İlhami SEL

Dr. İlhami SEL

Inonu University, Malatya, Turkey (Türkiye)
orcid

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
8 October 2024 in Mathematics
Verified
Journal Article
Open access
PDF file
İSİlhami SEL
İlhami SEL
DHDavut Hanbay
Davut Hanbay
This study focuses on the neural machine translation task for the TR-EN language pair, which is considered a low-resource language pair. We investigated fine-tuning strategies for pre-trained language models. Specifically, we explored the effectiveness of parameter-efficient adapter methods for fine-tuning multilingual pre-trained language models. Various combinations of LoRA and bottleneck adapters were experimented with. The combination of LoRA and bottleneck adapters demonstrated superior performance compared to other methods. This combination required only 5% of the pre-trained language model to be fine-tuned. The proposed method enhances parameter efficiency and reduces computational costs. Compared to the full fine-tuning of the multilingual pre-trained language model, it showed only a 3% difference in the BLEU score. Thus, nearly the same performance was achieved at a significantly lower cost. Additionally, models using only bottleneck adapters performed worse despite having a higher parameter count. Although adding LoRA to pre-trained language models alone did not yield sufficient performance, the proposed method improved machine translation. The results obtained are promising, particularly for low-resource language pairs. The proposed method requires less memory and computational load while maintaining translation quality.

15 September 2021 in Fırat Üniversitesi Mühendislik Bilimleri Dergisi
Verified
Journal Article
Closed access
İSİlhami SEL
İlhami SEL
DHDavut Hanbay
Davut Hanbay
Yazar profili oluşturma (Author Profiling) bir metnin üslup ve içeriğine bakarak yazarın çeşitli özelliklerinin ortaya çıkarılmasına yönelik bir metin kümesi analizidir. Bu özellikler yaş, cinsiyet, kişilik özellikleri ve hatta meslek gibi unsurları barındırır. Cinsiyet belirleme yazar profili oluşturma çalışmalarının alt alanlarından birisidir. Siber suçlar başta olmak üzere sahte haber yayma gibi adli olayların yanında pazarlama (reklamcılık), sosyolojik ve psikolojik olayların incelenmesinde cinsiyet belirleme oldukça önemlidir. Twitter gönderileri dil kurallarına uymayan, kısaltılmış kelimeler ve anlamsız cümle yapıları da içerme ihtimallerine rağmen cinsiyet belirleme görevi için yaygın bir şekilde kullanılmaktadır. Bu çalışmada Türkçe Twitter gönderilerinden cinsiyet tespiti yapılmaya çalışılmıştır. Problem bir sınıflandırma görevi olarak ele alınmıştır. Yapılan çalışmada makine öğrenmesi metotları(TF-IDF + SVM), derin öğrenme yöntemleri (LSTM, CNN) ve Türkçe için ön eğitimli dil modelleri(BERT, DistilBert, Electra) kullanılmıştır. Yapılan deneyler sonucunda en yüksek başarımı (%80.1) kelime boyutunun 128k olduğu Bert modeli sağlamıştır. Bu çalışma diğer metin sınıflandırma görevleri için de detaylı bir çalışma olma özelliği göstermektedir.

1 September 2019
Verified
Conference Paper
Closed access
İSİlhami SEL
İlhami SEL
AKAli Karci
Ali Karci
DHDavut Hanbay
Davut Hanbay
The feature selection can be defined as the selection of the best subset to represent the data set, that is, the removal of unnecessary data that does not affect the result. The efficiency and accuracy of the system can be increased by decreasing the size and the feature selection in classification applications. In this study, text classification was applied by using “20 news group” data published by Reuters news agency. The pre-processed news data were converted into vectors using the Doc2Vec method and a data set was created. This data set is classified by the Maximum Entropy Classification method. Afterwards, a subset of data sets was created by using the Mutual Information Method for the feature selection. Reclassification was performed with the resulting data set and the results were compared according to the performance rates. While the success of the system with 600 features was (0.9285) before the feature selection, (0.9285), then, the performance rates of the 200, 100, 50, 20 models were obtained as (0.9454, 0.9426, 0.9407, 0.9123), respectively. When the results were examined, the success of the 50-featured model was higher than the 600-featured model initially created.

11 November 2022 in Applied Sciences
Verified
Journal Article
Open access
PDF file
İSİlhami SEL
İlhami SEL
DHDavut Hanbay
Davut Hanbay
English is accepted as an academic language in the world. This necessitates the use of English in their academic studies for speakers of other languages. Even when these researchers are competent in the use of the English language, some mistakes may occur while writing an academic article. To solve this problem, academicians tend to use automatic translation programs or get assistance from people with an advanced level of English. This study offers an expert system to enable assistance to the researchers throughout their academic article writing process. In this study, Turkish which is considered among low-resource languages is used as the source language. The proposed model combines the transformer encoder-decoder architecture model with the pre-trained Sci-BERT language model via the shallow fusion method. The model uses a Fully Attentional Network Layer instead of a Feed-Forward Network Layer in the known shallow fusion method. In this way, a higher success rate could be achieved by increasing the attention at the word level. Different metrics were used to evaluate the created model. The model created as a result of the experiments reached 45.1 BLEU and 73.2 METEOR scores. In addition, the proposed model achieved 20.12 and 20.56 scores, respectively, with the zero-shot translation method in the World Machine Translation (2017–2018) test datasets. The proposed method could inspire other low-resource languages to include the language model in the translation system. In this study, a corpus composed entirely of academic sentences is also introduced to be used in the translation system. The corpus consists of 1.2 million parallel sentences. The proposed model and corpus are made available to researchers on our GitHub page.

30 January 2022 in Balkan Journal of Electrical and Computer Engineering
Verified
Journal Article
Open access
PDF file
IÇipek balıkçı çiçek
ipek balıkçı çiçek
İSİlhami SEL
İlhami SEL
Fatma Hilal YaginFatma Hilal Yagin
Fatma Hilal Yagin
Cemil ColakCemil Colak
Cemil Colak
Classification; biomedical, bioinformatics, medicine, engineering etc. It is a fundamental approach that is frequently used in many research areas, such as especially in the field of health; it has become common to classify diseases with machine learning methods using risk factors of these diseases and to determine the effect levels of these risk factors on the related disease. There are both commercial and free software tools that researchers can analyze their data with classification methods. The aim of this study is to develop a user-friendly web-based software for classification analysis. Python sklearn and Dash libraries were used during the development of the software. Among the classification algorithms in the developed software; Logistic regression, Decision trees, Support vector Machines, Random Forest, LightGBM, Gaussian Naive Bayes, AdaBoost and XGBoost methods are available. In order to show how the software works, a classification model was created with the Random forest algorithm using the cervical cancer data set. Different metric values were evaluated for the models. Obtained from a random forest classification model;accuracy, sensitivity, specificity, negative predictive value, matthews correlation coefficient, and F1 score values obtained from the model were 94.44%, 100%, 93.33%, 100%, 83.67%, and 94.44 respectively. It is thought that the classification software developed in this study will provide great convenience to clinicians and researchers in the field of medicine, in terms of applying predictive classification algorithms for the disease without any software knowledge.

1 September 2019
Verified
Conference Paper
Closed access
İSİlhami SEL
İlhami SEL
CYCelaleddin Yeroğlu
Celaleddin Yeroğlu
DHDavut Hanbay
Davut Hanbay
Feature selection can be defined as the selection of the best subset to represent the data set in machine learning applications, in other words extraction of the unnecessary data that has no effect on the result. In classification problems efficiency and accuracy of the system can be increased when the dimension is reduced by feature selection. In this study, text classifying application is performed by using the data set of “20 News Group” released in Reuters News Agent. The pre-processed news data were converted to vectors by using Doc2Vec method and the data set was created and classified by Naive Bayes method. Subsequently, a subset of the data set was formed by using heuristic methods that were inspired by nature (Whale and Gray Wolf Optimization Algorithms) and Chi-square method for feature selection. Then the reclassification was applied and the results were compared. While the success of the system with 600 features before the feature selection is 0.9214, the performance ratio of the 100 featured models created later is figured higher (0.94095 - 0.93833- 0.93619).