Identification of Word Level Information Based Semantic Similarity Using Extended Glove Embeddings for Clustering and Classification Analysis

In this article, an enhanced methodology for document representation and classification leveraging the Extended GloVe (ExGloVe) algorithm is presented. The ExGloVe algorithm extends the traditional GloVe model by incorporating subword information and domain-specific adaptations, addressing limitations in capturing semantic nuances and domain-specific language variations. The incorporation of subword information enables the algorithm to better represent rare and out-of-vocabulary words, enhancing the expressiveness and robustness of the embeddings. Domain-specific adaptations tailor the embeddings to specific domains, capturing domain-specific semantics and improving performance in domain-specific tasks. Document-level embeddings obtained through the aggregation process are utilized as input features for clustering algorithms such as K-Means, DBSCAN, and Hierarchical Clustering, as well as classification models including Support Vector Machine, Logistic Regression, and Neural Networks. These models leverage the semantic richness encoded in the ExGloVe embeddings for effective document analysis. Experiments with various evaluation metrics are conducted to validate the efficacy of the proposed methodology in document similarity measurement, clustering, and classification tasks.