Interaction and Intelligence: From China's Active Open Interfaces to Agent Design

Abstract

Large language model agents could serve as autonomous assistants that excel in both interacting with users and providing intelligent responses to them. However, despite their potential as the next generation of human-computer interaction interfaces, critical gaps remain for both human-computer interaction researchers and industry practitioners in China: How can we learn from user interfaces to design agents with better interaction and intelligence attributes for Chinese users? First, contextually, existing empirical studies predominantly focus on Western user interfaces, leaving the interaction and intelligence attributes of emerging successful Chinese digital platforms lacking in-depth exploration. Without such complete interaction and intelligence analysis grounded in the local environment, it is difficult to extract transferable design implications from them for successful agent system design in China. Second, practically, current agent designs often prioritise algorithmic or model performance optimisation over more user-centric design considerations. One major reason is the limited practice guidance on how to transfer empirical findings from interfaces study into the construction of user-centric agent systems.

To address this gap, this thesis conducts a comprehensive analysis of interaction and intelligence characteristics of two kinds of emerging popular Chinese user interfaces: live-streaming interface and virtual goods trading interface through constructing two empirical-theoretical analysis frameworks. Leveraging the proposed analysis framework, we formulated a set of design principles aimed at interaction and intelligence improvement. We utilize the proposed design implications to present a complete prototype design and implementation process of a retrieval-response agent system as well as analyze and evaluate it from an engineering perspective. Ultimately, this thesis connects interface interaction insights with agent engineering practices. By leveraging in-depth empirical evidence derived from emerging Chinese platforms, it offers a practical pathway for designers and developers to build intelligent agents that effectively cater to the needs of Chinese users.

Date of Award	18 May 2026
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Jiawei MA (Supervisor) & Zhicong LU (External Co-Supervisor)