Daniel Daza's research focused on improving how computers understand and use large, interconnected datasets called knowledge graphs (KGs). These graphs organize information about entities, like people, organizations, or molecules, and the relationships between them. While they are excellent for storing and retrieving data, they often miss subtle patterns and connections.
His research developed new ways for computers to analyze and make sense of these graphs, enabling them to recognize patterns, answer complex questions, and predict missing connections. A key motivation was to make better use of the rich context in KGs, such as patterns involving groups of entities or additional information like textual descriptions. This has practical applications, such as answering natural language questions, classifying entities, and exploring biomedical data like protein-molecule interactions. His work helps expand the potential of KGs in solving real-world problems.
The research has shown that we can teach computers to better understand and use large networks of information, called knowledge graphs, by looking at more than just simple connections between pieces of data. For example, instead of just knowing that one person is related to another, we can explore groups of relationships and additional details, like descriptions or characteristics, to uncover hidden patterns.
This means computers can now answer more complex questions, like finding connections that aren’t directly stated but are likely based on context. They can also handle new types of information, such as classifying data or analyzing biomedical details like protein interactions, even when there isn’t much information available. Overall, the work demonstrates that we can make smarter use of the information we already have, helping computers solve more advanced and practical problems.
Daza conducted the research by designing and testing computer models to analyze and learn from large datasets called knowledge graphs. These models simulate how computers can discover patterns in the data, such as relationships between people, places, or molecules. He developed new methods that look at groups of connections (subgraphs) and additional information, like descriptions or molecular structures, to improve how computers understand and predict complex patterns.
To test these methods, Daza used real-world datasets from various fields, such as social networks and biomedical data. He evaluated how well the models could answer questions, classify information, and make accurate predictions. Through these experiments, he refined the methods to ensure they worked efficiently and could generalize to new, unseen data. The work relied heavily on computer simulations and data analysis to create practical tools that can be applied to real-world problems.
More information on the thesis