Date of Award

2024

Document Type

Thesis

Degree Name

MS in Science

Department

Division of Computer Science, Mathematics and Science

First Advisor

Syed Ahmad Chan Bukhari

Abstract

The rise in online biomedical information has led to challenges in data retrieval, primarily due to insufficient semantic metadata. While semantic annotators have advanced in addressing this problem, they still lack accuracy, speed, and, most importantly, dynamic knowledge representation—an essential aspect of AI that enables human-like reasoning. Several techniques to achieve knowledge representation, such as Knowledge Graphs, have been introduced to improve current systems in many domains, including the biomedical domain. Knowledge Graphs, recognized for their semantic richness, show promise in enhancing the biomedical semantic annotation process. We propose a knowledge graph-based recommendation system, augmented with NLP for search queries, designed for the Semantically biomedical content annotation platform. This system aims to provide quick and easy access to optimal, machine-readable recommendations and graphically visualize recommendations in a knowledge graph format. Still, to build the knowledge graph and the interface, we had to address challenges and issues with the current system. Semantically uses a relational schema for data storage, which does not support machine-readable content that would allow search engines to understand and suggest annotations easily. Inherent differences in the sequential query language and cipher language mean that the backend for data manipulation must be rewritten entirely to support Neo4j operations. A knowledge graph interface with graph visualizations was also needed to present optimal recommendations to users. Our approach to these challenges is to migrate the MySQL database to a Neo4j database to create a knowledge graph database. Then, we rewrote the backend to connect to Neo4j and support cipher language queries. Using data visualization JavaScript libraries like D3.js, we wrote a knowledge graph interface that returns a graph visualization of annotation recommendations obtained from the knowledge graph. We verify the effectiveness of the knowledge graph through an evaluation survey taken by 20 people, testing its ability to represent knowledge dynamically. The results show a mostly positive reception for all positive and negative system characteristics, with crucial metrics such as speed, precision, and clarity with agreement percentages of 100%, 95%, and 95%, respectively. A demo of the knowledge graph-based recommendation system is available at https://github.com/bukharilab/KG4BioMedCntnAuthoring.

Download

Share

COinS