Device-Adaptive Conversation Decomposition for Generative AI: A Context-Preserving Framework for Lightweight Cross-Device Access to Long Conversations (DACD-GAI)

Generative artificial intelligence systems increasingly support long-running conversations involving research, programming, document creation, problem-solving, and multimodal interactions. As conversations accumulate messages, images, files, generated content, and contextual information, their overall size can increase substantially. Although accessing a large conversation may remain manageable on a desktop computer, the same conversation can become difficult to open, navigate, and interact with on a smartphone or another resource-constrained device. This problem is not exclusively attributable to model context-window limitations. It can also arise from excessive data transfer, client-side parsing, interface rendering, memory consumption, attachment processing, and other device-dependent operations. This paper proposes Device-Adaptive Conversation Decomposition for Generative AI (DACD-GAI), a framework that transforms an excessively large logical conversation into multiple smaller, independently accessible conversations while allowing users to choose whether the original conversation should be retained or deleted. Unlike conventional conversation branching, which primarily creates an alternative continuation from an existing message, DACD-GAI focuses on partitioning an established conversation into independently manageable units. Its central objective is to reduce the amount of information required to open and interact with an individual conversation without unnecessarily sacrificing historical continuity, message integrity, or contextual accessibility. The proposed framework introduces device-aware decomposition, user-controlled preservation policies, independent conversation loading, context-preserving partitioning, multimodal attachment management, and traceable relationships between original and decomposed conversations. It further distinguishes physical data partitioning from interface-level rendering optimization and model-context compression, establishing a unified architecture for addressing the different sources of long-conversation overhead. To the best of our knowledge, this paper proposes one of the first integrated frameworks specifically combining user-controlled original-conversation retention or deletion, device-adaptive conversation decomposition, and independently loaded conversation partitions as a unified approach to cross-device accessibility in generative AI systems. We formulate the framework, define its architecture and operational procedures, establish evaluation metrics, and propose an experimental methodology for validating its effectiveness.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24270976
Primary Topic
Speech and dialogue systems
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Device-Adaptive Conversation Decomposition for Generative AI: A Context-Preserving Framework for Lightweight Cross-Device Access to Long Conversations (DACD-GAI)

Jincheng Zhang
20 links
Zenodo (CERN European Organization for Nuclear Research)
Speech and dialogue systems
preprint

Device-Adaptive Conversation Decomposition for Generative AI: A Context-Preserving Framework for Lightweight Cross-Device Access to Long Conversations (DACD-GAI)

Jincheng Zhang
preprint en

Abstract

Generative artificial intelligence systems increasingly support long-running conversations involving research, programming, document creation, problem-solving, and multimodal interactions. As conversations accumulate messages, images, files, generated content, and contextual information, their overall size can increase substantially. Although accessing a large conversation may remain manageable on a desktop computer, the same conversation can become difficult to open, navigate, and interact with on a smartphone or another resource-constrained device. This problem is not exclusively attributable to model context-window limitations. It can also arise from excessive data transfer, client-side parsing, interface rendering, memory consumption, attachment processing, and other device-dependent operations. This paper proposes Device-Adaptive Conversation Decomposition for Generative AI (DACD-GAI), a framework that transforms an excessively large logical conversation into multiple smaller, independently accessible conversations while allowing users to choose whether the original conversation should be retained or deleted. Unlike conventional conversation branching, which primarily creates an alternative continuation from an existing message, DACD-GAI focuses on partitioning an established conversation into independently manageable units. Its central objective is to reduce the amount of information required to open and interact with an individual conversation without unnecessarily sacrificing historical continuity, message integrity, or contextual accessibility. The proposed framework introduces device-aware decomposition, user-controlled preservation policies, independent conversation loading, context-preserving partitioning, multimodal attachment management, and traceable relationships between original and decomposed conversations. It further distinguishes physical data partitioning from interface-level rendering optimization and model-context compression, establishing a unified architecture for addressing the different sources of long-conversation overhead. To the best of our knowledge, this paper proposes one of the first integrated frameworks specifically combining user-controlled original-conversation retention or deletion, device-adaptive conversation decomposition, and independently loaded conversation partitions as a unified approach to cross-device accessibility in generative AI systems. We formulate the framework, define its architecture and operational procedures, establish evaluation metrics, and propose an experimental methodology for validating its effectiveness.

Zenodo (CERN European Organization for Nuclear Research)
Speech and dialogue systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon