SMDKGG: A Socially Aware Metadata Driven Knowledge Graph Generation for Disaster Tweets

Abstract

Knowledge Graph Generation is a requisite in the modern era, where Knowledge is scarce whereas the data is exponentially surplus and extensively high. So, in this paper Knowledge Graph Generation framework, the SMDKGG has been proposed, is a Semantically Inclined Metadata driven model for Knowledge Graph Generation, which is a text to Knowledge Graph generation strategy where the document dataset is subjected to a lateral mapping of categories and alignment of generated ontologies with that of the categories for auxiliary Knowledge enhancement. Subsequently, applying Structural Topic Modeling (STM) and harvesting entities from LOD Cloud, NELL, DBPedia, and CYC Knowledge sources ensures the cohesiveness of the auxiliary Knowledge encompassed in the framework becomes extensively high, making the density of the harvested auxiliary Knowledge high. Apart from this, the Metadata generation and its classification using Transformers ensure that many more auxiliary knowledge substrates are selectively added to the framework. Its classification by a Deep Learning Transformer classifier ensures its ease of handling. Subsequently, the classification of the dataset using the AdaBoost, the encompassment of Twitter Semantic Similarity (TSS) for intra-class similarity computation, and the integration of Normalized Google Distance (NGD) with selective thresholds at various stages of the model ensure robust relevance computation mechanism for conceiving the best-in-class Knowledge Graph. Moreover, the optimization of the initial solution set is carried out using the Chemical Reaction metaheuristic optimization algorithm with NGD as the criterion function, ensuring the most relevant entities are alone conceived in the finalized Knowledge Graph. The proposed SMDKGG yields an overall Precision of 96.19%, Recall of 98.33%, Accuracy of 97.26%, F-Measure of 97.25%, and an FNR of 0.02.