AutoPR: Let's Automate Your Academic Promotion! (2510.09558v1)

Abstract: As the volume of peer-reviewed research surges, scholars increasingly rely on social platforms for discovery, while authors invest considerable effort in promoting their work to ensure visibility and citations. To streamline this process and reduce the reliance on human effort, we introduce Automatic Promotion (AutoPR), a novel task that transforms research papers into accurate, engaging, and timely public content. To enable rigorous evaluation, we release PRBench, a multimodal benchmark that links 512 peer-reviewed articles to high-quality promotional posts, assessing systems along three axes: Fidelity (accuracy and tone), Engagement (audience targeting and appeal), and Alignment (timing and channel optimization). We also introduce PRAgent, a multi-agent framework that automates AutoPR in three stages: content extraction with multimodal preparation, collaborative synthesis for polished outputs, and platform-specific adaptation to optimize norms, tone, and tagging for maximum reach. When compared to direct LLM pipelines on PRBench, PRAgent demonstrates substantial improvements, including a 604% increase in total watch time, a 438% rise in likes, and at least a 2.9x boost in overall engagement. Ablation studies show that platform modeling and targeted promotion contribute the most to these gains. Our results position AutoPR as a tractable, measurable research problem and provide a roadmap for scalable, impactful automated scholarly communication.

• The paper introduces the AutoPR task by formalizing automated, platform-specific academic promotion using a multimodal benchmark and multi-agent system.
• The paper presents the PRAgent framework which extracts, synthesizes, and adapts content through specialized agents for improved fidelity, engagement, and alignment.
• The paper demonstrates significant performance gains in engagement metrics, validating the multi-agent approach over direct prompting and traditional summarization methods.

Automating Academic Promotion: The AutoPR Task, PRBench Benchmark, and PRAgent Framework

Introduction and Motivation

The exponential growth in peer-reviewed research output has rendered manual tracking and dissemination of scientific advances increasingly impractical. Social media and digital platforms have become primary channels for research discovery, yet the creation of high-quality, engaging, and platform-optimized promotional content remains labor-intensive and highly variable in quality. The paper introduces the Automatic Promotion (AutoPR) task, which formalizes the end-to-end transformation of research papers into effective, audience- and platform-tailored promotional materials. To enable systematic evaluation and progress in this domain, the authors present PRBench, a multimodal benchmark linking 512 peer-reviewed articles to curated promotional posts, and propose PRAgent, a multi-agent system for automated, high-fidelity academic promotion.

Figure 1: Overview of the AutoPR task, PRBench benchmark, and the PRAgent multi-agent framework.

Formalization of the AutoPR Task

The AutoPR task is defined as a multi-objective optimization problem: given a research document $\mathbb{D}$ (comprising text, visuals, and supplementary materials), a target platform $\mathbb{T}_P$, and an intended audience $\mathbb{T}_A$, the goal is to generate a promotional post $\hat{P}$ that maximizes a weighted sum of Fidelity (accuracy and completeness), Alignment (platform-specific style and conventions), and Engagement (audience appeal and interaction potential):

$$\hat{P} = \operatorname*{argmax}\limits_{P}\left\{\alpha_1 \mathcal{S}_\text{Fidelity}(P) + \alpha_2 \mathcal{S}_\text{Align}(P) + \alpha_3 \mathcal{S}_\text{Engage}(P)\right\}$$

This formalization captures the inherent trade-offs between factual accuracy, platform optimization, and audience engagement, and provides a rigorous basis for both system design and evaluation.

Figure 2: Definition and overview of the Automatic Promotion (AutoPR) task.

PRBench: A Multimodal Benchmark for Academic Promotion

PRBench is constructed to enable robust, fine-grained evaluation of AutoPR systems. The dataset comprises 512 paper–promotion pairs, each linking an arXiv paper (with full text and visuals) to a human-authored promotional post from Twitter (X) or RedNote. The benchmark is annotated along three axes:

• Fidelity: Authorship/title accuracy and a weighted factual checklist, with expert-assigned importance weights.
• Engagement: Multi-dimensional scoring of hook strength, logical attractiveness, visual appeal, call-to-action, and pairwise preference for professional and broader audiences.
• Alignment: Contextual relevance, visual–text integration, and hashtag/mention strategy, with platform-specific rubrics.

The evaluation protocol leverages both expert human annotation and LLM-based judging, with Qwen-2.5-VL-72B-Ins selected for its high correlation with human judgments.

PRAgent: Multi-Agent Framework for Automated Promotion

PRAgent is a modular, three-stage multi-agent system designed to address the limitations of direct LLM prompting for AutoPR:

1. Content Extraction and Structuring: Hierarchical summarization and multimodal parsing convert unstructured PDFs into structured, machine-readable representations. Visual elements are extracted using DocLayout-YOLO and paired with captions via nearest-neighbor matching.
2. Multi-Agent Content Synthesis: Four specialized agents operate in parallel:
   • Logical Draft Agent: Produces a factually dense, style-agnostic draft with explicit analytical modules (research question, contributions, method, results).
   • Visual Analysis Agent: Uses a multimodal LLM to generate expert-level figure analyses.
   • Textual Enriching Agent: Adapts the logical draft to platform-specific style, incorporating hooks, CTAs, and hashtags.
   • Visual-Text-Interleaved Combination Agent: Integrates text and visuals into a cohesive, platform-optimized narrative.
3. Platform-Specific Adaptation and Orchestration: Final adaptation aligns tone, formatting, and engagement strategies with platform norms, producing publication-ready content.

   Figure 3: PRAgent architecture: content extraction, multi-agent synthesis, and platform-specific adaptation.

Empirical Evaluation and Analysis

Baseline Performance and Bottlenecks

Comprehensive benchmarking on PRBench reveals that even state-of-the-art LLMs (e.g., GPT-5, Qwen-2.5-VL-32B-Ins) exhibit significant deficiencies:

• Fidelity: Top models miss over 40% of key facts; errors are concentrated in omission or misstatement of core methods, results, and terminology.
• Engagement: AI-generated posts often lack authentic narrative structure and effective hooks; 42% of outputs omit any engagement device.
• Alignment: Hashtag and mention strategies are superficial, with low overlap with human-authored posts (Jaccard similarity 0.03).

  Figure 4: Analysis of AI-generated academic promotion, highlighting fidelity, engagement, and alignment limitations.

PRAgent vs. Direct Prompting

PRAgent consistently outperforms direct LLM prompting across all metrics and model backbones. On PRBench-Core, PRAgent yields at least a 7.15% improvement in average score, with gains exceeding 20% on smaller models. Ablation studies confirm that each stage—content extraction, multi-agent synthesis, and platform adaptation—contributes distinctly to final performance, with platform adaptation being most critical for alignment and engagement.

Real-World Deployment

A 10-day live experiment on RedNote demonstrates that PRAgent-generated posts achieve a 604% increase in total watch time, 438% more likes, and at least a 2.9x boost in overall engagement compared to direct-prompted baselines, controlling for account and posting variables.

Figure 5: PRAgent outperforms direct-prompt baseline in a 10-day real-world RedNote paper, with substantial gains in engagement metrics.

Human Preference and Component Analysis

Human preference studies indicate that PRAgent-generated posts are favored over human-authored posts in 64.8% of cases, with only 11.7% preferring the human version. Component ablations show that removing platform adaptation or content extraction leads to substantial drops in fidelity and alignment, underscoring the necessity of a structured, multi-agent approach.

Methodological Insights and Scaling Laws

• Parameter scaling: Larger models generally yield better AutoPR performance, but improvements are not strictly monotonic across model families.
• Long CoT and ICL: Chain-of-thought reasoning and in-context learning do not consistently improve AutoPR outcomes, and may even degrade performance due to specification drift.
• LLM evaluation: Open-source LLMs (e.g., Qwen-2.5-VL-72B-Ins) align more closely with human judgments than closed-source models, especially for engagement and alignment metrics.

Implications and Future Directions

The formalization of AutoPR and the release of PRBench establish a rigorous foundation for research on automated scholarly communication. PRAgent demonstrates that a modular, multi-agent approach is essential for high-fidelity, engaging, and platform-optimized academic promotion. The empirical results highlight that generic summarization and direct prompting are insufficient for this complex, multi-objective task.

Practical implications include scalable, automated pipelines for research dissemination, integration with institutional repositories, and potential for real-time, adaptive promotion strategies. Theoretically, the work motivates further research on multi-agent coordination, multimodal content synthesis, and platform-aware generation. Open challenges remain in robust factual grounding, authentic engagement modeling, and cross-platform generalization.

Conclusion

This work introduces the AutoPR task, PRBench benchmark, and PRAgent framework, providing a comprehensive solution for automated academic promotion. The results demonstrate that structured, multi-agent systems can substantially surpass both direct LLM prompting and human baselines in fidelity, engagement, and alignment. The release of PRBench and PRAgent will catalyze further research on automated, scalable, and effective scholarly communication, with broad implications for the future of scientific dissemination and altmetrics-driven impact.