Can LLMs Reason About Attention? Towards Zero-Shot Analysis of Multimodal Classroom Behavior

Automated classroom engagement recognition holds substantial promise for scalable learning analytics, yet the suitability of modern Vision-Language Models (VLMs) for this task under zero-shot conditions remains largely unexplored. We present a systematic benchmark that evaluates five widely-used VLMs: CLIP, BLIP-VQA, GPT-4o, LLaVA-1.5-7B, and Qwen2.5VL-7B-Instruct across two complementary educational datasets: DAiSEE, an individual-student video dataset (300 sampled test clips), and the Student Classroom Behaviour dataset (SCB, 1,168 scene-level images). Each model is probed with three prompt variants spanning minimal, rubric-anchored, and chain-of-thought designs. Our experiments reveal three primary failure modes of zero-shot VLMs for engagement recognition: (1) near-random performance on individual students, with Cohen's kappa never exceeding 0.10 on DAiSEE; (2) severe class collapse, where models assign 85-100% of predictions to a single engagement level regardless of visual content; and (3) extreme prompt sensitivity, with accuracy swings of up to 32 percentage points on identical images depending solely on prompt phrasing. Remarkably, scene-level classification on SCB is substantially more tractable: CLIP and GPT-4o achieve kappa approximately 0.60 when prompted with behaviorally-grounded rubrics. We also document a practical barrier for deployment: GPT-4o's safety filters reject 98% of chain-of-thought requests involving individual student faces. Our findings provide a calibrated baseline and surface critical design considerations for the use of VLMs in educational observation systems.

Educational videos are a cornerstone of remote and blended learning. However, learners' fluctuating attention remains a significant barrier to effective information retention. Prior research has attempted to mitigate this by detecting and reacting to attention loss at runtime using eye tracking. Such detection has been based so far on classical machine learning classifiers trained on engineered features, such as summary statistics over learners' fixations and saccades. These methods have struggled to capture the complex, temporal nature of learner engagement, thus exhibiting moderate prediction performance. In this study, we aim to advance the detection of attention by shifting from standard engineered features to a multimodal foundation models. Using an educational eye-tracking dataset (N = 70), we investigate a novel methodology that utilizes a Vision-Language Model (VLM) to analyze video content directly with superimposed gaze data. This approach aims to leverage the semantic reasoning capabilities of foundation models to contextualize learner focus within the video stream. We evaluate the performance of this VLM-based approach using several prompting strategies with Gemini 3, but ultimately found that none of them could outperform statistical baselines. Our results provide new insights into the limitations of using VLMs for real-time educational diagnostics.

Although Multimodal Large Language Models (MLLMs) have made substantial progress, their spatial reasoning may still produce intermediate judgments inconsistent with the input image, allowing errors to propagate through the reasoning chain and affect the final answer. Existing methods mainly improve spatial reasoning through training or additional spatial information, without considering whether the reasoning process itself is faithful to the model input. Our study shows that unfaithful reasoning chains significantly reduce final-answer accuracy. To address this issue, we propose a modular and training-free framework for spatial reasoning verification and correction. The framework constructs a Spatial Evidence Graph (SEG), which associates atomic spatial evidence extracted from Chain-of-Thought reasoning with visual entities, spatial relations, source steps, and visual evidence. Spatial Evidence Reliability Assessment (SERA) evaluates the reliability of visual evidence based on object existence, localization, and geometric measurements. The framework then identifies the earliest spatial evidence unit contradicted by reliable visual evidence and guides the original MLLM to revise the subsequent reasoning and final answer. Across 15 model-dataset settings, our method achieves an average accuracy of 68.94%, outperforming the compared baselines by 8.55 percentage points on average. Our code will be open-sourced.