ISEE: Interactive Semantic Enrichment for Database Fields
A novel system named ISEE (Interactive SEmantic Enrichment) has been launched to enhance the precision and thoroughness of data field descriptions, which frequently suffer from ambiguity or incompleteness. This system, outlined in an arXiv paper (2608.02604), employs LLM-based agents to aid in various data tasks, including exploration, retrieval, and sense-making. Nevertheless, the efficacy of these agents is compromised by vague semantics, as critical context typically stems from users' domain expertise and is seldom documented. ISEE tackles this issue by evaluating the quality of field descriptions with a scoring mechanism, accumulating domain knowledge, and collaboratively enriching semantics with users. The paper includes a user study, automated simulations, quantitative assessments, and a case study, showing that ISEE notably decreases cognitive load and enhances performance in downstream tasks such as entity-linking.
Key facts
- ISEE is a system for interactive semantic enrichment of database fields.
- It targets LLM-based agents used for data-related tasks.
- Field descriptions are often ambiguous or incomplete due to undocumented domain knowledge.
- ISEE scores field description quality and gathers domain knowledge.
- It collaborates with users to enrich semantics.
- The system was evaluated through user study, automated simulation, quantitative evaluation, and case study.
- Results show significant reduction in cognitive load and improved task performance.
- The paper is available on arXiv with ID 2608.02604.
Entities
Institutions
- arXiv