Research

Paper

AI LLM March 04, 2026

Evolutionary Multimodal Reasoning via Hierarchical Semantic Representation for Intent Recognition

Authors

Qianrui Zhou, Hua Xu, Yunjin Gu, Yifan Wang, Songze Li, Hanlei Zhang

Abstract

Multimodal intent recognition aims to infer human intents by jointly modeling various modalities, playing a pivotal role in real-world dialogue systems. However, current methods struggle to model hierarchical semantics underlying complex intents and lack the capacity for self-evolving reasoning over multimodal representations. To address these issues, we propose HIER, a novel method that integrates HIerarchical semantic representation with Evolutionary Reasoning based on Multimodal Large Language Model (MLLM). Inspired by human cognition, HIER introduces a structured reasoning paradigm that organizes multimodal semantics into three progressively abstracted levels. It starts with modality-specific tokens capturing localized semantic cues, which are then clustered via a label-guided strategy to form mid-level semantic concepts. To capture higher-order structure, inter-concept relations are selected using JS divergence scores to highlight salient dependencies across concepts. These hierarchical representations are then injected into MLLM via CoT-driven prompting, enabling step-wise reasoning. Besides, HIER utilizes a self-evolution mechanism that refines semantic representations through MLLM feedback, allowing dynamic adaptation during inference. Experiments on three challenging benchmarks show that HIER consistently outperforms state-of-the-art methods and MLLMs with 1-3% gains across all metrics. Code and more results are available at https://github.com/thuiar/HIER.

Metadata

arXiv ID: 2603.03827
Provider: ARXIV
Primary Category: cs.MM
Published: 2026-03-04
Fetched: 2026-03-05 06:06

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.03827v1</id>\n    <title>Evolutionary Multimodal Reasoning via Hierarchical Semantic Representation for Intent Recognition</title>\n    <updated>2026-03-04T08:25:05Z</updated>\n    <link href='https://arxiv.org/abs/2603.03827v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.03827v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Multimodal intent recognition aims to infer human intents by jointly modeling various modalities, playing a pivotal role in real-world dialogue systems. However, current methods struggle to model hierarchical semantics underlying complex intents and lack the capacity for self-evolving reasoning over multimodal representations. To address these issues, we propose HIER, a novel method that integrates HIerarchical semantic representation with Evolutionary Reasoning based on Multimodal Large Language Model (MLLM). Inspired by human cognition, HIER introduces a structured reasoning paradigm that organizes multimodal semantics into three progressively abstracted levels. It starts with modality-specific tokens capturing localized semantic cues, which are then clustered via a label-guided strategy to form mid-level semantic concepts. To capture higher-order structure, inter-concept relations are selected using JS divergence scores to highlight salient dependencies across concepts. These hierarchical representations are then injected into MLLM via CoT-driven prompting, enabling step-wise reasoning. Besides, HIER utilizes a self-evolution mechanism that refines semantic representations through MLLM feedback, allowing dynamic adaptation during inference. Experiments on three challenging benchmarks show that HIER consistently outperforms state-of-the-art methods and MLLMs with 1-3% gains across all metrics. Code and more results are available at https://github.com/thuiar/HIER.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.MM'/>\n    <published>2026-03-04T08:25:05Z</published>\n    <arxiv:comment>Accepted by CVPR 2026</arxiv:comment>\n    <arxiv:primary_category term='cs.MM'/>\n    <author>\n      <name>Qianrui Zhou</name>\n    </author>\n    <author>\n      <name>Hua Xu</name>\n    </author>\n    <author>\n      <name>Yunjin Gu</name>\n    </author>\n    <author>\n      <name>Yifan Wang</name>\n    </author>\n    <author>\n      <name>Songze Li</name>\n    </author>\n    <author>\n      <name>Hanlei Zhang</name>\n    </author>\n  </entry>"
}