Abstract Speech emotion recognition is a challenging task in speech processing field. For this reason, feature extraction process has a crucial importance to demonstrate and process the speech signals. In this work, we represent a model, which feeds raw audio files directly into the deep neural networks without any feature extraction stage for the recognition of emotions utilizing six different data sets, namely, The Berlin Database of Emotional Speech (EMO-DB), Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), Toronto Emotional Speech Database (TESS), Crowd-sourced Emotional Multimodal Actors (CREMA), Surrey Audio-Visual Expressed Emotion (SAVEE), and TESS+RAVDESS. To demonstrate the contribution of proposed model, the performance of traditional feature extraction techniques namely, mel-scale spectogram, mel-frequency cepstral coefficients, are blended with machine learning algorithms, ensemble learning methods, deep and hybrid deep learning techniques. Support vector machine, decision tree, naive Bayes, random forests models are evaluated as machine learning algorithms while majority voting and stacking methods are assessed as ensemble learning techniques. Moreover, convolutional neural networks, long short-term memory networks, and hybrid CNN-LSTM model are evaluated as deep learning techniques and compared with machine learning and ensemble learning methods. To demonstrate the effectiveness of proposed model, the comparison with state-of-the-art studies are carried out. Based on the experiment results, CNN model excels existent approaches with 95.86% of accuracy for TESS+RAVDESS data set using raw audio files, thence determining the new state-of-the-art. The proposed model performs 90.34% of accuracy for EMO-DB with CNN model, 90.42% of accuracy for RAVDESS with CNN model, 99.48% of accuracy for TESS with LSTM model, 69.72% of accuracy for CREMA with CNN model, 85.76% of accuracy for SAVEE with CNN model in speaker-independent audio categorization problems.

Data wisdom is the ability to think critically about data and create data-based judgments by combining domain, mathematical, and methodological expertise with experience, comprehension, common sense, insight, and sound judgment. This special issue focuses on data wisdom which is a mix of mathematical, scientific, and humanistic abilities and combines science with art. In this special issue, we invited high-quality and unpublished work in computing, intelligence, data analytics, and related areas, demonstrating both research contributions and solving real-world problems. We selected top papers from (http://iccida.net/index.htmlICCIDA 2023). ICCIDA 2023 was organized by Universidad de Castilla-La Mancha, Spain and Istinye University, Türkiye. The conference served as an interdisciplinary forum that may help the research community to take a step forward and share the research findings. This special issue focuses on-data analysis and visualization (e.g., EDA, BI, statistical and mathematical modelling, computer vision from different domains such as remote sensing & GIS, medicine, and military, artificial intelligence and machine learning, big data applications, algorithms, and systems, IoT, VR, digital twins and metaverse, distributed intelligence, smart communication systems and networks for big data, software engineering for big data, AI for cyber-security, other topics related to computing, intelligent systems, and data analytics. From around 21 submitted articles to this particular section, one paper was selected based on the reviews. Each paper was reviewed by at least two reviewers and went through at least two rounds of reviews. The brief contribution of the paper is discussed below. Nagihan Yagmur, Idiris Dag, and Hasan Temurtas, “A new computer-aided diagnostic method for classifying anaemia disease: Hybrid use of Tree Bagger and metaheuristics”. Anaemia occurs when the blood doesn't have enough haemoglobin or red blood cells. In their study, they use the hemogram data of 15,300 patients obtained from Tokat Gaziosmanpaşa University Faculty of Medicine. The dataset consists of records of individuals with Hgb-anaemia, vitamin B12 deficiency anaemia, iron deficiency anaemia, folate deficiency anaemia, and non-anaemia. They propose six different hybrid models for anaemia disease prediction and classification using TreeBagger, Crow Search Algorithm (CSA), Chicken Swarm Optimization Algorithm (CSO) and JAYA methods. The result of the study is expected to be helpful for medical students and doctors in predicting five different types of anaemia. The guest editors hope that the research contributions and findings in this special issue will benefit the readers by enhancing their knowledge and encouraging them to work on various aspects of intelligent systems and applications. We want to express our sincere thanks to the editor-in-chief for allowing us to organize this particular issue. The editorial office staff are excellent, and thanks for their support. We are also thankful to all the authors who made this special issue possible and to the reviewers for their thoughtful contributions.

Nowadays, detection of mispronunciations and providing feedback to users are provided by systems called computer-assisted language learning (CALL). In this study, we propose a model that detects the correct pronunciation of Arabic phonemes and supports Arabic language learning by blending it with deep and hybrid deep learning algorithms and mel-frequency cepstrum coefficients (MFCC) and Mel spectrogram feature extraction methods. In order to demonstrate the performance of the proposed model, 29 letters in the Arabic alphabet, 7 of which are hafiz, are voiced by a total of 10 different people. The amount of data has been increased by using the methods of adding noise, time shifting, time stretching, pitch shifting, and its performance on the model has been observed. Experiment results show that the combination of Mel-spectrogram and ConvLSTM exhibit the best results with 97.3% of accuracy when compared with the studies in the literature.

Stocks give significant clues to the investors, analysts, and researchers for the purpose of predicting the direction of stock market. In this work, we propose financial sentiment analysis to forecast the direction of Borsa Istanbul (BIST100) by investigating user comments from Twitter about high volume stocks in BIST100. For this purpose, Word2Vec, GloVe, and FastText as word embedding models, Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long Short-Term Memory Networks (LSTMs) as deep learning algorithms, Bidirectional Encoder Representations from Transformers (BERT), Embeddings from Language Models (ELMo), and Universal Language Model Fine-tuning (ULMFiT) as new generation deep contextualized word embedding models are employed for the classification task. To the best of our knowledge, this is the first study to predict the direction of stock market BIST100 using Twitter sentiments and new generation deep contextualized word embedding models. Another novel side of this study is also the inclusion of traditional word embedding models and deep learning approaches to the proposed system for the purpose of comparing the performance of new generation deep contextualized word embedding models. To demonstrate the contribution of proposed model, nine different bank stocks having large volume in stock market which can be orient the direction of BIST100 are evaluated. Extensive experiments show that the usage of ELMo as a new generation deep contextualized word embedding model on Twitter dataset exhibits remarkable classification results for predicting the direction of stock market.

Customer churn is one of the most important problems for many industries, including banking, telecommunications, and gaming. In the gaming market, it is observed that the demand on game applications rises with the usage of mobile devices such as smartphones. Because of this, it is important to predict when players tend to leave a game. Studies so far focus on churn prediction in mobile or online games by analyzing demographic, economic, and behavioral data about their customers. In this work, we introduce a sentiment analysis-based churn prediction model in mobile games using word embedding models and deep learning algorithms. To the best of our knowledge, this is the first study to evaluate the churn tendency of customers by analyzing sentiments of players from their comments on games using deep learning and word embedding models. For this purpose, we use deep learning algorithms for classification and word embedding models for text representation. The applied deep learning algorithms include convolutional neural networks, recurrent neural networks, long short-term memory networks. Word2Vec, GloVe, and FastText word embedding models are employed for text representation. To demonstrate the impact of proposed model, comprehensive experiments are carried out on Turkish four different game datasets. The experiment results show that sentiment analysis of players in mobile games can be powerful indicator in terms of predicting customer churn.

Text data mining is the process of extracting and analyzing valuable information from text. A text data mining process generally consists of lexical and syntax analysis of input text data, the removal of non-informative linguistic features and the representation of text data in appropriate formats, and eventually analysis and interpretation of the output. Text categorization, text clustering, sentiment analysis, and document summarization are some of the important applications of text mining. In this study, we analyze and compare the performance of text categorization by using different single classifiers, an ensemble of classifiers, a neural probabilistic representation model called word2vec on English texts. The neural probabilistic based model namely, word2vec, enables the representation of terms of a text in a new and smaller space with word embedding vectors instead of using original terms. After the representation of text data in new feature space, the training procedure is carried out with the well-known classification algorithms, namely multivariate Bernoulli naïve Bayes, support vector machines and decision trees and an ensemble algorithm such as bagging, random subspace and random forest. A wide range of comparative experiments are conducted on English texts to analyze the effectiveness of word embeddings on text classification. The evaluation of experimental results demonstrates that an ensemble of algorithms models with word embeddings performs better than other classification algorithms that uses traditional methods on English texts.

The basic idea behind the classifier ensembles is to use more than one classifier by expecting to improve the overall accuracy. It is known that the classifier ensembles boost the overall classification performance by depending on two factors namely, individual success of the base learners and diversity. One way of providing diversity is to use the same or different type of base learners. When the same type of base learners is used, the diversity is realized by using different training data subsets for each of base classifiers. When different type of base classifiers used to achieve diversity, then ensemble system is called heterogeneous. In this paper, we focus on the heterogonous ensembles that use different types of base learners. An ensemble system based on classification algorithms, naïve Bayes, support vector machine and random forest is used to measure the effectiveness of heterogeneous classifier ensembles by conducting experiments on Turkish texts. Experiment results demonstrate that the usage of heterogeneous ensembles improves classification performance for Turkish texts and encourages to evaluate the impact of heterogeneous ensembles for the other agglutinative languages.

Text categorization has become more and more popular and important problem day by day because of the large proliferation of documents in many fields. To come up with this problem, several machine learning techniques are used for categorization such as naîve Bayes, support vector machines, artificial neural networks, etc. In this study, we concentrate on ensemble of multiple classifiers instead of using only a single one. We perform a comparative analysis of the impact of the ensemble techniques for text categorization domain. To carry out this, the same type of base classifiers but diversified training sets are used which is referred as homogenous ensembles. In order to diversify the training dataset, various ensemble algorithms are utilized such as Bagging, Boosting, Random Subspace and Random Forest. Multivariate Bernoulli Naîve Bayes is preferred as a base classifier due to its superior classification performance compared to the success of the other single classifiers. A wide range of comparative and extensive empirical studies are conducted on four widely-used datasets in text categorization domain in both Turkish and English. Finally, the effectiveness of ensemble algorithms is discussed.

Naïve Bayes is a commonly used algorithm in text categorization because of its easy implementation and low complexity. Naïve Bayes has mainly two event models used for text categorization which are multivariate Bernoulli and multinomial models. A very large number of studies choose multinomial model and Laplace smoothing just based on the assumption that it performs better than multivariate model under almost any conditions. This study aims to shed some light into this widely adopted assumption by analyzing Naïve Bayes event models and smoothing methods from a different perspective. To clarify the difference between events models of Naïve Bayes, their classification performance are compared on different languages - English and Turkish - datasets. Results of our extensive experiments demonstrate that superior performance of multinomial model does not observed all the time. On the other hand, multivariate Bernoulli model can perform well when combined with an appropriate smoothing method under different training data size conditions.

It has been shown that Latent Semantic Indexing (LSI) takes advantage of implicit higher-order (or latent) structure in the association of terms and documents. Higher order relations in LSI capture "latent semantics". Inspired by this, a novel Bayesian framework for classification named Higher Order Naïve Bayes (HONB), which can explicitly make use of these higher-order relations, has been introduced previously. We present a novel semantic smoothing method named Higher Order Smoothing (HOS) for the Naive Bayes algorithm. HOS is built on a similar graph based data representation of HONB which allows semantics in higher order paths to be exploited. Additionally, we take the concept one step further in HOS and exploited the relationships between instances of different classes in order to improve the parameter estimation when dealing with insufficient labeled data. As a result, we have not only been able to move beyond instance boundaries, but also class boundaries to exploit the latent information in higher-order paths. The results of our extensive experiments demonstrate the value of HOS on several benchmark datasets.