [1], the largest publicly available retinal dataset that contains images from 500 subjects with diverse retinal conditions. To address the class imbalance, we employed k-fold cross-validation and grouped various other conditions under the 'OTHERS' class. Additionally, we compared the performance of the ResNet50 model with OCTA inputs to that of the ResNet50 and RetFound (Vision Transformer) models with OCT inputs to assess the efficiency of OCTA in retinal condition classification. In the three-class (AMD, D R, Normal) classification, ResNet50-OCTA o utperformed ResNet50-OCT, but slightly underperformed compared to RetFound-OCT, which was pretrained on a large OCT dataset. In the four-class (AMD, DR, Normal, Others) classification, ResNet50-OCTA and RetFound-OCT achieved similar classification a ccuracies. This study establishes a baseline for retinal condition classification using the OCTA-500 dataset and provides a comparison between OCT and OCTA input modalities.
Segmentation based classification of retinal diseases in OCT images
(Institute of Electrical and Electronics Engineers Inc., 2024) Eren, Öykü; Tek, Faik Boray; Turkan, Yasemin
Volumetric optical coherence tomography (OCT) scans offer detailed visualization of the retinal layers, where any deformation can indicate potential abnormalities. This study introduced a method for classifying ocular diseases in OCT images through transfer learning. Applying transfer learning from natural images to Optical Coherence Tomography (OCT) scans present challenges, particularly when target domain examples are limited. Our approach aimed to enhance OCT-based retinal disease classification by leveraging transfer learning more effectively. We hypothesize that providing an explicit layer structure can improve classification accuracy. Using the OCTA-500 dataset, we explored various configurations by segmenting the retinal layers and integrating these segmentations with OCT scans. By combining horizontal and vertical cross-sectional middle slices and their blendings with segmentation outputs, we achieved a classification a ccuracy of 91.47% and an Area Under the Curve (AUC) of 0.96, significantly outperforming the classification of OCT slice images.
Integrating the focusing neuron model with N-BEATS and N-HiTS
(Institute of Electrical and Electronics Engineers Inc., 2024) Özçelik, Şuayb Talha; Tek, Faik Boray
The N-BEATS (Neural Basis Expansion Analysis for Time Series) model is a robust deep learning architecture designed specifically for time series forecasting. Its foundational idea lies in the use of a generic, interpretable architecture that leverages backward and forward residual links to predict time series data effectively. N - BEATS influenced the development of N-HiTS (Neural Hierarchical Interpretable Time Series), which builds upon and extends the foundational ideas of N-BEATS. This paper introduces new integrations to enhance these models using the Focusing Neuron model in blocks of N-BEATS and N-HiTS instead of Fully Connected (Dense) Neurons. The integration aims to improve the forward and backward forecasting processes in the blocks by facilitating the learning of parametric local receptive fields. Preliminary results indicate that this new usage can significantly improve model performances on datasets that have longer sequences, providing a promising direction for future advancements in N-BEATS and N-HiTS.
Implementing lightweight, dynamic hierarchical key assignment scheme for cloud computing
(IEEE, 2024-03-25) Çelikbilek, İbrahim; Çeliktaş, Barış; Özdemir, Enver
In this paper, we propose the implementation and adaptation of a hierarchical key assignment scheme (HKAS) previously developed in our research to improve access control in cloud computing environments. The secret keys generated and managed by this scheme can be utilized for various purposes within the cloud computing, including data encryption, integrity checks, secure communications, and accessing critical infrastructures or services. Our implementation performs dynamic update operations with minimal computational cost and storage demands, as users within the hierarchical structure do not store any key components. Through security analysis, the scheme demonstrates strong key indistinguishability security (S-KI-security), effectively safeguarding keys against various cryptographic attacks. The scheme's flexibility allows it to be tailored to specific organizational needs, whether for securing sensitive data, ensuring compliance with regulatory standards, or facilitating secure data sharing and collaboration in cloud environments. Thus, we advocate for the practical implementation of the HKAS in transitioning to cloud environments.
Leveraging transformer-based language models for enhanced service insight in tourism
(IEEE, 2023-12-22) Er, Aleyna; Özçelik, Şuayb Talha; Yöndem, Meltem Turhan
Customer feedback is a valuable resource for enhancing customer experience and identifying areas that require improvement. Utilizing user insights allows a tourism company to identify and address problematic points in its service delivery, provide feedback to partner companies regarding their product offerings, and even reconsider agreements by incorporating these opinions when curating their product portfolio. Setur implemented a systematic approach to collecting customer feedback by distributing "after-stay surveys'' to its customers via email following the completion of the agency services provided. Guest answers to open-ended questions that gather opinions about travel experience are analyzed by four tasks: user intention for answering, the sentiment of the review, subjects touched upon, and whom it concerned. For these tasks, transformer-based natural language processing (NLP) models BERT, DistilBERT, RoBERTa, and Electra are fine-Tuned to classify customer reviews. Based on the test results, it is observed that best practices could be gathered using Bert. In addition, we showed that different insights can be obtained from text comments made for two hotels in Aydin, Turkiye. Some users made complaints using neutral sentences. In some cases, people gave high scores to the numerical rating questions, but their open-ended questions could have a negative meaning.
Forecasting and analysis of energy consumption and waste generation in Antalya with SVR
(IEEE, 2023-12-24) Özçelik, Şuayb Talha; Tek, Faik Boray; Şekerci, Erdal
Antalya, a rapidly expanding coastal city in Türkiye, has experienced significant changes due to urbanization and increasing tourism activities. Comprehending tourism trends is crucial for the city's sustainable development and environmental management. Based on this perspective, this paper aims to present a comprehensive retrospective analysis of Antalya's energy consumption, domestic solid waste generation, wastewater generation, population growth, and tourist numbers over the years. Antalya faces significant challenges due to escalating trends in listed areas. Utilizing the Support Vector Regression, this study projects a need for an additional 1715 GWh of electricity production capacity, an expansion of wastewater capacity by 85639 thousand m3, and an increase in domestic solid waste disposal capacity by 597745 tons by 2028 to accommodate growing demands. We emphasize the importance of adopting effective policies and strategies to support energy efficiency, waste reduction, and wastewater management alongside sustainable urban planning and tourism management for Antalya's long-Term environmental sustainability and development. The findings presented in this study provide valuable insights for policymakers, urban planners, and stakeholders to make informed decisions, ensuring a balanced approach toward economic growth and environmental conservation.
Application of ChatGPT in the tourism domain: potential structures and challenges
(IEEE, 2023-12-23) Kılıçlıoğlu, Orkun Mehmet; Özçelik, Şuayb Talha; Yöndem, Meltem Turhan
The tourism industry stands out as a sector where effective customer communication significantly influences sales and customer satisfaction. The recent shift from traditional natural language processing methodologies to state-of-The-Art deep learning and transformer-based models has revolutionized the development of Conversational AI tools. These tools can provide comprehensive information about a company's product portfolio, enhancing customer engagement and decision-making. One potential Conversational AI application can be developed with ChatGPT. In this study, we explore the potential of using ChatGPT, a cutting-edge Conversational AI, in the context of Setur's products and services, focusing on two distinct scenarios: intention recognition and response generation. We incorporate Setur-specific data, including hotel information and annual catalogs. Our research aims to present potential structures and strategies for utilizing Language Model-based systems, particularly ChatGPT, in the tourism domain. We investigate the advantages and disadvantages of three different architectures and evaluate whether a restrictive or more independent model would be suitable for our application. Despite the impressive performance of Large Language Models (LLMs) in generating human-like dialogues, their end-To-end application faces limitations, such as system prompt constraints, fine-Tuning challenges, and model unavailability. Moreover, semantic search fails to deliver satisfactory performance when searching filters that require clear answers. To address these issues, we propose a hybrid approach that employs external interventions, the assignment of different GPT agents according to intent analysis, and traditional methods at specific junctures, which will facilitate the integration of domain knowledge into these systems.
AnlamVer: Semantic model evaluation dataset for Turkish - word similarity and relatedness
(Association for Computational Linguistics (ACL), 2018-08-26) Ercan, Gökhan; Yıldız, Olcay Taner
In this paper, we present AnlamVer, which is a semantic model evaluation dataset for Turkish designed to evaluate word similarity and word relatedness tasks while discriminating those two relations from each other. Our dataset consists of 500 word-pairs annotated by 12 human subjects, and each pair has two distinct scores for similarity and relatedness. Word-pairs are selected to enable the evaluation of distributional semantic models by multiple attributes of words and word-pair relations such as frequency, morphology, concreteness and relation types (e.g., synonymy, antonymy). Our aim is to provide insights to semantic model researchers by evaluating models in multiple attributes. We balance dataset word-pairs by their frequencies to evaluate the robustness of semantic models concerning out-of-vocabulary and rare words problems, which are caused by the rich derivational and inflectional morphology of the Turkish language.
TUR2SQL: A cross-domain Turkish dataset for Text-to-SQL
(IEEE, 2023-09-15) Kanburoğlu, Ali Buğra; Tek, Faik Boray
The field of converting natural language into corresponding SQL queries using deep learning techniques has attracted significant attention in recent years. While existing Text-to-SQL datasets primarily focus on English and other languages such as Chinese, there is a lack of resources for the Turkish language. In this study, we introduce the first publicly available cross-domain Turkish Text-to-SQL dataset, named TUR2SQL. This dataset consists of 10,809 pairs of natural language statements and their corresponding SQL queries. We conducted experiments using SQLNet and ChatGPT on the TUR2SQL dataset. The experimental results show that SQLNet has limited performance and ChatGPT has superior performance on the dataset. We believe that TUR2SQL provides a foundation for further exploration and advancements in Turkish language-based Text-to-SQL research.
Hotel sales forecasting with LSTM and N-BEATS
(IEEE, 2023-09-15) Özçelik, Şuayb Talha; Tek, Faik Boray; Şekerci, Erdal
Time series forecasting aims to model the change in data points over time. It is applicable in many areas, such as energy consumption, solid waste generation, economic indicators (inflation, currency), global warming (heat, water level), and hotel sales forecasting. This paper focuses on hotel sales forecasting with machine learning and deep learning solutions. A simple forecast solution is to repeat the last observation (Naive method) or the average of the past observations (Average method). More sophisticated solutions have been developed over the years, such as machine learning methods that have linear (Linear Regression, ARIMA) and nonlinear (Polynomial Regression and Support Vector Regression) methods. Different kinds of neural networks are developed and used in time series forecasting problems, and two of the successful ones are Recurrent Neural Networks and N-BEATS. This paper presents a forecasting analysis of hotel sales from Türkiye and Cyprus. We showed that N-BEATS is a solid choice against LSTM, especially in long sequences. Moreover, N-BEATS has slightly better inference time results in long sequences, but LSTM is faster in short sequences.
ISIKSumm at BioLaySumm task 1: BART-based summarization system enhanced with Bio-entity labels
(Association for Computational Linguistics (ACL), 2023-07-13) Çolak, Çağla; Karadeniz, İlknur
Communicating scientific research to the general public is an essential yet challenging task. Lay summaries, which provide a simplified version of research findings, can bridge the gap between scientific knowledge and public understanding. The BioLaySumm task (Goldsack et al., 2023) is a shared task that seeks to automate this process by generating lay summaries from biomedical articles. Two different datasets that have been created from curating two biomedical journals (PLOS and eLife) are provided by the task organizers. As a participant in this shared task, we developed a system to generate a lay summary from an article’s abstract and main text.
Auto Train Brain increases the variance of the gamma band sample entropy in the left hemisphere in dyslexia: a pilot study
(Springer Science and Business Media Deutschland GmbH, 2023) Eroğlu, Günet
Auto Train Brain is a mobile app that improves reading speed and reading comprehension in dyslexia. The efficacy of Auto Train Brain was proven with a clinical trial. We have analyzed the long-term training effects of the Auto Train Brain on dyslexic children. We have collected QEEG data from 14 channels from 21 dyslexic children for 100 sessions and calculated the sample entropy in the gamma bands for the left posterior brain (T7, P7, and O1). Although the gamma band values fluctuate and no permanent increase in the gamma band values is detected after Auto Train Brain training at T7, P7, and O1, the variance of gamma band sample entropy increases as the neurofeedback session number increases. We have concluded that the Auto Train Brain increases the flexibility of the left brain in dyslexia.
BOUN-ISIK participation: an unsupervised approach for the named entity normalization and relation extraction of Bacteria Biotopes
(Association for Computational Linguistics (ACL), 2019-11-04) Karadeniz, İlknur; Tuna, Ömer Faruk; Özgu, Arzucan
This paper presents our participation at the Bacteria Biotope Task of the BioNLP Shared Task 2019. Our participation includes two systems for the two subtasks of the Bacteria Biotope Task: the normalization of entities (BB-norm) and the identification of the relations between the entities given a biomedical text (BB-rel). For the normalization of entities, we utilized word embeddings and syntactic re-ranking. For the relation extraction task, pre-defined rules are used. Although both approaches are unsupervised, in the sense that they do not need any labeled data, they achieved promising results. Especially, for the BB-norm task, the results have shown that the proposed method performs as good as deep learning based methods, which require labeled data.
Convolutional neural network (CNN) algorithm based facial emotion recognition (FER) system for FER-2013 dataset
(IEEE, 2022-11-18) Ezerceli, Özay; Eskil, Mustafa Taner
Facial expression recognition (FER) is the key to understanding human emotions and feelings. It is an active area of research since human thoughts can be collected, processed, and used in customer satisfaction, politics, and medical domains. Automated FER systems had been developed and have been used to recognize humans’ emotions but it has been a quite challenging problem in machine learning due to the high intra-class variation. The first models were using known methods such as Support Vector Machines (SVM), Bayes classifier, Fuzzy Techniques, Feature Selection, Artificial Neural Networks (ANN) in their models but still, some limitations affect the accuracy critically such as subjectivity, occlusion, pose, low resolution, scale, illumination variation, etc. The ability of CNN boosts FER accuracy. Deep learning algorithms have emerged as the greatest way to produce the best results in FER in recent years. Various datasets were used to train, test, and validate the models. FER2013, CK+, JAFFE and FERG are some of the most popular datasets. To improve the accuracy of FER models, one dataset or a mix of datasets has been employed. Every dataset includes limitations and issues that have an impact on the model that is trained for it. As a solution to this problem, our state-of-the-art model based on deep learning architectures, particularly convolutional neural network architectures (CNN) with supportive techniques has been implemented. The proposed model achieved 93.7% accuracy with the combination of FER2013 and CK+ datasets for FER2013.
Analysis of single image super resolution models
(IEEE, 2022-11-18) Köprülü, Mertali; Eskil, Mustafa Taner
Image Super-Resolution (SR) is a set of image processing techniques which improve the resolution of images and videos. Deep learning approaches have made remarkable improvement in image super-resolution in recent years. This article aims and seeks to provide a comprehensive analysis on recent advances of models which has been used in image superresolution. This study has been investigated over other essential topics of current model problems, such as publicly accessible benchmark data-sets and performance evaluation measures. Finally, The study concluded these analysis by highlighting several weaknesses of existing base models as their feeding strategy and approved that the training technique which is Blind Feeding, which led several model to achieve state-of-the art.
Machine learning-based model categorization using textual and structural features
(Springer Science and Business Media Deutschland GmbH, 2022-09-08) Khalilipour, Alireza; Bozyiğit, Fatma; Utku, Can; Challenger, Moharram
Model Driven Engineering (MDE), where models are the core elements in the entire life cycle from the specification to maintenance phases, is one of the promising techniques to provide abstraction and automation. However, model management is another challenging issue due to the increasing number of models, their size, and their structural complexity. So that the available models should be organized by modelers to be reused and overcome the development of the new and more complex models with less cost and effort. In this direction, many studies are conducted to categorize models automatically. However, most of the studies focus either on the textual data or structural information in the intelligent model management, leading to less precision in the model management activities. Therefore, we utilized a model classification using baseline machine learning approaches on a dataset including 555 Ecore metamodels through hybrid feature vectors including both textual and structural information. In the proposed approach, first, the textual information of each model has been summarized in its elements through text processing as well as the ontology of synonyms within a specific domain. Then, the performances of machine learning classifiers were observed on two different variants of the datasets. The first variant includes only textual features (represented both in TF-IDF and word2vec representations), whereas the second variant consists of the determined structural features and textual features. It was finally concluded that each experimented machine learning algorithm gave more successful prediction performance on the variant containing structural features. The presented model yields promising results for the model classification task with a classification accuracy of 89.16%.
El yazısı rakam sınıflandırması için gözetimsiz benzerlik tabanlı evrişimler
(Institute of Electrical and Electronics Engineers Inc., 2022) Erkoç, Tuğba; Eskil, Mustafa Taner
Effective training of filters in Convolutional Neural Networks (CNN) ensures their success. In order to achieve good classification results in CNNs, filters must be carefully initialized, trained and fine-tuned. We propose an unsupervised method that allows the discovery of filters from the given dataset in a single epoch without specifying the number of filters hyper-parameter in convolutional layers. Our proposed method gradually builds the convolutional layers by a discovery routine that extracts a number of features that adequately represent the complexity of the input domain. The discovered filters represent the patterns in the domain, so they do not require any initialization method or backpropagation training for fine tuning purposes. Our method achieves 99.03% accuracy on MNIST dataset without applying any data augmentation techniques.
Comparison of choreography vs orchestration based Saga patterns in microservices
(Institute of Electrical and Electronics Engineers Inc., 2022) Aydın, Şahin; Çebi, Cem Berke
Microservice Architecture (MSA) is a design and architecture pattern created to deal with the challenges of conventional software programs in terms of stream processing, highly available flexibility, and infrastructural agility. Despite the many advantages of MSA, designing isolated services using the autonomous Databases per Services paradigm is difficult. We realized that because each microservice will have its repository, ensuring data coherence between databases becomes difficult, especially in reversals, where operations transcend different sites. Distributed networked transactions and rollbacks can be efficiently handled using two-phase commitment methods in hardware virtualization using RDBMS databases. However, these approaches can't be used in micro-services with segregated NoSQL servers. Three issues have been addressed in this study: (i) investigate the implementation of event choreography and orchestration methods for the Saga pattern execution in MSA, (ii) existing reality suggestions on the saga pattern adoption and implementation besides the use cases, and (iii) introduce the disbursed transaction records and rollbacks challenges in isolated No-SQL databases with reliant collections in MSA.
Unreasonable effectiveness of last hidden layer activations for adversarial robustness
(Institute of Electrical and Electronics Engineers Inc., 2022) Tuna, Ömer Faruk; Çatak, Ferhat Özgür; Eskil, Mustafa Taner
In standard Deep Neural Network (DNN) based classifiers, the general convention is to omit the activation function in the last (output) layer and directly apply the softmax function on the logits to get the probability scores of each class. In this type of architectures, the loss value of the classifier against any output class is directly proportional to the difference between the final probability score and the label value of the associated class. Standard White-box adversarial evasion attacks, whether targeted or untargeted, mainly try to exploit the gradient of the model loss function to craft adversarial samples and fool the model. In this study, we show both mathematically and experimentally that using some widely known activation functions in the output layer of the model with high temperature values has the effect of zeroing out the gradients for both targeted and untargeted attack cases, preventing attackers from exploiting the model's loss function to craft adversarial samples. We've experimentally verified the efficacy of our approach on MNIST (Digit), CIFAR10 datasets. Detailed experiments confirmed that our approach substantially improves robustness against gradient-based targeted and untargeted attack threats. And, we showed that the increased non-linearity at the output layer has some ad-ditional benefits against some other attack methods like Deepfool attack.

Güncel Gönderiler