Multi-modal Large Model Driven Whole House Decoration KnowledgeQuestion and Answer and Scheme Automatic Generation
Keywords:
Question and answer, Multi-modal large model, Whole house decorationAbstract
Focusing on the pain points existing in traditional decoration services, such as asymmetric information, vague demand
expression and high iterative cost of schemes, this paper puts forward a whole house decoration knowledge quiz and scheme automatic generation system based on multi-modal large model. The system adopts a hierarchical micro-service architecture,covering four core modules: multi modal input understanding, knowledge and intention center, scheme generation and optimization, and presentation and interaction. In the process of multi modal input understanding, by integrating heterogeneous inputs such as text, sketch, photo and voice, with the help of advanced technologies such as large language model, visual Transformer and graph neural network, the user's spatial constraints, style preferences and functional requirements are accurately analyzed, and a unified semantic representation is generated. The knowledge and intention center combines the domain knowledge base to deeply analyze and refine the user's intention, which provides strong support for the scheme generation. The scheme generation and optimization layer automatically generates a complete decoration scheme including 3D layout, bill of materials and budget estimation based on the parsed user intention and knowledge base information, and supports dynamic adjustment and optimization. The experimental results show that the system is superior to the traditional manual design and single-mode and two-mode systems in terms of intention accuracy, constraint satisfaction and material recommendation performance. The generation time of the first draft of the scheme is greatly shortened, the cost of a single iteration is reduced to zero, and the user satisfaction is improved.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Zhang Lijun (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.
Authors retain copyright and grant the journal the right of first publication. This work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Provided that appropriate credit is given and the original publication in this journal is properly cited, others may copy, distribute, transmit, and adapt the work, including for commercial purposes. Authors may also deposit the published version in institutional repositories or on personal websites with a full citation to the final published article and a link to the journal page.