My research group is actively working in different areas of recommendation systems, prediction models, signal processing, machine learning, and deep learning. Recently, we focus on developing deep learning models for emotion recognition/classification, stock market prediction, dynamic pricing and task extraction and user modeling for recommendation systems. Some of the topics that we worked on are:
Emotion Classification:
My PhD student Deger Ayata worked on emotion classification and finished hihs PhD in 2019. Our initial work compared the time domain based features with wavelet based features using different window sizes [1]. An extended version of this paper is invited for a special issue and published in [2]. Later, we did multi channel emotion classification using signal fusion in [3]. We have utilized different learning algorithms and applied Autoencoder based features using deep learning frameworks and compared their results in [4].
Stock Market Prediction:
Stock market price data have non-linear, noisy and non-stationary structure, and therefore prediction of the price or its direction are both challenging tasks. Most of the work that has been conducted on stock market prediction uses indicators or price information. In one of our initial work we analyzed the effect of feature selection for daily return prediction [5]. Then the daily movement directions of three frequently traded stocks in Borsa Istanbul were predicted using CNN [6]. We have extended our initial works and we have extracted a feature set using different indicators, price and temporal information. Correlations between instances and features are utilized to order the features before they are presented as inputs to the CNN [7]. We have shown that our deep learning architecture outperforms both Logistic Regression and CNN that utilizes randomly ordered features. Recently, we investigated the performance of Graph Neural Networks on stock market prediction
Recommendation Systems:
Task extraction on query logs is one of the important and interesting topics used on search engines and many search-based applications. Task extraction is important for providing suggestions in the direction of the user's intention, in the search text completion, in returning the correct results for the domain being searched. Session information, clicked document contents and query entities are used for feature extraction in existing approaches. In some of the recent studies, in addition to the existing features the Wikipedia/DBPedia category hierarchies are used as feature for task extraction. Previously, we have shown that Word2Vec representation of the category hierarchies improves the task extraction results and achieved very promising results[8]. Currently, my PhD student Nurullah Ates has been working on task extraction problem and he has applied Siamese networks for task classification problems[9]. We are currently investigating the deep clustering approaches in task extraction problems.
References:
[1] Deger Ayata, Yusuf Yaslan, Mustafa Ersel Kamasak (2016). Emotion Recognition via Random Forest and Galvanic Skin Response: Comparison of Time Based Feature Sets, Window Sizes and Wavelet Approaches. Tip Teknolojileri Kongresi, 27-29 October 2016, Antalya Turkey
[2] Deger Ayata, Yusuf Yaslan, Mustafa Ersel Kamasak (2017). Emotion Recognition via Galvanic Skin Response: Comparison of Machine Learning Algorithms and Feature Extraction
Methods. Istanbul University-Journal of Electrical and Electronics Engineering, 17(1), 3129-3136.
[3] Deger Ayata, Yusuf Yaslan, Mustafa Ersel Kamasak (2017). Emotion Recognition via Multi Channel EEG Signal Fusion and Pattern Recognition. EUSIPCO 2017 Multi-Learn 2017: Multimodal
processing, modeling and learning approaches for human-computer/robot interaction Workshop, 2 September 2017, Kos Island Greece.
[4] Deger Ayata, Yusuf Yaslan, Mustafa Ersel Kamasak (2017). Multi Channel Brain EEG Signals Based Emotional Arousal Classification with Unsupervised Feature Learning using Autoencoders.
2017 25th Signal Processing and Communications Applications Conference(SIU), 15-18 May 2017 Antalya, Turkey
[5] Hakan Gunduz, Zehra Cataltepe, Yusuf Yaslan (2017). Stock daily return prediction using expanded features and feature selection. Turkish Journal of Electrical Engineering Computer Sciences, 25, 4829-4840.
[6] Hakan Gunduz, Yusuf Yaslan, Zehra Cataltepe (2017). Stock market direction prediction using deep neural networks. 25th Signal Processing and Communications Applications Conference (SIU),15-18 May 2017 Antalya, Turkey
[7] Hakan Gunduz, Yusuf Yaslan, Zehra Cataltepe (2017). Intraday prediction of Borsa Istanbul using convolutional neural networks and feature correlations. Knowledge-Based Systems, 137, 138-148.
[8] Almila Selcen Akgun, Yusuf Yaslan (2018), Task-Based Clustering on Search Queries, 2018 26th Signal Processing and Communications Applications Conference(SIU), 2-5 May 2018 Izmir, Turkey
[9] Nurullah Ates, and Yusuf Yaslan (2021), Labeling Consecutive Search Query Pairs Using Siamese Networks. 11th International Conference of Pattern Recognition Systems (ICPRS 2021). Vol. 2021. IET, 2021.
For a detailed list of my recent publications and active research areas, please see: