Enhancing Free-Text Information Extraction For Improved Knowledge Acquisition and Retrieval from Social Media Networks

Abstract

The exponential growth of user-generated content on social media has led to an urgent need for intelligent systems that can extract structured knowledge from noisy, informal, and emotionally charged text. Traditional information extraction approaches often struggle with the nuances of free-text data, including code-switching, slang, sarcasm, and multimodal signals. This paper presents a novel framework that enhances free-text information extraction by integrating emotion-aware representation (EMO), multi-context attention mechanisms (MCAM), and multimodal fusion to bridge the gap between unstructured text and structured knowledge. Our approach jointly addresses named entity recognition, relation extraction, and knowledge retrieval from multimodal social media posts. Extensive experiments on benchmark datasets—MMHS-DS1, Hatebase-DSII, and HSOL-DSIII—demonstrate significant improvements over baseline models in F1 score, precision@1, and mean reciprocal rank (MRR). The proposed system also exhibits cross-domain robustness and provides a viable foundation for real-world applications such as hate speech detection, misinformation analysis, and dynamic knowledge graph population.