| 摘要: | 隨著自動駕駛技術在近年快速發展，「自駕車」逐漸成為大眾關注的焦點議題之一。本研究聚焦臺灣最大網路論壇 PTT 中 2016年第二季至 2025年第一季期間與自駕車相關的貼文，透過主題分析方法探討該議題在討論內容上的演變。研究以 BERTopic 主題模型識別出討論中的 12 個主題主題，並進一步將這些主題歸納對應至技術 (TECH)、商業 (BUS)、法規 (REG) 與安全 (SAFE) 共四大框架 (Frame)。此外，本研究引入 Shannon 熵作為主題多樣性的指標，衡量不同框架下討論主題分佈的多元程度，並結合斷點分析辨識討論議題變化的關鍵時間節點。
分析結果顯示，自駕車相關討論在 PTT 上的熱度與主題內容均隨時間產生顯著變化。技術框架方面，自 2021 年起貼文量大幅攀升，對應的主題熵值亦顯著提高，代表討論內容變得更加多元；詞彙分析發現技術討論焦點已由早期的感測器、演算法轉移至近年的多模態感知、生成式 AI 與 L4 自動駕駛架構等新興概念。商業框架的討論量在經歷 2018–2020年的低迷後，於 2023年再度上揚並在 2025年初劇增，主題熵自 2021年起同步上升，顯示隨著技術突破，商業情境與想像也日益多元 (如 Robotaxi 營運模式、產業投資動態等)。法規框架的討論呈現週期性高峰：在政府研擬相關法規草案期間討論熱度上升，但長期而言其熵值逐步下降，意味討論重心趨於聚焦於特定法條細節。安全框架的討論量一直維持相對低水平，且熵值長期近乎 0，顯示此類討論主要被少數幾起重大事故事件所主導。透過雙區段線性迴歸進行轉折點檢測，本研究發現在 2021 年第一季左右出現議題多樣性的顯著提升拐點，其中技術與商業框架的討論在該時點前後呈現斜率陡增的變化。上述發現呼應了技術創新擴散的趨勢：隨著自駕車技術突破出現，相關商業議題的討論也隨之急劇擴散。總體而言，本研究不僅描繪了自駕車議題在 PTT 討論熱度上的變遷，更量化展現了討論主題內容的演進，提供政策制定者與產業決策者觀測技術議題輿情的新視角。
;In recent years, rapid advancements in autonomous driving technology have made “self-driving cars” a focal topic of public attention. This study examines the evolution of discussions about self-driving cars on PTT – the largest online forum in Taiwan – from Q2 2016 to Q1 2025. We employ semantic analysis methods to investigate how the content of these discussions has changed over time. A BERTopic model is used to identify 12 distinct latent topics in the posts, which are further consolidated into four major frames: Technology (TECH), Business (BUS), Regulation (REG), and Safety (SAFE). We then introduce Shannon entropy as a measure of topic diversity to quantify how the distribution of discussion topics within each frame varies over time. Additionally, we apply a piecewise regression approach to detect critical breakpoints in the temporal trends of topic diversity.
The results reveal significant temporal shifts in both the volume and semantic focus of PTT discussions on autonomous vehicles. In the Technology frame, post volume surged from 2021 onward, accompanied by a marked increase in topic entropy, indicating more diversified content. Keyword analysis shows that the technical discussion focus has shifted from early emphasis on sensors and algorithms to newer concepts in recent years such as multi-modal perception, generative AI, and L4 autonomous driving architectures. In the Business frame, after a lull during 2018–2020, discussion volume rose again in 2023 and spiked in early 2025; the topic entropy likewise increased from 2021, suggesting that alongside technological breakthroughs, business-related narratives (e.g., Robotaxi service models, industry investment trends) have become more diverse. Discussions in the Regulation frame showed periodic peaks corresponding to government legislative efforts, but the long-term entropy trend declined, implying that conversations became increasingly concentrated on specific legal details as regulatory frameworks matured10. Discussions in the Safety frame remained low in volume and yielded near-zero entropy, indicating that safety discourse was dominated by a few major accident events throughout the period. A two-segment linear regression identified a prominent change point around 2021 Q1 in the topic diversity trends: both the TECH and BUS frames exhibited substantially steeper post-2021 slopes (approximately 3–4 times higher than pre-2021), signifying a rapid broadening of discussed topics after that time. These findings echo patterns of innovation diffusion, wherein technological breakthroughs catalyze a proliferation of related business discourse. Overall, this research not only charts the changing popularity of the self-driving car topic on PTT, but also quantitatively illustrates the evolution of the discussion content. The study offers a novel semantic monitoring indicator for policymakers and industry stakeholders to better understand public discourse dynamics surrounding emerging technologies. |
