Agentic AI-Empowered Dynamic Survey Framework
Organizations: University of South Florida · Mitsubishi Electric Research Laboratories (MERL)
Abstract
Survey papers play a central role in synthesizing and organizing scientific knowledge, yet they are increasingly strained by the rapid growth of research output. As new work continues to appear after publication, surveys quickly become outdated, contributing to redundancy and fragmentation in the literature. We reframe survey writing as a long-horizon maintenance problem rather than a one-time generation task, treating surveys as living documents that evolve alongside the research they describe. We propose an agentic Dynamic Survey Framework that supports the continuous updating of existing survey papers by incrementally integrating new work while preserving survey structure and minimizing unnecessary disruption. Using a retrospective experimental setup, we demonstrate that the proposed framework effectively identifies and incorporates emerging research while preserving the coherence and structure of existing surveys, even over extended sequences of updates.
Figures & tables
| Similarity | Property-Based Quality | Disruption | |||||
|---|---|---|---|---|---|---|---|
| Survey | BLEU | ROUGE | BERT | Semantic Align. | Local Coherence | Tokens | Out |
| Survey 1 | 7.74 | 0.216 | 0.867 | 0.809 | 0.787 | 233.3 | 0.0 |
| Survey 2 | 4.83 | 0.191 | 0.855 | 0.823 | 0.792 | 225.3 | 0.0 |
| Survey 3 | 2.22 | 0.149 | 0.849 | 0.804 | 0.777 | 224.2 | 0.0 |
| Survey 4 | 4.98 | 0.189 | 0.862 | 0.803 | 0.775 | 216.1 | 0.0 |
| Survey 5 | 1.61 | 0.158 | 0.847 | 0.791 | 0.787 | 229.9 | 0.0 |
| Similarity | Property-Based Quality | Disruption | |||||
|---|---|---|---|---|---|---|---|
| Method | BLEU | ROUGE | BERT | Semantic Align. | Local Coherence | Tokens | Out |
| Baseline 1 (Single-Step) | 1.12 | 0.100 | 0.791 | 0.730 | 0.717 | 6110.4 | 563.3 |
| Baseline 2 (Oracle Routing) | 2.84 | 0.148 | 0.842 | 0.797 | 0.760 | 410.5 | 225.1 |
| Ours | 4.28 | 0.181 | 0.856 | 0.806 | 0.783 | 225.8 | 0.0 |
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
| Survey topic | Year | Venue | #Sections / #Tables / #Late papers |
|---|---|---|---|
| Generic object detection | 2018 | IJCV | 10 / 2 / 9 |
| Adversarial attacks in computer vision | 2018 | IEEE Access | 7 / 1 / 15 |
| Remote sensing image super resolution | 2026 | ISPRS JPRS | 9 / 2 / 20 |
| Digital twin robotic arm manipulation with RL | 2026 | RCIM | 13 / 2 / 17 |
| Single scene video anomaly detection | 2020 | IEEE TPAMI | 6 / 3 / 16 |
| Backbone | Survey | BLEU | ROUGE-L | BERT | Sem. Align. | Local Coh. | Tokens | Out |
|---|---|---|---|---|---|---|---|---|
| Qwen3-4B | Survey 1 | 6.186 | 0.216 | 0.864 | 0.930 | 0.796 | 220.2 | 0 |
| Survey 2 | 4.618 | 0.211 | 0.857 | 0.944 | 0.758 | 210.9 | 0 | |
| Survey 3 | 1.929 | 0.151 | 0.849 | 0.925 | 0.799 | 148.7 | 0 | |
| Survey 4 | 2.510 | 0.176 | 0.855 | 0.915 | 0.729 | 180.2 | 0 | |
| Survey 5 | 2.862 | 0.165 | 0.845 | 0.899 | 0.772 | 199.5 | 0 | |
| Average | 3.621 | 0.184 | 0.854 | 0.923 | 0.771 | 191.9 | 0 |
| Method | Survey | BLEU | ROUGE-L | BERT | Sem. Align. | Local Coh. | Tokens | Out |
|---|---|---|---|---|---|---|---|---|
| Baseline 1 | Survey 1 | 0.538 | 0.095 | 0.809 | 0.891 | 0.751 | 3998.3 | 739.3 |
| Survey 2 | 1.497 | 0.138 | 0.815 | 0.891 | 0.745 | 525.2 | 394.1 | |
| Survey 3 | 0.718 | 0.058 | 0.815 | 0.927 | 0.782 | 7519.6 | 725.4 | |
| Survey 4 | 0.465 | 0.070 | 0.811 | 0.899 | 0.718 | 584.1 | 281.9 | |
| Survey 5 | 0.309 | 0.070 | 0.796 | 0.857 | 0.700 | 602.5 | 482.0 | |
| Average | 0.706 | 0.086 | 0.809 | 0.893 | 0.739 | 2645.9 | 524.5 |
| Method | Survey | BLEU | ROUGE-L | BERT | Sem. Align. | Local Coh. | Tokens | Out |
|---|---|---|---|---|---|---|---|---|
| Baseline 1 | Survey 1 | 0.256 | 0.103 | 0.816 | 0.863 | 0.753 | 7544.0 | 584.5 |
| Survey 2 | 1.431 | 0.135 | 0.815 | 0.893 | 0.748 | 919.3 | 394.1 | |
| Survey 3 | 0.610 | 0.059 | 0.812 | 0.926 | 0.772 | 8642.1 | 726.5 | |
| Survey 4 | 0.619 | 0.069 | 0.813 | 0.894 | 0.720 | 578.4 | 245.2 | |
| Survey 5 | 0.251 | 0.113 | 0.797 | 0.850 | 0.695 | 884.9 | 619.4 | |
| Average | 0.633 | 0.096 | 0.811 | 0.885 | 0.738 | 3713.7 | 513.9 |
| Method | Survey | BLEU | ROUGE-L | BERT | Sem. Align. | Local Coh. | Tokens | Out |
|---|---|---|---|---|---|---|---|---|
| Baseline 1 | Survey 1 | 0.599 | 0.099 | 0.814 | 0.893 | 0.756 | 3940.5 | 990.0 |
| Survey 2 | 1.241 | 0.134 | 0.819 | 0.876 | 0.729 | 5555.9 | 422.6 | |
| Survey 3 | 0.367 | 0.051 | 0.813 | 0.928 | 0.768 | 4401.1 | 1215.2 | |
| Survey 4 | 0.386 | 0.063 | 0.807 | 0.815 | 0.682 | 4282.7 | 459.8 | |
| Survey 5 | 2.342 | 0.096 | 0.810 | 0.838 | 0.759 | 9828.3 | 830.3 | |
| Average | 0.987 | 0.089 | 0.813 | 0.870 | 0.739 | 5601.7 | 783.6 |
| Survey | # Papers | Top–1 | Top–3 |
|---|---|---|---|
| Object Detection | 9 | 0.89 | 0.94 |
| Adversarial Attacks | 15 | 0.87 | 0.93 |
| Remote Sensing SR | 20 | 0.90 | 0.95 |
| Robotic Arms | 17 | 0.90 | 0.95 |
| Video Anomaly | 16 | 0.92 | 0.97 |
| Survey | Top–1 | Top–3 |
|---|---|---|
| Object Detection | 0.72 | 0.87 |
| Adversarial Attacks | 0.80 | 0.92 |
| Remote Sensing SR | 0.93 | 0.97 |
| Robotic Arms | 0.95 | 0.98 |
| Video Anomaly | 0.93 | 0.98 |
| Survey | Total | TP | TN | FP | FN | Precision | Recall |
|---|---|---|---|---|---|---|---|
| Object Detection | 45 | 13 | 28 | 3 | 1 | 81.25% | 92.86% |
| Adversarial Attacks | 45 | 15 | 29 | 1 | 0 | 93.75% | 100% |
| Remote Sensing SR | 45 | 23 | 20 | 1 | 1 | 95.83% | 95.83% |
| Robotic Arms | 78 | 17 | 61 | 0 | 0 | 100% | 100% |
| Video Anomaly | 78 | 16 | 62 | 0 | 0 | 100% | 100% |
| Dataset | Samples ( ) | Total Fidelity ( ) | Exact Match |
|---|---|---|---|
| Survey 1 | 11 | 0.91 | 0.55 |
| Survey 2 | 21 | 0.86 | 0.62 |
| Survey 3 | 132 | 0.90 | 0.68 |
| Survey 4 | 2 | 1.00 | 0.50 |
| Survey 5 | 14 | 0.88 | 0.57 |
| Weighted Avg. | 180 | 0.90 | 0.65 |
| Similarity | Property-Based Quality | Disruption | ||||||
|---|---|---|---|---|---|---|---|---|
| Backbone | Survey | BLEU | ROUGE | BERT | Semantic Align. | Local Coherence | Tokens | Out |
| Qwen3-4B | Survey 1 | 5.6395 | 0.2002 | 0.8626 | 0.7705 | 0.7458 | 260.667 | 0.0 |
| Survey 2 | 3.4556 | 0.1876 | 0.8510 | 0.7819 | 0.7430 | 285.133 | 0.0 | |
| Survey 3 | 2.0293 | 0.1326 | 0.8435 | 0.7961 | 0.7587 | 253.714 | 0.0 | |
| Survey 4 | 3.2538 | 0.1669 | 0.8537 | 0.7652 | 0.7332 | 270.765 | 0.0 | |
| Survey 5 | 2.0971 | 0.1609 | 0.8471 | 0.7788 | 0.7685 | 251.611 | 0.0 | |
| Similarity | Property-Based Quality | Disruption | ||||||
|---|---|---|---|---|---|---|---|---|
| Backbone | Method | BLEU | ROUGE | BERT | Semantic Align. | Local Coherence | Tokens | Out |
| Qwen3-4B | Baseline 1 | 0.7712 | 0.0583 | 0.8143 | 0.7448 | 0.7476 | 4082.326 | 684.192 |
| Baseline 2 | 1.5012 | 0.1006 | 0.8319 | 0.7770 | 0.7475 | 807.707 | 319.273 | |
| Gemma-4B | Baseline 1 | 1.0136 | 0.0753 | 0.8143 | 0.7192 | 0.7310 | 4213.640 | 678.858 |
| Baseline 2 | 1.3050 | 0.0957 | 0.8328 | 0.7627 | 0.7547 | 782.200 | 335.625 | |
| Qwen3-14B | Baseline 1 | 0.6809 | 0.0525 | 0.8133 | 0.7410 | 0.7429 | 5707.472 | 571.503 |