Semantic Zero-Shot Text Classification for Social Media Using Knowledge Graph Embeddings

Abstract

The exponential growth of social media platforms has resulted in an unprecedented volume of unstructured textual data, making automated text classification a critical task in natural language processing. Traditional supervised learning approaches rely heavily on labeled datasets, which are often expensive, time-consuming, and impractical to obtain for every emerging category or domain. This limitation becomes particularly evident in dynamic environments such as social media, where new topics and trends emerge continuously. To address this challenge, this paper proposes a zero-shot text classification framework leveraging knowledge graph embedding techniques to classify unseen categories without requiring labeled training data.The proposed system integrates semantic knowledge extracted from structured knowledge graphs with advanced embedding techniques to establish meaningful relationships between textual inputs and predefined class labels. By representing both input text and category labels within a shared embedding space, the model effectively captures contextual and semantic similarities, enabling accurate classification even for previously unseen classes. The approach eliminates the dependency on extensive labeled datasets while maintaining competitive performance.The implementation is developed using Python and machine learning frameworks, with a web-based interface built using a lightweight backend architecture. The system processes social media text data, performs preprocessing steps such as tokenization and normalization, and maps the textual content into embedding vectors. Simultaneously, class labels are represented using knowledge graph embeddings, allowing the model to compute similarity scores and assign the most relevant category.Experimental evaluation demonstrates that the proposed zero-shot classification model achieves reliable performance across diverse social media datasets. The system effectively handles ambiguous and context-dependent text, providing meaningful classifications without prior exposure to labeled examples. Additionally, the integration of knowledge graph embeddings enhances semantic understanding, leading to improved classification accuracy compared to baseline approaches.The results indicate that the proposed method is scalable, adaptable, and suitable for real-time applications in dynamic environments. This research contributes to the advancement of zero-shot learning methodologies by combining structured knowledge representation with natural language processing techniques. The framework can be extended to various domains such as sentiment analysis, topic categorization, and content moderation, making it a valuable solution for modern data-driven applications.