Yazar "Yildirim, Savas" seçeneğine göre listele

Listeleniyor 1 - 13 / 13
  • Küçük Resim Yok
    Öğe
    Acquisition of Turkish meronym based on classification of patterns
    (Springer, 2016) Yildiz, Tugba; Diri, Banu; Yildirim, Savas
    The identification of semantic relations from a raw text is an important problem in Natural Language Processing. This paper provides semi-automatic pattern-based extraction of part-whole relations. We utilized and adopted some lexico-syntactic patterns to disclose meronymy relation from a Turkish corpus. We applied two different approaches to prepare patterns; one is based on pre-defined patterns that are taken from the literature, second automatically produces patterns by means of bootstrapping method. While pre-defined patterns are directly applied to corpus, other patterns need to be discovered first by taking manually prepared unambiguous seeds. Then, word pairs are extracted by their occurrence in those patterns. In addition, we used statistical selection on global data that is obtaining from all results of entire patterns. It is a whole-by-part matrix on which several association metrics such as information gain, T-score, etc., are applied. We examined how all these approaches improve the system accuracy especially within corpus-based approach and distributional feature of words. Finally, we conducted a variety of experiments with a comparison analysis and showed advantage and disadvantage of the approaches with promising results.
  • Küçük Resim Yok
    Öğe
    Building A Non-Personalized Recommender System by Learning Product and Basket Representation
    (IEEE, 2020) Yildirim, Savas; Soyler, Sebnem Gunes; Akarsu, Ozgur
    In this paper, we addressed the problem of learning product and basket representation for a non-personalized recommendation system where the baskets do not have a specific owner. The recommendation models tend to exploit as much information as possible along with basket patterns to improve performance. We focus on the representation problem for the baskets without any customer information. Deep learning-based architectures have solved many representation problems such as natural language processing (NLP) and computer vision (CV) so far. While the NLP model takes a bag of words as input, the recommendation models take a basket of products as input. The learning algorithm uses co-occurrence information and therefore exploits the idea that the things that appear in a similar environment share similar meaning. But traditional representation approaches such as one-hot encoding have dimensionality problems when the number of entities increases. On the other hand, neural models can solve this dimensionality curse and transform each entity into a short and dense vector, namely embeddings. We successfully designed unsupervised and supervised architectures to solve the product and basket embeddings for a recommendation engine. Our experiments show that the proposed deep learning architecture showed better performance than baseline approaches in terms of many metrics. We also discussed and addressed many product representation related problems throughout the paper.
  • Küçük Resim Yok
    Öğe
    A cascaded framework for identification and extraction of antonym for Turkish language
    (Springer, 2019) Yildiz, Tugba; Yildirim, Savas
    Identification and extraction of semantic relations are challenging tasks in Natural Language Processing. In this paper, we design and propose three different models for the two separate tasks of identifying and extracting antonyms. In the first model, we develop two methods to identify antonyms: the first method consists of a probabilistic approach to calculate the probability of a given target/candidate pair being an antonym, whereby two distinct scoring functions are proposed to decide about the correct candidate for each target word; the second method consists of learning word embeddings and measuring embedding similarity to identify antonym pairs. In the second proposed model, we represent target/candidate pairs by a set of features that are compatible with those that are used by a supervised machine learning algorithm. The first and second models both especially well-suited for the identification of antonymy. In the last and third model, we adopt a minimally supervised bootstrapping approach, which operates by starting with a few antonym pairs and producing, thereafter, both seeds and patterns in an iterative fashion. Our study is deemed to be a significant contribution toward enriching the lexicon of the Turkish language.
  • Küçük Resim Yok
    Öğe
    Classification of Hot News for Financial Forecast Using NLP Techniques
    (IEEE, 2018) Yildirim, Savas; Jothimani, Dhanya; Kavaklioglu, Can; Basar, Ayse
    Complex dynamics of stock market could be attributed to various factors ranging from company's financial ratios to investors' sentiment and reaction to Financial news. The paper aims to classify Financial news articles as hot (significant) and non-hot (non-significant). The study is carried out using Dow Jones newswires text feed for a period of four years spanning from 2013 till 2017. Bag-of-ngrams appraoch and Term Frequency-Inverse Document Frequency (TF-IDF) were used for text representation and text weighting, respectively. Four linear classifiers, namely, Logistic Regression (LR), Support Vector Machine (SVM), k Nearest Neighbours (kNN) and multinomial Naive Bayes (mNB) were used. Grid search was used for hyperparameter optimisation. Performance of the classifiers was evaluated using five measures, namely, success rate, precision, recall, F1 measure and area under receiver operating characteristics curve. LR and SVM outperformed other models in terms of all five performance measures for both Bag-of-ngrams model and Bag-of-ngrams model with TF-IDF approach. Use of TF-IDF improved performance of the classifiers, especially, in case of mNB. This study serves as a stepping stone in identification of important/relevant news, which could used as predictors for stock price forecasting.
  • Küçük Resim Yok
    Öğe
    Comparing Deep Neural Networks to Traditional Models for Sentiment Analysis in Turkish Language
    (Springer-Verlag Singapore Pte Ltd, 2020) Yildirim, Savas
    Traditional bag-of-words (BOW) draws advantage from distributional theory to represent document. The drawback of BOW is high dimensionality. However, this disadvantage has been solved by various dimensionality reduction techniques such as principal component analysis (PCA) or singular value decomposition (SVD). On the other hand, neural network-based approaches do not suffer from dimensionality problem. They can represent documents or words with shorter vectors. Especially, recurrent neural network (RNN) architectures have gained big attractions for short sequence representation. In this study, we compared traditional representation (BOW) with RNN-based architecture in terms of capability of solving sentiment problem. Traditional methods represent text with BOWapproach and produce one-hot encoding. Further well-known linear machine learning algorithms such as logistic regression and Naive Bayes classifier could learn the decisive boundary in the data points. On the other hand, RNN-based models take text as a sequence of words and transform the sequence using hidden and recurrent states. At the end, the transformation finally represents input text with dense and short vector. On top of it, a final neural layer maps this dense and short representation to a sentiment of a list. We discussed our findings by conducting several experiments in depth. We comprehensively compared traditional representation and deep learning models by using a sentiment benchmark dataset of five different topics such as books and kitchen in Turkish language.
  • Küçük Resim Yok
    Öğe
    Deep Learning Approaches for Sentiment Analysis on Financial Microblog Dataset
    (IEEE, 2019) Yildirim, Savas; Jothimani, Dhanya; Kavaklioglu, Can; Basar, Ayse
    Sentiment analysis of financial news and social media messages along with movement of stock prices could aid in improving the forecasting accuracy of stock prices. In this regard, we aim to perform sentiment analysis of a financial microblog, namely, StockTwits. We carried out the analysis on labelled messages of twelve stocks for a period of five months ranging from May 2019 to September 2019 using various Deep Learning (DL) approaches. We compared the performance of the DL classifiers with traditional machine learning approaches. Long Short Term Memory (LSTM) model and its variations such as bidirectional LSTM and bidrirectional LSTM with dropout outperformed other classifiers. Though use of dropout mechanism did not improve the performance of the model but there was a decrease in bias and variance. Further, we evaluated the performance of various optimizers such as rmsprop, adam, adagrad, adamax and nadam on LSTM. The success rate of all optimizers was similar.
  • Küçük Resim Yok
    Öğe
    A Knowledge-Poor Approach to Turkish Text Categorization
    (Springer-Verlag Berlin, 2014) Yildirim, Savas
    Document categorization is a way of determining a category for a given document. Supervised methods mostly rely on a training data and rich linguistic resources that are either language-specific or generic. This study proposes a knowledge-poor approach to text categorization without using any sets of rules or language specific resources such as part-of-speech tagger or shallow parser. Knowledge-poor here refers to lack of a reasonable amount of background knowledge. The proposed system architecture takes data as-is and simply separates tokens by space. Documents represented in vector space models are used as training data for many machine learning algorithm. We empirically examined and compared a several factors from similarity metrics to learning algorithms in a variety of experimental setups. Although researchers believe that some particular classifiers or metrics are better than others for text categorization, the recent studies disclose that the ranking of the models purely depends on the class, experimental setup and domain as well. The study features extensive evaluation, comparison within a variety of experiments. We evaluate models and similarity metrics for Turkish language as one of the agglutinative language especially within poor-knowledge framework. It is seen that output of the study would be very beneficial for other studies.
  • Küçük Resim Yok
    Öğe
    Learning-based pronoun resolution for Turkish with a comparative evaluation
    (Academic Press Ltd- Elsevier Science Ltd, 2009) Kilicaslan, Yilmaz; Guner, Edip Serdar; Yildirim, Savas
    The aim of this paper is twofold. On the one hand, it attempts to explore several machine learning models for pronoun resolution in Turkish, a language not sufficiently studied with respect to anaphora resolution and rarely being subjected to machine learning experiments. On the other hand, this paper offers an evaluation of the classification performances of the learning models in order to gain insight into the question of how to match a model to the task at hand. In addition to the expected observation that each model should be tuned to an optimum level of expressive power so as to avoid underfitting and overfitting, the results also suggest that non-linear models properly tuned to avoid overfitting outperform linear ones when applied to the data used in our experiments. (C) 2008 Elsevier Ltd. All rights reserved.
  • Küçük Resim Yok
    Öğe
    Listening to the organization: change evaluation with discourse analysis
    (Emerald Group Publishing Ltd, 2018) Akarsu, Ozgur; Gencer, Mehmet; Yildirim, Savas
    Purpose Change is continuous and leaves many digital traces in contemporary organizations, while research on change usually lacks such continuity. The purpose of this paper is to test and explore the claim that change can be monitored through employee discourse. In doing so, the authors introduce basic text mining methods to detect prevailing keywords and their changes over time. Such monitoring of content and its change promises a continuous feedback and improvement for change management efforts. Design/methodology/approach The authors use a mixed research design, combining an ethnographic approach with digital methods. The quantitative element of the method involves applying text mining techniques to a document corpus that is representative of people in organizations, and is originally collected as part of a relatively common performance management system. The findings about discursive categories and their change patterns through time are then combined with observations and secondary information about change management for interpretation. Findings By combining these measurements with additional information about the change program in focus, the authors develop an interpretation of the dynamics of organizational change. Results showed that even in a successfully implied change effort that realize the planned targets, change does not occur directly and fully, with some elements of discourse being more persistent than others. Research limitations/implications Method of the research presents a new way of monitoring discursive change. Its incorporation into practice potentially allows for timely correction of change efforts and increasing possibility of success. Originality/value This research provides a framework for understanding how, and to what extent, planned change efforts effect organizations. Furthermore, the method developed in this research presents an innovative approach to monitor discursive change and timely managerial intervention.
  • Küçük Resim Yok
    Öğe
    Pronoun Resolution in Turkish Using Decision Tree and Rule-Based Learning Algorithms
    (Springer-Verlag Berlin, 2009) Yildirim, Savas; Kilicaslan, Yilmaz; Yildiz, Tugba
    This paper reports on the results of some pronoun resolution experiments performed by applying a decision tree and a rule-based algorithm on an annotated Turkish text. The text has been compiled mostly from various popular child stories in a semi-automatic way. A knowledge-lean learning model has been devised using only nine most commonly employed features. An evaluation and comparison of the performances achieved with the two different algorithms is offered in terms of the recall, precision and f-measure metrics.
  • Küçük Resim Yok
    Öğe
    Sentiment Analysis Using Learning Approaches over Emojis for Turkish Tweets
    (IEEE, 2018) Velioglu, Riza; Yildiz, Tugba; Yildirim, Savas
    With the rise of the usage and interest on social media platforms, emojis have become an increasingly important part of the written language and one of the most important signals for micro-blog sentiment analysis. In this paper, we employed and evaluated classification models using two different representations based on hag-of-words and fasilText to address the problem of sentiment analysis over emojis/emoticons for Turkish positive, negative and neutral tweets. At first, the hag of-words approach is used as a simple and efficient baseline method for tweet representation, where the classifiers such as Naive Bayes, Logistic Regression, Support Vector Machines, Decision Trees have been applied to these tweets. Secondly, we utilized fastText to represent tweets as word n-grams for sentiment analysis problem. The results show that there is no significant difference between the two models. While fastText shows 79% and the Linear Regression classifier obtains 77% Fl-score for binary classification, fastText performs 62% and Linear Regression has 58% Fl-score for multi-class classification. This study is considered as the first study that contributes to the literature by applying different vector representations such as bag-of-words and fastText to predict Turkish tweets over emojis. This study can also be utilized to predict emojis on social media context in the future.
  • Küçük Resim Yok
    Öğe
    Sequence Labeling for Disambiguating Medical Abbreviations
    (Springernature, 2023) Cevik, Mucahit; Jafari, Sanaz Mohammad; Myers, Mitchell; Yildirim, Savas
    Abbreviations are unavoidable yet critical parts of the medical text. Using abbreviations, especially in clinical patient notes, can save time and space, protect sensitive information, and help avoid repetitions. However, most abbreviations might have multiple senses, and the lack of a standardized mapping system makes disambiguating abbreviations a difficult and time-consuming task. The main objective of this study is to examine the feasibility of sequence labeling methods for medical abbreviation disambiguation. Specifically, we explore the capability of sequence labeling methods to deal with multiple unique abbreviations in a single text. We use two public datasets to compare and contrast the performance of several transformer models pre-trained on different scientific and medical corpora. Our proposed sequence labeling approach outperforms the more commonly used text classification models for the abbreviation disambiguation task. In particular, the SciBERT model shows a strong performance for both sequence labeling and text classification tasks over the two considered datasets. Furthermore, we find that abbreviation disambiguation performance for the text classification models becomes comparable to that of sequence labeling only when postprocessing is applied to their predictions, which involves filtering possible labels for an abbreviation based on the training data.
  • Küçük Resim Yok
    Öğe
    A Study on Turkish Meronym Extraction Using a Variety of Lexico-Syntactic Patterns
    (Springer International Publishing Ag, 2016) Yildiz, Tugba; Yildirim, Savas; Diri, Banu
    In this paper, we applied lexico-syntactic patterns to disclose meronymy relation from a huge Turkish raw text. Once, the system takes a huge raw corpus and extract matched cases for a given pattern, it proposes a list of whole-part pairs depending on their co-occur frequencies. For the purpose, we exploited and compared a list of pattern clusters. The clusters to be examined could fall into three types; general patterns, dictionary-based pattern, and bootstrapped pattern. We evaluated how these patterns improve the system performance especially within corpusbased approach and distributional feature of words. Finally, we discuss all the experiments with a comparison analysis and we showed advantage and disadvantage of the approaches with promising results.