Bidirectional Communication Preference Management for Generative AI Systems: A Unified Framework for Language, Interaction Timing, and Communication Modality Adaptation
Generative artificial intelligence (GenAI) systems increasingly participate in sustained, personalized, and context-dependent communication with human users. However, communication personalization is frequently implemented through isolated mechanisms, such as response-style instructions, language selection, interaction history, or modality preferences, rather than through a unified system that explicitly records, manages, and applies the communication preferences of both interacting parties. This paper proposes the Bidirectional Communication Preference Management Framework (BCPMF), a unified conceptual and architectural framework for representing, maintaining, and applying communication preferences between generative AI systems and their users. The framework introduces three foundational preference dimensions: language preferences, communication-time preferences, and communication-method preferences. Its defining characteristic is the separation of preference ownership from preference application: users and AI systems can each maintain independent preference profiles, while the system supports bidirectional, unidirectional, and asymmetric configurations. The framework further introduces explicit preference declaration, implicit preference inference, preference confirmation, temporal validity, contextual adaptation, conflict resolution, preference synchronization, historical versioning, authorization, and revocation. Communication preferences are represented as structured, attributable, and updateable records rather than being treated exclusively as informal instructions embedded in conversational histories. A preference orchestration mechanism evaluates applicable preferences before initiating or generating an interaction, while preserving higher-priority system requirements, user consent, privacy constraints, and contextual appropriateness. The proposed architecture distinguishes stable preferences from temporary requests and separates a party's stated preferences from inferred behavioral tendencies. This distinction is particularly important when communication preferences change across tasks, relationships, time periods, and interaction environments. The paper establishes a conceptual foundation, formalizes the preference representation and management processes, identifies research propositions, and develops an evaluation methodology for future empirical investigation. BCPMF aims to advance conversational personalization from isolated response customization toward structured, persistent, and potentially reciprocal communication preference management.
Authors
- Jincheng Zhang (ORCID: https://orcid.org/0009-0005-1860-0009)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24280885
- Primary Topic
- Speech and dialogue systems
- Type
- preprint