Multi-round scientific research workflow evaluation system and method for multi-modal intelligent agent

By constructing a multi-round scientific research workflow evaluation system, the shortcomings of existing technologies in evaluating multimodal intelligent agents in multi-round scientific research workflows are addressed. This system enables a realistic and fine-grained assessment of intelligent agents in scientific research assistance scenarios, thereby improving the effectiveness and guidance of the evaluation.

CN122347401APending Publication Date: 2026-07-07HARBIN INST OF TECH
1 Cites 0 Cited by

Patent Information

Application Number
CN202610492139.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-15
Publication Date
2026-07-07

AI Technical Summary

Technical Problem

Existing technologies lack effective evaluation methods for proactive literature discovery, cross-paper multi-chart evidence integration, and evidence memory reuse when evaluating multi-modal intelligent agents' multi-round scientific research workflows. This makes it difficult to truly reflect the characteristics of scientific research tasks, and the evaluation results are not sufficiently instructive or diagnostic.

Method used

A multi-round scientific research workflow evaluation system for multimodal intelligent agents is constructed, including a multi-round task generation module, a multimodal information processing module, and a process-level evaluation mechanism. Through active search, evidence extraction and alignment, cross-paper evidence integration, and evidence round reuse, the system combines process-level evaluation of the intelligent agent's task processing process.

Benefits of technology

It enables the realism evaluation of multimodal intelligent agents in scientific research assistance scenarios, and can analyze their capability boundaries in a fine-grained manner, thereby improving the realism, interpretability and diagnostic value of the evaluation, and providing a reliable basis for the optimization of intelligent agents.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to a multi-round scientific research workflow evaluation system and method for a multi-modal intelligent agent, and belongs to the technical field of large language models and multi-modal large models. The multi-round task generation module is used for constructing a multi-round task scene closer to a real scientific research process; and the multi-modal information processing module is used for enabling the tested agent to complete active paper retrieval, multi-modal evidence extraction and alignment, cross-paper evidence integration and round reuse of the evidence. The protection points of the method are not limited to a single task sample or a single scoring index, but include a multi-round evaluation benchmark constructed around a real scientific research workflow, a process-level evaluation protocol matched with the multi-round evaluation benchmark, and an overall implementation scheme for supporting landing execution and fine-grained diagnosis of the benchmark.
Need to check novelty before this filing date? Find Prior Art