Multimodal Intent Acquisition: Integrating Textual, Visual, Auditory, and Contextual Signals for Intent-Driven Internet Interaction

The development of intent-driven Internet systems introduces a fundamental requirement: computational systems must be capable of understanding what users intend to accomplish before determining which Internet capabilities should be invoked. However, human intentions are not exclusively communicated through natural language. Users frequently express their objectives through images, screenshots, voice recordings, videos, maps, documents, gestures, and contextual information. In many situations, these non-textual inputs contain essential information that cannot be adequately represented through a short textual instruction. Consequently, an intent-driven Internet architecture based exclusively on textual requests may impose unnecessary interaction requirements and fail to capture important aspects of users' underlying objectives. This paper proposes Multimodal Intent Acquisition (MIA) as a research framework for acquiring human intentions from heterogeneous input modalities in intent-driven Internet environments. The proposed framework investigates how textual, visual, auditory, video-based, and contextual signals can be independently interpreted and jointly analyzed to identify the objectives that users intend to accomplish. Rather than treating multimodal understanding as an isolated perception or content-recognition task, MIA conceptualizes it as an upstream interaction process that transforms heterogeneous human-provided signals into task-oriented intent hypotheses suitable for subsequent intent representation, capability discovery, and autonomous Internet task orchestration. To operationalize this approach, the paper introduces the Multimodal Intent Acquisition Framework (MIAF), comprising seven interconnected functional components: multimodal input acquisition, modality-specific signal interpretation, contextual intent inference, cross-modal alignment, multimodal evidence fusion, missing-modality compensation, and intent hypothesis generation. The framework further introduces an uncertainty-aware acquisition mechanism designed to distinguish between complementary evidence, conflicting signals, insufficient information, and genuinely absent modalities. It also considers situations in which a system should request clarification rather than infer an unsupported intention. A central distinction of the proposed framework is that it addresses how user intentions are acquired from heterogeneous signals rather than how those intentions are subsequently formalized into a unified computational representation. This distinction positions MIA as a complementary research direction within the broader Intent-Driven Internet architecture. Its potential advantages include reduced dependence on textual instruction, more natural human–Internet interaction, improved utilization of contextual information, cross-modal disambiguation, more flexible task initiation, and greater accessibility across different interaction conditions. The paper examines the framework's architecture, functional mechanisms, theoretical advantages, research propositions, evaluation methodology, and implementation challenges. It also discusses the limitations associated with multimodal ambiguity, contradictory evidence, modality imbalance, privacy, hallucination, and uncertainty calibration. Importantly, the proposed advantages are presented as theoretically motivated possibilities and testable research hypotheses rather than experimentally established outcomes. This work establishes a conceptual research framework that requires subsequent implementation and empirical evaluation.