PIVOTSBench: Evaluating Fine-Grained Interpersonal Relationship Reasoning in Multimodal Large Language Models

2026-06-22

AI Summary

This work addresses the limited capacity of existing vision-language foundation models to reason about fine-grained, bidirectional interpersonal relationships. To this end, we introduce Social-IQ 2.0, the first multimodal benchmark grounded in psychological theory for evaluating interpersonal reasoning, constructed using YouTube data. The benchmark systematically assesses models’ ability to interpret bidirectional interpersonal dimensions and leverage critical visual cues. We propose auxiliary tasks to explicitly measure visual cue recognition and conduct comprehensive evaluations through multimodal large model testing, vision–language alignment analysis, ablation studies, and joint versus paired prediction frameworks, comparing both open- and closed-source models. Our findings demonstrate that visual modality and social role information critically influence the performance of interpersonal relationship reasoning.

Abstract

Humans possess an innate ability to understand fine-grained interpersonal relationships, which is central to everyday social interactions. Although such reasoning is inherently multimodal, it remains largely unexplored by existing multimodal large language models (MLLMs). To address this gap, we introduce PIVOTS, the first benchmark built from Social-IQ 2.0 and YouTube data to evaluate MLLMs' ability to predict bidirectional interpersonal relationship dimensions grounded in established psychology research. In addition, PIVOTS includes auxiliary tasks that assess models' ability to identify and leverage the critical visual cues underlying such predictions. We evaluate both proprietary and open-source MLLMs and conduct detailed ablation studies to analyze the effects of visual modalities and explicit social role information in conversational utterances. We further examine how joint and pairwise prediction settings benefit MLLMs in scoring bidirectional PIVOTS dimensions. Project page and resources: https://flynnzhangsx.github.io/PIVOTSBench/ .

Problem

• interpersonal relationship reasoning
• multimodal large language models
• fine-grained evaluation
• visual cues
• bidirectional relationship

Innovation

• interpersonal relationship reasoning
• multimodal large language models
• fine-grained social understanding
• visual cue grounding
• bidirectional relationship prediction