Esra Kahya Özyirmidokuz

Across digital platforms, large-scale assessment of customer sentiment has become integral to brand management, service recovery, and data-driven marketing in e-commerce. Still, most studies center on single-language settings, with bilingual and culturally diverse environments receiving comparatively limited attention. In this study, a bilingual sentiment analysis of consumer feedback on X (formerly Twitter) was conducted for three global quick-service restaurant (QSR) brands—McDonald’s, Burger King, and KFC—using 145,550 English tweets and 15,537 Turkish tweets. After pre-processing and leakage-safe augmentation for low-resource Turkish data, both traditional machine learning models (Naïve Bayes, Support Vector Machines, Logistic Regression, Random Forest) and a transformer-based deep learning model, BERT (Bidirectional Encoder Representations from Transformers), were evaluated. BERT achieved the highest performance (macro-F1 ≈ 0.88 in Turkish; ≈0.39 in temporally split English), while Random Forest emerged as the strongest ML baseline. An apparent discrepancy was observed between pseudo-label agreement (Accuracy > 0.95) and human-label accuracy (EN: 0.75; TR: 0.49), indicating the limitations of lexicon-derived labels and the necessity of human validation. Beyond methodological benchmarking, linguistic contrasts were identified: English tweets were more polarized (positive/negative), whereas Turkish tweets were overwhelmingly neutral. These differences reflect cultural patterns of online expression and suggest direct managerial implications. The findings indicate that bilingual sentiment analysis yields brand-level insights that can inform strategic and operational decisions.

This study addresses the growing concern of digital game addiction (DGA) by developing an information system for detection and early intervention, to predict an individual’s propensity toward DGA. The study employed machine learning models to process unstructured data collected from online game platforms between May 1, 2023, and June 1, 2023. A total of 660 comments were automatically extracted. Following an expert review process, 245 ambiguous or inconsistent entries were removed, resulting in a final dataset of 415 textual comments for model training. The comments were manually annotated into two categories: addiction-related (1) and non-addiction-related (0) to ensure data consistency and reliability. Data augmentation techniques, such as BERT-based augmentation, deep translation, and synonym replacement, were applied to increase dataset diversity and enhance the model’s generalization capability. After extensive preprocessing, we trained and evaluated six machine learning models: XGBoost, Support Vector Machines (SVM), Logistic Regression, K-Nearest Neighbors (KNN), Decision Trees, and Random Forests. XGBoost consistently outperformed all other models, achieving the highest accuracy of 93.06% on the binary-labeled dataset and 92.49% on the expanded dataset. Random Forests and Logistic Regression also exhibited strong classification performance, making them viable alternatives depending on the requirements. In contrast, SVM and Decision Trees showed lower predictive reliability, indicating potential challenges in classifying text-based addiction indicators effectively. As a practical outcome, we developed an interface using the best-performing model, capable of performing preliminary diagnoses based on user-provided sentences. This innovative approach introduces a new dimension in addiction studies by presenting a practical tool for early detection and intervention in DGA.

The rapid depletion of global resources and the increasing challenges associated with waste management have underscored the importance of the circular economy—a framework designed to balance resource utilization, align production with consumption, and effectively address waste-related issues. Within this context, social complementary factors in waste management are crucial for overcoming these significant challenges. This study investigates the current state of solid waste management within the circular economy, with a particular focus on socio-cultural and behavioral dynamics, as well as the socio-technical implications of these processes. The research aims to elucidate the influence of socio-cultural dynamics on waste management practices. A comprehensive search of the Web of Science (WoS) database, using the keyword “solid waste management” up to August 10, 2024, yielded 33,247 relevant studies. To extract key themes from this extensive body of literature, Latent Dirichlet Allocation (LDA) topic modeling was applied. Subsequently, BERT-based deep learning methods were employed for text classification to identify which socio-cultural dynamics are most emphasized in the selected studies. The model achieved an overall accuracy of 87.5% in predicting the dominant topics. The findings from this study provide significant insights into the socio-cultural factors influencing waste management practices within the circular economy. By identifying these dynamics, the research contributes to a deeper understanding of how social and cultural factors can be leveraged to enhance the effectiveness of waste management strategies.

Learning objectives: The most important problem of dynamic societies is waste management (WM). For a healthy, happy and sustainable life, WM in cities should be carried out accurately, quickly and effectively with artificial intelligence-based systems. The most important negative output of developed societies is waste, the more production and consumption is made, the more waste will be generated. Based on the principle of Green energy, where and what kind of work should be done in WM will be put forward by practices.

Pazarlama ile ilgili bir yazılım sistem kalitesini artırmak için, tüketici geribildirimleri sistem tasarımı sırasında değerlendirilmelidir. Bu araştırmanın amacı tüketicilerin kullanacağı iyi-olma yazılımları ile ilgili gereksinimlerini keşfetmektir. Tüketicilerin yazılım ile mutluluklarını nasıl artırabilecekleri ve ikna teknolojileri ile oyunlaştırma teknikleri dahil tüketicileri sistemde tutan ve mutluluklarını etkileyen detaylar irdelenmektedir. Çalışmada karma araştırma yöntemi uygulandı. Yapılan tematik analiz sonucunda, katılımcıların bahsettikleri bir pazarlama iyi-olma uygulamasının mutlu eden özelliklerinin yanı sıra, işlevsel özellikleri, interaktif özellikleri ve ürün-fiyat özellikleri ayrı temalarda gruplandırıldı.

BACKGROUND Measuring online Turkish happiness requires a Turkish happiness dictionary which could reflect norms and social values more culturally and linguistically instead of using a translation-oriented method. Analyzing data without neglecting cultural characteristics will not be reliable. Turkish translation of an English word in the Affective Norms of English Words (ANEW) dictionary does not express the same feeling of a Turkish word. In addition, existing emotional dictionaries are not developed for specifically for the social networks with emoticons. OBJECTIVE This research presents the Turkish Happiness Index (THI) which is a set of psychological normative happiness scores to measure an average level of happiness of Turkish online unstructured large-scale data. A well-being informatics analytics research is also done by using THI. METHODS Turkish Happiness Index was completely generated on social networks. 20000 words were extracted with web text mining from social networks. Natural Language Processing algorithms were applied. After data reduction quantitative research methodology is applied. The happiness scores were based detected based on 667 participants’ subjective happiness levels and their thoughts about the 1874 Turkish words. Alexithymia scale was also used to identify the emotional awareness of the participants. The evaluations of the words were done in the dimension of valence using the Self-Assessment Manikin in an online platform. NLP was used to measure online Turkish happiness of data. Data was collected from Facebook with negative #war and positive #family hashtags in a duration of one month using a 3rd party software tool. Natural language processing algorithms including tokenization, transformation, filtering and stemming after converting data to documents. The happiness levels of the documents based on hashtags were determined using the Turkish Happiness Index dictionary. RESULTS THI which contains 345 words and their happiness scores in the Turkish language was developed. The THI is given in Appendix 1. We also put a comparison between words of dictionaries to understand the cultural differences. CONCLUSIONS THI provide researchers with standard materials through which they can automatically measure online happiness of Turkish large-scale data. THI can be used in in real-time big data analytics.