TL;DR
This paper proposes a method for merging traditional databases with knowledge bases to enable more sophisticated queries. By creating "hybrid knowledge bases/databases," users can leverage general domain knowledge to query data in ways previously impossible, simplifying complex queries and opening new avenues for data mining.
What Problem Does This Paper Address?
Traditional databases, while excellent at storing structured data, often lack general domain knowledge. For instance, a medical database might store patient records but wouldn't inherently understand the relationships between diseases, their causes (like bacteria), or symptoms. This limitation restricts the types of queries users can perform, forcing them to rely on pre-existing, often complex, knowledge to formulate queries. The paper addresses this by proposing a way to integrate structured data with rich, general domain knowledge, thereby enhancing query capabilities and making them more accessible.
Methods
The core of the proposed solution involves merging a knowledge base (KB) with a database (DB) to create a hybrid KB/DB. The authors utilize a frame-based semantic network called Parka, developed at the University of Maryland, as the knowledge representation tool to facilitate this merger. The research focuses on developing methodologies for this integration, including techniques to automatically convert relational DB schemas into semantic networks and to extract and integrate existing knowledge bases, such as medical ontologies like UMLS.
Datasets
The paper describes an example application using a medical database of approximately 20,000 OB/GYN patients. This database is a single table with over 70 columns. A hybrid OB/GYN KB/DB was developed from this data, containing 60,000 frames and 1.3 million assertions.
Key Insights and Ideas
- Bridging the Knowledge Gap: The fundamental insight is that combining structured data with general domain knowledge unlocks powerful new query capabilities that are impossible with either component alone.
- Simplifying Complex Queries: By integrating domain knowledge, queries that would previously require extensive user expertise and complex disjunctions can be expressed much more simply. For example, querying for patients with infections caused by bacteria becomes straightforward when the KB understands the concept of bacteria and its relation to infections.
- Enabling Broader User Access: This simplification democratizes access to sophisticated data analysis, allowing users without deep domain expertise to ask more insightful questions.
- Foundation for Data Mining: Hybrid KB/DBs are seen as a promising foundation for attribute-oriented data mining techniques, suggesting future research directions.
Key Results
The paper demonstrates how a complex query, such as listing patients with infections caused by bacteria, can be simplified. Instead of a lengthy disjunction of specific bacterial infections, the hybrid KB/DB allows for a query that directly references the concept of "bacteria" as a cause of infection. Figure 1 visually illustrates this simplified query using the Parka system. The developed hybrid OB/GYN KB/DB is substantial, comprising 60,000 frames and 1.3 million assertions.
Limitations and Future Work
The paper acknowledges that this research is ongoing. Future work includes developing general methodologies for merging DBs and KBs to make this technique more widely applicable. Specific areas of investigation include automatically converting relational DB schemas into semantic networks and extracting/integrating parts of existing KBs, such as medical ontologies.
Summary updated