Ecem Şimşek

Selected publications are highlighted. The list of my publications can also be found in Google Scholar.

Emotion Classification with Visibility Graphs (EmoVis)

Ecem Şimşek, Atakan Topcu, Emirhan Koç, Emine Ülkü Sarıtaş, Aykut Koç

IEEE Signal Processing Letters (SPL)

abstract

Transformers have gained prominence in natural language processing due to their representational capabilities and performances. Transformers process natural language as a sequence on finite context windows; however, global relationships among words beyond these windows cannot be completely modeled via sequence processing only. Graph neural network (GNN) based models have been proposed to alleviate this problem, as they provide geometric extensions to neural networks, enabling models to learn associations within a text. However, regular graph-based methods ignore the sequential nature of underlying texts. In this paper, we propose EmoVis, the first generic graph-based neural network that utilizes visibility graphs, which converts classical time-series information to graph representations. We cast the problem as an emotion classification task, enabling the proposed model to learn associations between the labels and words in a sentence. Moreover, EmoVis can be used as a highly modular graph-based extension to any transformer-based model, significantly improving their performance and learning capabilities in various languages. We experimentally show that EmoVis enables transformer-based models to outperform the state-of-the-art baselines across three diverse datasets in different languages in the SemEval2018 competition datasets and the GoEmotions dataset.

Fractional Fourier-Guided Visibility Graphs

Ecem Şimşek, Haldun Özaktaş, Aykut Koç

To be submitted to IEEE Signal Processing Letters (SPL)

abstract

EpiGraphNet: Epilepsy Recognition Architecture with Graph-Based EEG Analysis

Ecem Şimşek, Emirhan Koç, Aykut Koç

IEEE Signal Processing and Communications Applications Conference (SIU)

abstract

Early and accurate detection of epileptic seizures is crucial for patients' quality of life and the treatment process. In this context, our study proposes EpiGrafNet, an innovative hybrid model that delivers high performance in epileptic seizure detection using electroencephalography (EEG) signals. The proposed method extracts both local and long-term temporal features from EEG signals via a one-dimensional convolutional neural network (CNN) and long short-term memory (LSTM) module; these features are then transformed into a graph structure by the utilization of correlation connection matrices (CCM), and integrated with graph convolutional neural networks (GCN). Experiments conducted with different sparsity values demonstrate that the developed model achieves superior accuracy, recall, precision, and F1 scores in both binary and multi-class epileptic seizure detection tasks, indicating that EpiGrafNet provides higher and more reliable results compared to existing methods. This approach, by effectively modeling the complex spatial and temporal relationships in EEG signals, significantly contributes to the development of automatic epileptic seizure detection systems in clinical applications.

Wind Power Prediction Using Machine Learning and Deep Learning Algorithms

Ecem Şimşek, Ayşemüge Güngör, Öykü Karavelioğlu, Mustafa Tolga Yerli, Nejat Göktuğ Kuyumcuoğlu

IEEE Signal Processing and Communications Applications Conference (SIU)

abstract

In this study, it has been tried to predict the wind power generation values in a long-term period by using a dataset containing the wind power generation values of 10 zones using machine learning and deep learning methods. In this context, the importance of accurately predicting renewable energy production was emphasized by associating it with machine learning and deep learning methods. The methods to be used in the study were selected based on the literature review and the characteristics of the time series datasets. Since the dataset includes the basic wind components, a detailed feature analysis was performed, and the dataset was enriched with the newly added features. The hyperparameters of the utilized models were optimized for all regions in the dataset separately and the models were run with these hyperparameters. The results of the models were evaluated with different error metrics and compared with each other, and the models with the lowest error scores were determined.

Automatic Detection of Basic Level Categories

Emirhan Koç, İrem Şanlı, Ecem Şimşek, Hasan Özekin, Aykut Koç

IEEE Signal Processing and Communications Applications Conference (SIU)

abstract

Basic level categories (BLCs), which can be defined as the most inclusive level at which a concrete mental image of the entire category can be formed, have proven to be useful in a variety of applications in natural language processing (NLP) and computer vision (CV) tasks, such as word sense disambiguation, image searches, image description, and retrieval. Limiting their practical applications, current methods for detecting BLCs predominantly rely on rule-based methods and external knowledge sources rather than the information extracted directly from the text. In this manuscript, we propose a novel approach to detect BLCs that is merely based on the information obtained from the word embeddings, including Gaussian word embeddings (W2G) and embeddings retrieved from transformer-based models such as BERT and GPT-2. The proposed method significantly outperforms existing works in performance and practicality, demonstrating the effectiveness of contextual word embeddings for BLC detection.

Ecem Şimşek

Education

Professional Experience

Publications

Honors and Awards

Teaching

Skills