由大型語言模型驅動的對話式介面與圖像介面的用戶使用行為探討

Abstract

這項研究評估了大型語言模型（LLMs）在增強對話式介面（CUIs）相較於圖形介面（GUIs）在線上購物中的應用效果。GUIs 以其直觀的視覺元素為主導，而 CUIs 允許自然語言互動，但在理解上下文方面面臨挑戰。本研究探討將這些介面整合到混合式介面（HUI）中對使用者行為和體驗的影響。

先前的研究顯示，GUIs 通過提供直觀導航來提升可用性，而 CUIs 因其對話功能而備受讚譽，但在準確性和上下文理解方面存在不足。本研究採用了混合方法，開發了三個原型——GUI、CUI 和 HUI。參與者進行了購物任務，他們的體驗通過感知有用性、易用性、信任、社會臨場感、愉悅感、感知控制和專注度等指標進行了測量。量化結果顯示，GUIs 具有最高的信任水平，而 CUIs 提供了最強的社會存在感。PLS-SEM 分析強調，愉悅感顯著影響了所有介面中的易用性、有用性和社會臨場感。質性結果指出了多樣的 CUI 應用場景，但也表明了對準確性和上下文的擔憂。

這項研究建議將視覺元素整合到 CUIs 中可以增強互動性和滿意度，並強調信任和社會存在感的重要性。該研究為改進混合介面設計提供了參考，並突顯了 LLM 驅動的 CUIs 在線上購物中的潛力。未來的研究應探索這些介面對使用者行為的長期影響及進一步的 LLM 整合。

This study evaluates the effectiveness of Large Language Models (LLMs) in enhancing Conversational User Interfaces (CUIs) compared to Graphical User Interfaces (GUIs) for online shopping. GUIs are known for their intuitive visual elements, while CUIs enable natural language interaction but struggle with context understanding. This research investigates the impact of integrating these interfaces into a Hybrid User Interface (HUI) on user behavior and experience.

Previous studies show that GUIs improve usability through intuitive navigation, whereas CUIs are valued for their conversational abilities but lack accuracy and context comprehension. This study employs a mixed-method approach, developing three prototypes: GUI, CUI, and HUI. Participants completed shopping tasks, and their experiences were measured using metrics such as perceived usefulness, ease of use, trust, social presence, enjoyment, perceived control, and concentration.

Quantitative results indicated that GUIs achieved the highest trust levels, while CUIs offered the strongest social presence. PLS-SEM analysis revealed that enjoyment significantly influenced perceived ease of use, usefulness, and social presence across all interfaces. Qualitative findings highlighted diverse CUI applications but also raised concerns about accuracy and context understanding.

This study suggests that integrating visual components into CUIs can enhance engagement and satisfaction, emphasizing the importance of trust and social presence. The findings provide insights for improving hybrid interface designs and underscore the potential of LLM-driven CUIs in online shopping. Future research should explore the long-term effects of these interfaces and further integration of LLMs.