Infogent: An Agent-Based Framework for Web Information Aggregation

Revanth Gangi Reddy , Sagnik Mukherjee , Jeonghwan Kim , Zhenhailong Wang , Dilek Hakkani-Tur , Heng Ji

🏛 Institutions: UIUC
📅 Date: April 29, 2025
📑 Publisher: Findings of NAACL 2025
💻 Env: Web
🔑 Keywords: framework information aggregation interactive visual access direct API-driven access AssistantBench Infogent

TLDR

Infogent studies web information aggregation rather than single-site completion, asking agents to visit multiple websites and decide when enough evidence has been collected for a complex query. Its Navigator-Extractor-Aggregator design is evaluated in both direct API-driven and browser-based visual settings, including gains on AssistantBench under interactive visual access.

Open paper Report issue