Researchers propose ISEE system to improve LLM-based agents’ performance on data tasks

LLM-based agents increasingly perform data-related tasks such as sense-making, exploration, and retrieval, but their performance depends heavily on the clarity and completeness of data semantics. In practice, many field descriptions remain ambiguous or incomplete because essential context—often derived from users’ domain knowledge—is rarely documented publicly.

To bridge this gap, researchers introduce the Interactive SEmantic Enrichment system (ISEE), which measures the quality of a given field description using a scoring system, gathers relevant domain knowledge, and collaboratively enriches the semantics with users.

The authors evaluate ISEE through a user study, automated user simulation, quantitative evaluation, and case studies. They report that ISEE significantly reduces cognitive load, improves description quality, and enhances downstream task performance, particularly in tasks like entity-linking where precise semantics are critical.

The system is positioned as a response to the practical challenge of undocumented domain knowledge in database fields, which currently limits the effectiveness of automated data processing workflows.