ProActLLM

About

From Reactive to Proactive: The Next Frontier in Conversational AI

While today's AI assistants excel at answering direct questions, they remain fundamentally reactive—waiting for users to ask before they act. What if conversational AI could anticipate your needs, offer timely suggestions, and guide you toward information you actually need or didn't even know you needed?

ProActLLM explores this transformative shift from reactive question-answering to proactive information-seeking assistants powered by Large Language Models (LLMs). Our workshop investigates how AI systems can:

Anticipate user needs before they're explicitly expressed
Model complex conversational context to understand deeper intent
Take helpful initiative while respecting user control
Seamlessly integrate external tools and knowledge sources
Personalize interactions based on individual preferences and expertise
Adapt and learn from user feedback over time
Optimise interaction interfaces and response representations for engaging user experience

Why This Matters Now

Recent advances in LLMs have opened unprecedented opportunities for building truly intelligent, context-aware dialogue systems. Yet current conversational AI largely remains stuck in a reactive paradigm. ProActLLM addresses this gap by bringing together leading researchers from Natural Language Processing, Information Retrieval, Human-Computer Interaction, and Cognitive Science to define and advance this emerging field.

Our Vision

We envision conversational assistants that don't just respond — they collaborate. Systems that understand not only what you're asking, but what you're trying to achieve. Assistants that can surface relevant information at the right moment, suggest productive next steps, and adapt their behavior to your expertise level and cognitive load. Join us in shaping the future of proactive conversational AI — where the conversation truly becomes a partnership between human intelligence and artificial intelligence.

Call for Papers

Shape the Future of Proactive Conversational AI

We invite original research papers (short and long), perspective papers, system demonstrations, and resource papers that advance our understanding of proactive conversational information seeking. Whether you're developing novel algorithms, discussing potential related future research directions, proposing evaluation frameworks, or building real-world applications, we want to hear from you.

Research Themes

🤖 Core Proactive Behaviors

Proactive question asking and clarification: How can agents formulate the right questions at the right time to better understand user intent?
Mixed-initiative dialogue management: What models best balance system initiative with user control in information-seeking conversations?
Anticipatory information retrieval: Techniques for predicting and surfacing relevant information before users explicitly request it.

🧠 Understanding Users and Context

User modeling and context understanding: Inferring user goals, knowledge, expertise levels, and preferences to drive intelligent proactive actions.
Personalized conversational search: Adapting search results, suggestions, and dialogue flow based on individual user characteristics, history, and preferences.
Cognitive load management: Designing systems that respect users' cognitive capacity and attention limitations while maximizing utility.
Metacognitive awareness: Building systems that recognize when users need different types of cognitive support or are experiencing confusion.

🔧 Technical Foundations

LLMs for conversational information access: Retrieval-augmented generation, knowledge grounding, and dynamic topic suggestion in dialogue.
Memory and forgetting models: Human-like memory mechanisms that prioritize, decay, and reconstruct information for better proactive suggestions.
Tool orchestration and API integration: Seamlessly coordinating multiple external tools and services for complex, multi-step information needs.

📊 Evaluation and Methodology

Evaluation metrics and methodologies: Novel benchmarks and user-centric evaluation protocols for measuring proactive system effectiveness.
Explainability and transparency: Making proactive decisions interpretable and helping users understand system reasoning.
Long-term interaction studies: Understanding how proactive behaviors affect user satisfaction and task success over time.

🌍 Applications and Impact

Domain-specific applications: Healthcare, education, customer support, enterprise search, personal assistants, and beyond.
Multi-modal proactive assistance: Integrating visual, audio, and textual cues across different interaction modalities.
Cultural and linguistic adaptation: Developing systems sensitive to diverse cultural contexts and communication styles.

⚖️ Responsible AI

Ethical and user experience considerations: Ensuring proactive behavior is helpful rather than intrusive.
Privacy and fairness: Protecting user data while ensuring equitable experiences across diverse user populations.
Bias mitigation: Addressing and preventing harmful biases in proactive suggestions and recommendations.

What We're Looking For

Novel Research: Cutting-edge algorithms, models, and techniques that advance the state-of-the-art in proactive conversational systems
Practical Systems: Working implementations, demos, and case studies showing real-world applications of proactive conversational AI
Theoretical Insights: Position papers, surveys, and conceptual frameworks that help define and structure this emerging field
Evaluation Studies: User studies, benchmarks, and evaluation frameworks that help us better understand and measure proactive behaviors
Interdisciplinary Perspectives: Work that bridges computer science with cognitive science, psychology, linguistics, or domain expertise

Submission Instructions

We welcome original research contributions that advance the field of proactive conversational information seeking. All submissions must be previously unpublished work, not currently under review at another venue.

Submission Types & Requirements

📄 Research Papers

Short Papers (5-7 pages): Preliminary results, novel ideas, or focused contributions.
Long Papers (10-12 pages): Comprehensive technical papers presenting novel algorithms, models, empirical studies, or theoretical insights.

📝 Perspective Papers (5-12 pages)

Opinion pieces, surveys, or conceptual frameworks that help define and structure the field.

🖥️ System Demonstrations (5-7 pages + demo)

Working systems with accompanying demonstration videos (max 5 minutes).

📊 Dataset & Resource Papers (5-12 pages)

Novel datasets, benchmarks, or evaluation resources for proactive conversational systems

All page limits exclude references and appendices

Format Requirements

Template: CEUR-WS single-column format
Language: English
File Format: PDF only
Submission Portal: https://easychair.org/conferences?conf=proactllm2025
Download Templates:
- Zip file containing ODT and LaTeX files
- Overleaf LaTeX CEUR Template (Recommended)

Template Usage Instructions:

LaTeX users: Use the 1-column CEUR-WS LaTeX template (preferred option). Requires TeX-Live 2020 or later for local installations.
LibreOffice users: Use the ODT template if you cannot use LaTeX. Ensure Libertinus font family is installed on your computer (instructions included in template).
Important: Do not use cloud-based editors (Office 365, Google Docs) as they don't support required fonts.
Microsoft Word: Not recommended due to PDF compatibility issues.
Note: All templates include the mandatory AI declaration section as of 2024.

Submission Guidelines

What to Include

Clear problem statement and motivation
Comprehensive related work positioning within existing literature
Detailed methodology with sufficient implementation details
Rigorous evaluation with appropriate baselines and statistical tests
Honest discussion of limitations and future work

Special Requirements by Type

Experimental Papers:
- Include statistical significance testing
- Provide implementation details for reproducibility
- Share code/data when possible (anonymized links acceptable)
System Papers:
- Include architecture diagrams and technical specifications
- Provide demo video for system demonstrations
- Discuss scalability and real-world deployment considerations
Position Papers:
- Clearly articulate the position or perspective
- Support arguments with evidence from literature
- Discuss implications for future research directions

Camera-Ready Instructions

For accepted papers:

Address all reviewer comments and concerns
Follow CEUR-WS single-column format exactly
Ensure compliance with CEUR-WS publication requirements
Submit source files (LaTeX preferred) along with PDF
Complete CEUR-WS copyright agreement

Publication: Accepted papers will be published in CEUR Workshop Proceedings (CEUR-WS.org), which is indexed by DBLP and Scopus.

Important Dates

Paper Submission Deadline:	August 31, 2025 (11:59 PM AoE)
Notification of Acceptance:	September 30, 2025
Camera-Ready Deadline:	October 10, 2025 (11:59 PM AoE)
Workshop Date:	November 14, 2025