Visualization and Analysis of Social Network-Based Diverse Datasets Using Multi-Viewpoint Similarity Metrics

This article focuses on visualizing and analyzing diverse datasets, with a specific emphasis on text data extracted from social networks such as Twitter, utilizing topic models with multi-viewpoint similarity metrics. Traditional topic modeling encounters challenges due to reliance on uncommon terms in tweets and manual selection of clustering parameters based on distance metrics. To address these limitations, this work employs hybrid topic models including Visual Latent Schematic Indexing (VLSI), Visual Probabilistic Latent Schematic Indexing (VPLSI), Visual Latent Dirichlet Allocation (VLDA), and Visual Non-negative Matrix Factorization (VNMF), alongside multi-viewpoint cosine-based similarity metrics (MVPs). Additionally, both Euclidean and Cosine-based measures are utilized in comparative clustering results obtained from the proposed model. Validation of results involves a combination of internal and external validity indices, memory usage, and execution time to assess the effectiveness of each method in revealing meaningful topics within the dataset. Through comprehensive analysis, this study offers valuable insights into the effectiveness of visual hybrid techniques, particularly when applied to diverse datasets sourced from social media platforms like Twitter.