efit of the injected signal.
Semantic alignment (RQ3): are RVLSMs’s summaries closely aligned with human-generated ones?
We empirically verify whether the RVLSM’s summaries are more closely aligned with human expectations
than baseline methods in terms of NLP entities retrieved in the ground-truth summary.
The core idea is to verify the semantic alignment in terms of matching with semantically rich concepts.
Table 2 reports the Precision, Recall, and F1 MicroAverages separately for each entity type.
On MixTyVSum, RVLSM attains the best aggregate scores (F1=0.170 to 0.081 for the VideoLLM baseline) and leads on most entity types, including work_of_art, date, and person, where the baselines recover no entity. Overall, RVLSM recovers salient entities that surface-overlap metrics tend to overlook. We also evaluate extracted named entities on V2Xum-generated summaries. However, only a few cardinal are detected, leading to all zeros matric values for all other entities. More results are reported in the Supplementary Document.
Entity type
RVLSM (ours)
Baseline VideoLLM
P ↑\uparrow
R ↑\uparrow
F1 ↑\uparrow
P ↑\uparrow
R ↑\uparrow
F1 ↑\uparrow
CARDINAL
0.393
0.204
0.268
0.125
0.024
0.041
GPE
0.188
0.300
0.231
0.333
0.250
0.286
WORK_OF_ART
0.308
0.190
0.235
0.000
0.000
0.000
DATE
0.222
0.182
0.200
0.143
0.100
0.118
PERSON
0.128
0.312
0.182
0.000
0.000
0.000
EVENT
0.000
0.000
0.000
0.000
0.000
0.000
ORG
0.083
0.235
0.123
0.000
0.000
0.000
NORP
0.000
0.000
0.000
0.200
0.500
0.286
ORDINAL
0.167
0.027
0.047
1.000
0.031
0.061
FAC
0.000
0.000
0.000
0.500
1.000
0.667
Mean
0.170
0.170
0.170
0.136
0.058
0.081
Table 2: Precision, Recall, and F1 of the entities detected on MixTyVSum by RVLSM (GPT-5.4 root, Qwen3-VL-30B sub-model) and the VideoLLM baseline (same sub-model).
Best value per entity type and metric in bold.
LLM-as-a-Judge
To complement automatic metrics, an LLM-as-a-Judge A/B test (Zheng et al. 2023) is conducted using Gemini 3.1 Pro (gemini-3.1-pro-preview) across all three configurations Qwen3-VL-30B compared with the baseline. Results in Table 3 confirm the trends
observed through automatic metrics, with the use of STD and the AMTC achieving the best overall performance.
Selected Methods
Wins
Win %
RVLSMw/ STD
106
34.2%
RVLSMAMTC
91
29.2%
RVLSMw/o STD
69
22.2%
Baseline
45
14.4%
Table 3: LLM-as-a-Judge evaluation on the VideoXum and MixTyVSum datasets, using Qwen3-VL-30B as sub-VLLM and GPT-5.4 as Root-LLM.
Figure 4: A qualitative example from VideoXum showing the summaries generated by RVLSM w/STD + AMTC, compared to the zero-shot baseline and the ground truth
Qualitative Analysis
Figure 5: A qualitative example from MixTyVSum showing the summaries generated by RVLSM w/ STD + AMTC, compared to the zero-shot baseline and the ground truth.
This section presents two representative examples highlighting scenarios in which the RVLSM approach shows clear advantages over the baseline. Figure 4 reports the result generated from a sample video of the VideoXum dataset, where the RVLSM summary exploits OCR injection to accurately identify the on-screen information and the ASR transcript to detail the specific technique demonstrated. In contrast, the baseline produces a generic description that mentions neither the instructor’s identity nor the specific content, confirming that without recursive decomposition relevant information dispersed over time tends to be lost. A second case is shown in Figure 5, a summary generated from EpicKitchens as part of the MixTyVSum annotations. Here the baseline captures only a subset of ingredients and omits secondary activities such as dishwashing and table setting, whereas the RVLSM, leveraging its recursive structure, reconstructs the full sequence of events in close alignment with the ground truth. This example illustrates how the framework scales with video duration, yielding greater improvements as content becomes longer and more complex.
Statistical Significance Tests
To rank the summarizers on each benc