Beyond a single dataset: connecting knowledge with ‘bonds’
How do you logically link ‘users’ to ‘movie genres’? We explore the theory of ‘bonds’, the mathematical ‘glue’ that connects two different datasets, providing new properties and experiments to make them more powerful.
In data analysis, we’re very good at understanding one dataset at a time. We can take a list of “users and the movies they like” and build a perfect map of their preferences (a concept lattice). We can do the same for “movies and their genres.”
But how do you connect the two? How do you build a logical bridge from the “user” map to the “genre” map? In our 2023 paper in Information Sciences (Q1), we explored the deep theory behind this “conceptual glue.”
🧐 The problem: connecting the dots
In Formal Concept Analysis (FCA), this “glue” is called a bond. A bond is a formal, mathematical structure that links the concepts from one context (like users) to the concepts of another (like genres).
This idea is critical for applications like recommender systems. However, to use bonds effectively, we need to know exactly how they work. Previous work had introduced two main types:
- Rigorous bonds: A “pessimistic” or strict link (e.g., “link user A to genre X only if all their movies are of that genre”).
- Benevolent bonds: An “optimistic” or loose link (e.g., “link user A to genre X if at least one of their movies is of that genre”).
But the full theory of how to build and use these bonds, especially with “fuzzy” or graded data, needed to be solidified.
💡 Our solution: using external info to build the bridge
Our paper provides a deep theoretical and experimental analysis of how to construct these bonds using a third, separate piece of information: an external context.
Let’s use the Netflix example: * Context 1: (Users, Movies) * Context 2: (Movies, Genres) * The “external information” that links them is the movies themselves.
Our work formalized how to use this external information to build both rigorous and benevolent bonds. We didn’t just propose the idea; we proved a host of new theoretical properties about how these bonds behave and what they represent.
*
🚀 The results: a solid foundation for bonds
This is a fundamental theory paper. Our results are the new mathematical properties and formal interpretations that make bonds a trustworthy tool for data scientists.
We provided: * Extended interpretations for both rigorous and benevolent operators. * New theoretical properties that show how these bonds formally connect the two concept lattices. * A set of experiments to validate these properties and show how they work on real data.
🔬 Why does this matter?
This work isn’t just an academic exercise. It provides the solid mathematical foundation needed to build more powerful, flexible, and reliable recommender systems and other complex AI tools.
By understanding exactly how these bonds work, data scientists can now choose the right kind of bond (strict or loose) for their specific task. It allows us to build smarter systems that can logically connect different knowledge bases, which is a critical step for linking disparate datasets and building more intelligent AI.
📖 The full paper
For the complete theoretical definitions, mathematical proofs, and experimental results, you can read the original journal article.
Connecting concept lattices with bonds induced by external information. Authors: Ondrej Krídlo, Domingo López-Rodríguez, Lubomir Antoni, Peter Eliaš, Stanislav Krajči, Manuel Ojeda-Aciego. Journal: Information Sciences (vol. 648, 119498)