An Interactive Paradigm for Deep Research

2026-05-22

AI Summary

This work addresses the limitations of existing deep research systems, which rely on rigid pipelines and struggle to adapt dynamically to user intent over extended tasks. To overcome this, we propose SteER, a novel framework that introduces the first interactive and interpretable control paradigm for deep research. SteER employs cost–benefit analysis to decide when to pause and solicit user feedback, integrates diversity-aware planning with utility signals to dynamically optimize research trajectories, and incorporates a real-time evolving user persona model to jointly enhance alignment, novelty, and coverage. Experimental results demonstrate that SteER outperforms current open- and closed-source baselines by up to 22.80% in alignment and achieves significant gains in breadth and balance, earning preference from over 85% of human evaluators.

Abstract

Recent advances in large language models (LLMs) have enabled deep research systems that synthesize comprehensive, report-style answers to open-ended queries by combining retrieval, reasoning, and generation. Yet most frameworks rely on rigid workflows with one-shot scoping and long autonomous runs, offering little room for course correction if user intent shifts mid-process. We present SteER, a framework for Steerable deEp Research that introduces interpretable, mid-process control into long-horizon research workflows. At each decision point, SteER uses a cost-benefit formulation to determine whether to pause for user input or to proceed autonomously. It combines diversity-aware planning with utility signals that reward alignment, novelty, and coverage, and maintains a live persona model that evolves throughout the session. SteER outperforms state-of-the-art open-source and proprietary baselines by up to 22.80\% on alignment, leads on quality metrics such as breadth and balance, and is preferred by human readers in 85\%+ of pairwise alignment judgments. We also introduce a persona-query benchmark and data-generation pipeline. To our knowledge, this is the first work to advance deep research with an interactive, interpretable control paradigm, paving the way for controllable, user-aligned agents in long-form tasks.

Problem

• deep research
• interactive control
• user alignment
• long-horizon workflows
• steerable AI

Innovation

• interactive deep research
• steerable LLMs
• mid-process control
• persona-aware generation
• cost-benefit decision making