Scopus İndeksli Yayınlar Koleksiyonu / Scopus Indexed Publications Collection
Permanent URI for this collectionhttps://hdl.handle.net/11147/7148
Browse
23 results
Search Results
Article Recognition of Counterfactual Statements in Turkish(Assoc Computing Machinery, 2025) Acar, Ali; Tekir, SelmaCounterfactual statements are examples of causal reasoning as they describe events that did not happen and, optionally, those events' consequences if they happened. SemEval-2020 introduces the counterfactual detection (CFD) task and shares an English dataset. Since then, a set of datasets has been released in English, German, and Japanese as part of Amazon product reviews. This work releases the first Turkish corpus of counterfactuals (TRCD). The data collection process is driven by a clue phrase list of counterfactuals, mainly in the form of verb inflections in Turkish. We use clue phrase-based filtering to collect sentences from the Turkish National Corpus (TNC). On the other hand, half of the collection is subject to random word filtering to avoid selection bias due to clue phrases. After the human annotation process with an Inter Annotator Agreement of 0.65, we have 5000 sentences, of which 12.8% contain counterfactual statements. Furthermore, we provide a comprehensive baseline of transformer-based models by testing the effect of clue phrases, cross-lingual performance comparisons using the available CFD datasets, and zero-shot cross-lingual classification experiments using fine-tuning on the different combinations of the existing datasets. The results confirm that TRCD is compatible with the other CFD datasets. Moreover, fine-tuning a Turkish-specific model (BERTurk) performs better than the multilingual alternatives (mBERT and XLM-R). BERTurk is more robust to clue phrase masking. This result emphasizes the importance of a language-specific tokenizer for contextual understanding, especially for low-resource languages. Finally, our qualitative analysis gives insights into errors by different models.Conference Object Citation - WoS: 4Citation - Scopus: 8LGPsolver - Solving Logic Grid Puzzles Automatically(Assoc Computational Linguistics-acl, 2020) Jabrayilzade, Elgun; Tekir, SelmaLogic grid puzzle (LGP) is a type of word problem where the task is to solve a problem in logic. Constraints for the problem are given in the form of textual clues. Once these clues are transformed into formal logic, a deductive reasoning process provides the solution. Solving logic grid puzzles in a fully automatic manner has been a challenge since a precise understanding of clues is necessary to develop the corresponding formal logic representation. To meet this challenge, we propose a solution that uses a DistilBERT-based classifier to classify a clue into one of the predefined predicate types for logic grid puzzles. Another novelty of the proposed solution is the recognition of comparison structures in clues. By collecting comparative adjectives from existing dictionaries and utilizing a semantic framework to catch comparative quantifiers, the semantics of clues concerning comparison structures are better understood, ensuring conversion to correct logic representation. Our approach solves logic grid puzzles in a fully automated manner with 100% accuracy on the given puzzle datasets and outperforms state-of-the-art solutions by a large margin.Conference Object A News Chain Evaluation Methodology Along With a Lattice-Based Approach for News Chain Construction(Association for Computational Linguistics (ACL), 2017) Toprak, Mustafa; Özkahraman,Ö.; Tekir, SelmaChain construction is an important requirement for understanding news and establishing the context. A news chain can be defined as a coherent set of articles that explains an event or a story. There's a lack of well-established methods in this area. In this work, we propose a methodology to evaluate the "goodness" of a given news chain and implement a concept latticebased news chain construction method by Hossain et al. The methodology part is vital as it directly affects the growth of research in this area. Our proposed methodology consists of collected news chains from different studies and two "goodness" metrics, minedge and dispersion coefficient respectively. We assess the utility of the lattice-based news chain construction method by our proposed methodology. © EMNLP 2017.All right reserved.Conference Object Improvements on a Multi-Task Bert Model(Ieee, 2024) Agrali, Mahmut; Tekir, SelmaPre-trained language models have introduced significant performance boosts in natural language processing. Fine-tuning of these models using downstream tasks' supervised data further improves the acquired results. In the fine-tuning process, combining the learning of tasks is an effective approach. This paper proposes a multi-task learning framework based on BERT. To accomplish the tasks of sentiment analysis, paraphrase detection, and semantic text similarity, we include linear layers, a Siamese network with cosine similarity, and convolutional layers to the appropriate places in the architecture. We conducted an ablation study using Stanford Sentiment Treebank (SST), Quora, and SemEval STS datasets for each task to test the framework and its components' effectiveness. The results demonstrate that the proposed multi-task framework improves the performance of BERT. The best results obtained for sentiment analysis, paraphrase detection, and semantic text similarity are accuracies of 0.534 and 0.697 and a Pearson correlation coefficient of 0.345.Article Citation - WoS: 2Citation - Scopus: 2Enrichment of Turkish Question Answering Systems Using Knowledge Graphs(Tubitak Scientific & Technological Research Council Turkey, 2024) Ciftci, Okan; Soygazi, Fatih; Tekir, SelmaRecent capabilities of large language models (LLMs) have transformed many tasks in Natural Language Processing (NLP), including question answering. The state-of-the-art systems do an excellent job of responding in a relevant, persuasive way but cannot guarantee factuality. Knowledge graphs, representing facts as triplets, can be valuable for avoiding errors and inconsistencies with real-world facts. This work introduces a knowledge graph-based approach to Turkish question answering. The proposed approach aims to develop a methodology capable of drawing inferences from a knowledge graph to answer complex multihop questions. We construct the Beyazperde Movie Knowledge Graph (BPMovieKG) and the Turkish Movie Question Answering dataset (TRMQA) to answer questions in the movie domain. We evaluate our proposed question answering pipeline against a baseline study. Furthermore, we compare it with a question answering system built upon GPT-3.5 Turbo to answer the 1-hop questions from TRMQA. The experimental results confirm that link prediction on a knowledge graph is quite effective in answering questions that require reasoning paths. Finally, we provide insights into the pros and cons of the provided solution through a qualitative study.Article Asking the Right Questions To Solve Algebraic Word Problems(TÜBİTAK - Türkiye Bilimsel ve Teknolojik Araştırma Kurumu, 2022) Çelik, Ege Yiğit; Orulluoğlu, Zeynel; Mertoğlu, Rıdvan; Tekir, SelmaWord algebra problems are among challenging AI tasks as they combine natural language understanding with a formal equation system. Traditional approaches to the problem work with equation templates and frame the task as a template selection and number assignment to the selected template. The recent deep learning-based solutions exploit contextual language models like BERT and encode the natural language text to decode the corresponding equation system. The proposed approach is similar to the template-based methods as it works with a template and fills in the number slots. Nevertheless, it has contextual understanding because it adopts a question generation and answering pipeline to create tuples of numbers, to finally perform the number assignment task by custom sets of rules. The inspiring idea is that by asking the right questions and answering them using a state-of-the-art language model-based system, one can learn the correct values for the number slots in an equation system. The empirical results show that the proposed approach outperforms the other methods significantly on the word algebra benchmark dataset alg514 and performs the second best on the AI2 corpus for arithmetic word problems. It also has superior performance on the challenging SVAMP dataset. Though it is a rule-based system, simple rule sets and relatively slight differences between rules for different templates indicate that it is highly probable to develop a system that can learn the patterns for the collection of all possible templates, and produce the correct equations for an example instance.Article Citation - WoS: 1Citation - Scopus: 1Author Reputation Measurement on Question and Answer Sites by the Classification of Author-Generated Content(World Scientific Publishing, 2021) Sezerer, Erhan; Tenekeci, Samet; Acar, Ali; Baloğlu, Bora; Tekir, SelmaIn the field of software engineering, practitioners' share in the constructed knowledge cannot be underestimated and is mostly in the form of grey literature (GL). GL is a valuable resource though it is subjective and lacks an objective quality assurance methodology. In this paper, a quality assessment scheme is proposed for question and answer (Q&A) sites. In particular, we target stack overflow (SO) and stack exchange (SE) sites. We model the problem of author reputation measurement as a classification task on the author-provided answers. The authors' mean, median, and total answer scores are used as inputs for class labeling. State-of-the-art language models (BERT and DistilBERT) with a softmax layer on top are utilized as classifiers and compared to SVM and random baselines. Our best model achieves 63.8% accuracy in binary classification in SO design patterns tag and 71.6% accuracy in SE software engineering category. Superior performance in SE software engineering can be explained by its larger dataset size. In addition to quantitative evaluation, we provide qualitative evidence, which supports that the system's predicted reputation labels match the quality of provided answers.Article Citation - WoS: 2Citation - Scopus: 2Incorporating Concreteness in Multi-Modal Language Models With Curriculum Learning(MDPI, 2021) Sezerer, Erhan; Tekir, SelmaOver the last few years, there has been an increase in the studies that consider experiential (visual) information by building multi-modal language models and representations. It is shown by several studies that language acquisition in humans starts with learning concrete concepts through images and then continues with learning abstract ideas through the text. In this work, the curriculum learning method is used to teach the model concrete/abstract concepts through images and their corresponding captions to accomplish multi-modal language modeling/representation. We use the BERT and Resnet-152 models on each modality and combine them using attentive pooling to perform pre-training on the newly constructed dataset, which is collected from the Wikimedia Commons based on concrete/abstract words. To show the performance of the proposed model, downstream tasks and ablation studies are performed. The contribution of this work is two-fold: A new dataset is constructed from Wikimedia Commons based on concrete/abstract words, and a new multi-modal pre-training approach based on curriculum learning is proposed. The results show that the proposed multi-modal pre-training approach contributes to the success of the model.Conference Object Citation - WoS: 2Citation - Scopus: 4Çok-etiketli Film Türü Sınıflandırması için Türkçe Konu Modellemesi Veri Kümesi(Institute of Electrical and Electronics Engineers, 2020) Jabrayilzade, Elgün; Poyraz Arslan, Algın; Para, Hasan; Polatbilek, Ozan; Sezerer, Erhan; Tekir, SelmaStatistical topic modeling aims to assign topics to documents in an unsupervised way. Latent Dirichlet Allocation (LDA) is the standard model for topic modeling. It shows good performance on document collections, documents being relatively long texts but it has poor performance on short texts. Topic modeling on short texts is on the rise due to the potential of social media. Thus, approaches that are able to nd topics on short texts as well as long texts are sought. However, there is a lack of datasets that include both long and short texts which have the same ground-truth categories. In this work, we release a Turkish movie dataset which contain both short lm descriptions and long subscripts where lm genre can be considered as topic. Furthermore, we provide multi-label movie genre classication results using a Feed Forward Neural Network (FFNN) taking LDA document-topic or Doc2Vec dense representations. © 2020 IEEE.Conference Object Doğal Dil Çıkarımı Modellerinde Bert Vektörlerinin Başarım Değerlendirmesi(Institute of Electrical and Electronics Engineers Inc., 2021) Oğul, İskender Ülgen; Tekir, SelmaDoğal dil çıkarımı, düşünce ifade eden cümlelerin arasındaki ilişkiyi; karşıtlık, gerekseme veya tarafsızlık olarak sınıflandırmayı hedefler. Sınıflandırma görevini gerçekleştirmek için metinsel kaynaklar, vektör ya da gömme olarak adlandırılan matematiksel gösterimlere dönüştürülür. Bu çalışmada, hem statik (Glove, OntoNotes5) hem de bağlamsal (BERT) kelime gömme yöntemleri kullanılmıştır. Fikirsel cümleler arasındaki mantıksal ilişkilerin sınıflandırılması zordur zira cümleler karmaşık gramer yapılarına sahiptir ve cümlelerin işlenerek mantıksal gösterimlere dönüştürülmesi geleneksel doğal dil işleme çözümleri ile yetersiz kalmaktadır. Bu çalışma, sınıflandırma görevini gerçekleştirmek için ayrıştırılabilir ilgi ve doğal dil çıkarımı için gelişmiş LSTM (ESIM) derin öğrenme modellerini kullanmıştır. En iyi sonuç olan %88 doğruluk değeri SNLI veri kümesi üzerinde ESIM-BERT ile elde edilmiştir.
- «
- 1 (current)
- 2
- 3
- »