İlknur Dönmez

TUBITAK, Information Technologies Institute, Senior Researcher

Followers

Following

Co-author

Mentions

Public Views

To apply my solid industrial experience and strong academic background in Machine Learning, Data Mining, Big Data, Natural Language Processing and Project Management to research, design and implementation of data-centric, state-of-the-art software solutions in an organization led by a growth mindset.

less

University of Oxford

Williams College

Ibb University

Northern Arizona University

Stephen Mackessy

University of Northern Colorado

Nikola Milosevic

The University of Manchester

Md Kafiul Islam

Independent University, Bangladesh

Hasan Davulcu

Arizona State University

Buddhika Kasthuriarachchy

Sri Lanka Institute of Information Technology

Mahmod Sahebi

Interests

Uploads

Papers by İlknur Dönmez

Multimodal Speech Emotion and Text Sentiment Analysis

2023 8th International Conference on Computer Science and Engineering (UBMK)

Analyzing Five Conscious and Unconscious Behaviors Using Google n-gram Database Generated from Millions of Books

Due to the exponential increase in the proliferation of information by humans and machines, the 2... more Due to the exponential increase in the proliferation of information by humans and machines, the 21st century is called the "Information Age" and the "Digital Age". Regardless of the size of the data at hand, the challenge always is to clear, combine, analyze and transform the data into meaningful information. In this article, the words "eating, hearing, seeing, guiding, humming" in Google n-gram corpus are examined along with their adjacent adjectives indicating to elucidate whether they refer to a conscious action or not. The Google n-gram corpus was generated from millions of scanned books published between 1500 and 2008. Books are the expression of human emotions, thoughts and behaviors. The frequency of the usage of certain words in books can be considered as a marker to estimate the importance of the concept represented by that word. Also, the adjective that is frequently used to modify a noun can provide us with the information about the characteristics of the concept that the noun represents. In this study, an extensive body of data derived from the published books have been used to analyze conscious and unconscious behaviors. Time-dependent conscious and unconscious behavior trends were obtained by using linear regression. The uniqueness of this study lies in the fact that it uses written statements for conscious and unconscious behavior analysis. Among the five behaviors examined, the most common words in the books are "conscious eating". The use of this phrase also showed the highest increase over time. Although "unconscious eating" has almost the same increments as "conscious seeing", the concept of "unconscious seeing" has found little place in the books.

Cancel-for-Any-Reason Insurance Recommendation Using Customer Transaction-Based Clustering

IEEE Access, 2021

In the travel insurance industry, cancel-for-any-reason insurance, also known as a cancellation p... more In the travel insurance industry, cancel-for-any-reason insurance, also known as a cancellation protection service (CPS), is a recent attempt to strike a balance between customer satisfaction and service provider (SP) profits. However, some exceptional circumstances, particularly the COVID-19 pandemic, have led to a dramatic decrease in SP revenues, especially for non-refundable tickets purchased early with CPS. This paper begins by presenting a risk group segmentation of customers in an online ticket reservation system. Then, a CPS fee is recommended depending on the different customer risk groups provided by the cluster segmentation via different clustering algorithms such as centroid-based K-means, hierarchical agglomerative, DBSCAN, and artificial neural network-based SOM algorithms. According to the implemented cluster metrics, which include the Silhouette index, Davies-Bouldin index, Entropy index, and DBCV index, the SOM algorithm presents the most appropriate result. After predicting the new customer cluster, a CPS fee will be calculated with the proposed adaptive CPS method based on the cluster segmentation weights. Determining the weight of each cluster is related to the total CPS revenue threshold for all clusters defined by the SP. Therefore, to avoid a loss for SPs, the total CPS revenue will be kept constant with the threshold that the SP has been adjusted. The experimental results based on real-world data show that the risk group segmentation of customers helps to maintain a balance between CPS fees and SP profits. Finally, according to the calculated weights, the proposed model pegs the SP gain/loss variation with a 0.00012 exchange ratio.

Download

An experimental study: using categorical or fuzzy inputs for classification problems with dimensionality reduction

Neural Computing and Applications, Jan 21, 2023

A Proposed Methodology for Risk Classification Using Fuzzy Group Decision Making and Fuzzy C-Means

Springer eBooks, 2022

Role of air pollution on COVID-19 in Istanbul

Sigma Journal of Engineering and Natural Sciences, 2022

There are many types of research on strong relations between air pollution and the respiration sy... more There are many types of research on strong relations between air pollution and the respiration system. This paper is related to the effect of air pollution on pandemic case numbers. By analyses of the number of patients and mean air pollution (µg/m³) data by using data mining, it is concluded that there was evidence of this relationship with a significant level alpha = 0,10. This paper estimates "the number of patients with coronavirus" as a function of daily air pollution and corona case numbers data using single ANN, LSTM, and W-ANN, W-LSTM hybrid methods. This paper explains small, meso, and large scale factors and their role on COVID-19 patients. This finding does not demonstrate a direct cause-effect relationship between air pollution and COVID-19 patients. This study shows the importance of pollution on the number of patients with COVID-19 infection. Although our results have significant uncertainties, we can clearly distinguish the contribution of air pollution to COVID-19 patients. When we used wavelet transformation of air pollution data to estimate COVID-19 patient's numbers, R 2 score is increased in both ANN and LSTM between [0.01-0.10]. Nevertheless, the actual number of patients is influenced by many additional factors such as the country's health system. After error analysis, sMAPE [3.6-5.9] changed, there is sufficient evidence of model results (M3, M4 Hybrid LSTM) and observation. (0.91<R 2 <0.96, alpha=<0.01). The performance of hybrid model is 10% better than the simple ANN model.

Download

Vectorial Approach For Analysing Turkish Sentence

Fen ve mühendislik bilimleri dergisi, Sep 11, 2015

Anahtar kelimeler Dilbilgisi çözümlemesi, anlamsal çözümleme, öbek-kavramın yüklem uyumluluğu, cü... more Anahtar kelimeler Dilbilgisi çözümlemesi, anlamsal çözümleme, öbek-kavramın yüklem uyumluluğu, cümlenin vektör temsili. Özet Tümcenin anlamsal ve dilbilgisi açısından çözümlenmesi Doğal Dil İşleme (DDİ)'nin ana konulardan biridir. Çalışmamızda, tümcedeki temel dilbilgisi ve anlamsal yanlışları saptamak için yüklemi temel alan yeni bir yöntem önerilmektedir. Türkçe tümcede yüklem özne ve zaman bilgisi içerir. Ayrıca yüklem, o tümcenin hangi öbeklerden oluşabileceği konusunda da belirleyicidir. Örneğin, "büyümek" yüklemi tümce içinde nesne almazken,-de ekiyle biten dolaylı tümleç öbeğini alır. Yüklem ayrıca her bir öbeğin içereceği kavram hakkında da bilgi içermektedir. Örneğin "düşünmek" yüklemi insanlara özgüdür. Dolayısıyla özne olarak insan kavramıyla ilişkilidir. Bu saptamalardan yola çıkarak çalışmamızda, tümcelerin öbekleri bulunmuş; her bir öbeğin hangi kavramla ilişkili olduğu belirlenmiş ve tümcenin dilbilgisi çözümlemesini ve anlam çözümlemesini yapan bir model tasarlanmıştır.

Download

Contex Free Grammer For Turkish

DergiPark (Istanbul University), Aug 15, 2018

Download

Sadness and Happiness Analysis Acording to Gender Using Google n-gram Database

DergiPark (Istanbul University), Jun 1, 2019

The currentera has been defined as "Dijital Age" and "Information Age" since it is characterized ... more The currentera has been defined as "Dijital Age" and "Information Age" since it is characterized by an exponential growth of data, generated by bothhuman, i.e. social environments, and machine, i.e. Internet of things. The challenge is to convert "data" into "information", by analyzing the data and discovering patterns hidden inside it. In this paper the two basic human feelings of Happiness and Sadness are extracted from a subset of Google n-grams corpus and analyze daccording to gender. Google n-gramscorpus is generated from millions of scanned boks published between year 1500 and 2008; it can be considered as an indicator forhuman specific feature and behavior. Under the hypothesis thatuser'semotion can be extrapolated by the frequency of the corresponding emotional words, this study applies regression to predict the importance of the Happiness and Sadness emotional states in future years. This study is enhanced version of the "Feeling Analysis for Sadness and Happiness using Google n-gram Database"study which includes gender fraction for each analysis.

Download

Türkçe Hal Ekleri ve Öbekleri Kapsayan Ba˘glamdan Bağımsız Dil Temsili Kuralları

Türkiye Bilişim Vakfı Bilgisayar Bilimleri ve Mühendisliği Dergisi, Oct 3, 2017

Bir Steganografik metodun değerlendirme kriterlerinden biri de dayanıklılıktır. Bu çalışmada, ile... more Bir Steganografik metodun değerlendirme kriterlerinden biri de dayanıklılıktır. Bu çalışmada, iletim hattı boyunca örtü verinin uğrayabileceği saldırılara karşı gizlenmiş veriyi korumak hedeflenmiştir. Resmin parlaklık değişimlerine karşı gizlenmiş mesajı korumak için yerel ikili örüntü (LBP) operatörünün kullanılması önerilmiştir. Daha önceden damgalama tekniklerinde kullanılan bu metodun, daha farklı bir yaklaşımla steganografide kullanılması önerilmiştir. Resme LBP operatörü uygulandıktan sonra LBP haritasının üzerinde, LSB metodu kullanılarak veri gizlenecektir. Önerilen algoritmanın saldırılara karşı dayanıklılığı, kapasitesi ve taşıyıcıdaki değişim ölçülecektir.

Download

2D Vector Representation of Binomial Hierarchical Tree Items

2022 International Conference on Theoretical and Applied Computer Science and Engineering (ICTASCE)

Parkinson hastalarında motor komplikasyonlar ve kortikal eksitabilite

Giriş: Parkinson Hastalığı, yavaş progresyon gösteren, Alzheimer hastalığından sonra en sık görül... more Giriş: Parkinson Hastalığı, yavaş progresyon gösteren, Alzheimer hastalığından sonra en sık görülen nörodejeneratif hastalıktır. Tanı için bradikinezi esastır. Rijidite, istirahat tremoru veya postural instabiliteden birinin eşlik etmesi gerekmektedir. Parkinson Hastalığı patogenezinde; substansiya nigra pars kompakta ve nigrostriatal yolaklardaki dopaminerjik nöron dejenerasyonu ile karakterizedir. Parkinson Hastalığının geç dönem komplikasyonları olarak motor dalgalanmalar ve diskineziler ortaya çıkabilmektedir. Motor dalgalanmalar; doz sonu kötüleşmesi, gecikmiş on yanıtı veya on yaralının olmaması, öngörülemeyen on-off dalgalanmalar, donmalar şeklinde karşımıza çıkabilmektedir. Diskineziler ise tepe doz diskinezisi, difazik diskinezi , off distonisi veya kare dalga diskinezisi şeklinde olabilmektedir. Motor dalgalanma ve diskinezilerin nigrostriatal dopaminerjik yoldaki presinaptik ve postsinaptik değişikliklere bağlı olduğu bilinmektedir. Ancak motor komplikasyonların oluş meka...

Dealing with the Uncertainty Between the Text Content and the Categories by a Proposed Wavelet Similarity Metric

Springer eBooks, Aug 24, 2021

The rise of the issues related to the uncertainty of decision-making has become a warm issue in o... more The rise of the issues related to the uncertainty of decision-making has become a warm issue in operation research. The study is dealing with the uncertainty of text contents and text categories. The complex and ambiguous structure of texts prevents us to have clear and precise categorization. The purpose of this paper is to find the similarity distance for the different categories that the text may be related to, using hidden semantic relations. Our study proposes a new method to reveal the hypernym relations (generic terms, upper classes of the term) of the words in the text and formalize the similarity distance metrics between the given text and different categories. We proposed an original and novel measurement formula to calculate the similarity of defined categories to a specific text using discrete Wavelet transformation. Utilizing the Wavelet transformation method that has been rarely used in text analysis, the upper class “hypernym” relation and their dominance on each other is found. The strength of the results examined on the sample text (I like giraffes and I am afraid of lions. I saw them when they are standing opposite of each other near a cactus in Hoanib Desert.). In the first version that the frequencies are taken into account, sample text is categorized as “warm-blooded animals” between the 8 categories with highest similarity distance. For the normalized binary-valued version, maximum and minimum similarities are calculated in the range from 0.8639 to 0.1360, respectively.

An experimental study: using categorical or fuzzy inputs for classification problems with dimensionality reduction

Neural Computing and Applications

Sadness and Happiness Analysis Acording to Gender Using Google n-gram Database

Istanbul University - DergiPark, Jun 1, 2019

Grammatical And Semantic Analysis Of Turkish Sentence Based On Predicate

Tez (Doktora) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 2016Thesis (PhD) -- İstan... more Tez (Doktora) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 2016Thesis (PhD) -- İstanbul Technical University, Institute of Science and Technology, 2016Çalışmamız tümcelerin anlamsal ve dilbilgisi çözümlemesini içermektedir. Tümcenin anlamsal ve dilbilgisi açısından çözümlenmesi Doğal Dil İşleme (DDİ)’nin ana konulardan biridir. Çalışmamızda, çözümleme yapılırken önce metin içindeki tümcelerin her biri basit tümce olacak şekilde alt tümcelere ayrıştırılmaktadır. Her bir alt tümceye ait öbek kavram çiftleri bulunmakta ve daha sonra her bir alt tümcedeki temel dilbilgisi ve anlamsal yanlışları saptamak için yüklemi temel alan yeni bir yöntem önerilmektedir. Türkçe tümcede yüklem özne ve zaman bilgisi içerir. Ayrıca yüklem, o tümcenin hangi öbeklerden oluşabileceği konusunda da belirleyicidir. örneğin, ``büyümek'' yüklemi tümce içinde nesne almazken, ``-de'' ekiyle biten dolaylı tümleç öbeğini alır. Örneğin ``Ayşeyi büyüdü.'' tümcesi sorunluyken, ``S...

Extracting phrase-content pairs for Turkish sentences

2015 9th International Conference on Application of Information and Communication Technologies (AICT), 2015

In this paper we present a framework for extraction of Turkish phrases and their concepts. The ob... more In this paper we present a framework for extraction of Turkish phrases and their concepts. The objective of the study is meeting the requirement of sources for Turkish Semantic Extractions and represent a Turkish sentence at phrase-concept level. The semantic and grammatical analysis of a sentence is a basic content of Natural Language Processing (NLP) which is a branch of Artificial Intelligence (AI). In our study Turkish Phrase-Content Finding system is formed as a source for the other application areas in NLP. This system can be used in Summarization Systems, Information Extraction, Automatic Question Answering System, Semantic Role Labeling, and other semantic application.

A Proposed Methodology for Risk Classification Using Fuzzy Group Decision Making and Fuzzy C-Means

Lecture notes in networks and systems, 2022

Document Sentiment Classification Using Hybrid Wavelet Methodologies

Sentiment and semantic analysis of a text are very important issues of today because of increasin... more Sentiment and semantic analysis of a text are very important issues of today because of increasing text data. Our study proposes a new method to reveal the hypernym relations (generic term) of the words in the text and to enhance the accuracy results of sentiment classification of the texts. We used wavelet transform method that has been rarely used in text analysis. In our study, the aim is to show how this method contributes the sentiment analysis classification problem. We used classical algorithms and hybrit wavelet algorithm for sentiment analysis problem. It has been observed that when wavelets are applied to classical classification algorithms, the accuracies increased approximately 5%