The rapid surge in social media usage has augmented the significance and value of data available on these platforms. As a result, analyzing community sentiment and opinions related to various topics and events using social media data has become increasingly crucial. However, the sheer volume of data produced on social media platforms surpasses human processing capabilities. Consequently, artificial intelligence-based models became frequently employed in social media analysis. In this study, deep learning (DL) and machine learning (ML) methods are applied to assess user opinions regarding airlines, and the effectiveness of these methods in social media analysis is comparatively discussed based on the performance results obtained. Due to the imbalanced nature of the dataset, synthetic data is produced using the Synthetic Minority Over-Sampling Technique (SMOTE) to enhance model performance. Before the SMOTE process, the dataset containing 14640 data points expanded to 27534 data points after the SMOTE process. The experimental results demonstrate that Support Vector Machines (SVM) achieved the highest performance among all methods with accuracy, precision, recall, and F-score values of 0.79 in the pre-SMOTE (imbalanced dataset). In contrast, Random Forest (RF) obtained the best performance among all methods, with accuracy, precision, recall, and F-score values of 0.88 in the post-SMOTE (balanced data set). Moreover, experimental findings demonstrate that SMOTE led to performance improvements in ML and DL models, ranging from a minimum of 3% to a maximum of 24% increase in F-Score metric.

In the recent past and in today’s world, the internet is advancing rapidly and is easily accessible; this growth has made the social media platforms such as Facebook, Instagram, Twitter, and LinkedIn widely used which produces big data. This requires both topic Detection applications in order to access the required information, as well as community detection practices in order to provide collective services to communities that can be referred to as individuals with similar interests and opinions over the same subject. Therefore, it is vital for researchers to conduct research on topic detection and community detection research areas in social networks and to develop methods and techniques for problem-solving. In this study, a systematic and in-depth literature review is provided on studies that conduct topic and community analysis on social media platforms to provide a comprehensive overview of the given areas. Most of the studies to be analyzed are selected from articles using machine learning-based models that are known to achieve successful results in practice. As a result of the analysis of these studies; it has been concluded that a single model cannot be proposed in the area of topic detection and that the appropriate model should only be selected or created in a problem-specific way, taking into account all the characteristics of the given problem, while the Louvain method seems to stand out with its results in terms of performance in the area of community detection.

Short-form video content has gained importance as a popular form of digital media due to the rising popularity of social media platforms and the decreasing attention spans of consumers. However, a major obstacle to popularity detection in short-form content is the heterogeneous nature of the data, encompassing textual, visual, and metadata components. To tackle this challenge, we propose FUSEPOP, a robust multi-modal architecture. The proposed framework utilizes ResNet-50 for visual feature extraction and XLM-RoBERTa for encoding multilingual textual information. FUSEPOP employs a mutual information-based modality weighting mechanism with logarithmic smoothing and a 0.7 weight ceiling to balance contributions from each input stream. Furthermore, FUSEPOP implements a robust stacked generalization strategy trained via stratified 5-fold cross-validation. This approach utilizes a logistic regression meta-learner to dynamically synthesize predictions from random forest, XGBoost, and a neural network-based classifier. Experimental results show that this architecture significantly outperforms benchmark models, achieving an accuracy of 0.980 and an average F1-score of 0.964 on the feature configuration selected for this study, and remains competitive on a literature-aligned alternative configuration. These findings confirm that the proposed model successfully detects popularity on short-form social media content.

Social media platforms are one of the most popular platforms for users to express their opinions and sentiments concerning various products, services, and organizations. This study presents the models that can successfully analyze sentiment through text on social media platforms using hybrid deep learning methods. The proposed models were applied to the "U.S. Airline Dataset" obtained from the Kaggle platform. After the text cleaning phase, the BERT embeddings were implemented on the dataset. The resulting preprocessed data set was used in the training of three deep learning-based hybrid algorithms, namely, CNN-GRU, BiLSTM-GRU, and RNN-GRU. The experimental results revealed that the best result was achieved by the CNN-GRU model, with an accuracy of 0.79 and an F-Score of 0.79.