Abstract: With the widespread integration of new loads and the continuous advancement of virtual power plant construction, diverse load structures and frequent grid-load interactions in new power system make user characterization applications to face critical challenges in data fusion, feature extraction, and business empowerment. Therefore, this paper first deeply analyzes the research status of user characteristics, then proposes a user characteristics stereoscopic application paradigm in new power system based on it, which links the application chain of "multi-source data - multi-dimensional characteristics - multi-faceted modules - multi-varied business". Next, aiming at the core technical problems of poor usability of massive multi-source data, low flexibility of characterization, and high complexity of business landing in the realization of the paradigm, based on the remarkable capabilities and agent application paradigms that have attracted much attention in the recent large language models, this paper proposesthe solutions in the aspects of adaptive data governance, conversational data interaction, automated data analysis and machine learning, and multi-agent application system architecture, respectively. Further, the case study of electricity consumption database interaction and business expansion access optimization demonstrates that the proposed solutions can implement the user characteristics application in an automated, intelligent, and personalized manner, which is expected to address the critical challenges faced by the user-side business in practice. Finally, this paper provides an outlook on the large language model empowering the control of massive resources on the user side and the leap of system intelligence level.