Evaluating Hybrid AI For Prediction Over Lung Cancer Knowledge Graphs

Abstract

Link prediction is of great importance in the field of knowledge graphs, as it plays a key role in facilitating knowledge discovery and supporting decision-making, especially in healthcare. Although knowledge graphs provide a structured representation of data, challenges arise from data integration and quality assurance issues. The presence of inaccuracies, outdated information and inconsistencies poses a threat to data quality, requiring ongoing efforts to address incomplete or missing data. The challenges posed by data quality issues are multifaceted and contribute to an overall reduction in the reliability of information. In the era of big data and artificial intelligence, dealing with incomplete information and missing data is a challenge. Inductive learning, a form of machine learning that involves making generalizations based on specific examples, can be a valuable approach for link prediction to overcome some obstacles associated with knowledge graphs in healthcare. In response to these challenges, link prediction is becoming as a valuable technique to improve the quality of knowledge graphs by filling in missing links. The state-of-the-art proposes various approaches for knowledge graph completion and link predictions involves the evaluation of different embeddings and symbolic learning models. Experimental benchmarks are designed to evaluate different models and relations types and provide insights into their effectiveness. This research aims to develop a framework for evaluation of hybrid AI models over lung cancer knowledge graph. The primary objectives include comparative analysis of embeddings and symbolic learning models, investigation of the impact of data modelling, exploration of the influence of relation types, and evaluation of the impact of knowledge graph enhancing.