The growing volume of opinion data presents a significant challenge for policymakers striving to distill public sentiment into actionable decisions. This study aims to explore the capability of large language models (LLMs) to synthesize public opinion data into coherent policy recommendations. We specifically leverage Mistral 7B and Mixtral 8x7B models for text generation and have developed an architecture to process vast amounts of unstructured information, integrate diverse viewpoints, and extract actionable insights aligned with public opinion. Using a retrospective data analysis of the Polis platform debates published by the Computational Democracy Project, this study examines multiple datasets that span local and national issues with 1600 statements posted and voted upon by over 3400 participants. Through content moderation, topic modeling, semantic structure extraction, insight generation, and argument mapping, we dissect and interpret the comments, leveraging voting data and LLMs for both quantitative and qualitative insights. A key contribution of this thesis is demonstrating how LLM reasoning techniques can enhance content moderation. Our content moderation approach shows performance improvements using comment deconstruction in multi-class classification, underscoring the trade-offs between moderation strategies and emphasizing a balance between precision and cautious moderation. Using comment clustering, we establish a hierarchy of semantically linked topics, facilitating an understanding of thematic structures and the generation of actionable insights. The generated argument maps visually represent the relationships between topics and insights, and highlight popular opinions. Future work will leverage advanced semantic extraction and reasoning techniques to enhance insight generation further. We also plan to generalize our techniques to other major discussion platforms, including Kialo. Our work contributes to the understanding of using LLMs for policymaking and offers a novel approach to structuring complex debates and translating public opinion into actionable policy insights.

隨著Web 2.0社群媒體服務的普及化，越來越多的民眾開始運用網際網路發表自身對於政府治理的需求與看法，大量的民意資訊在網絡的交互連結下，迅速集結成可觀的網路輿情。由於網路輿情具備巨量資料的特性，使得當前各政府部門熟悉的分析方法，似乎產生適用上的困難。因而網路輿情分析的出現，成為當前政府洞察民意的新興工具。更重要的是，如何運用網路輿情分析進一步與政策面產生實質的連結，如探究網路輿情分析當中情緒分析對於政策立場解讀的可能性，對公共管理者而言更為重要。再者，網路輿情分析目前尚缺乏一套檢測方法來驗證其分析結果的信效度。因此，本研究的目的在於，運用網路輿情分析所撈取的輿情資料，比較新聞網站、社群網站、討論區及部落格四類來源在情緒分析與立場分析之差異，最後運用情緒與立場來解讀網路輿情。 研究設計，本文採用次級資料分析法及內容分析法，次級資料來自2014年行政院國發會委託政治大學蕭乃沂教授所主持的「政府應用巨量資料精進公共服務與政策分析之可行性研究」，本文以「自由經濟示範區政策」作為個案分析。研究發現，在立場分析方面，新聞網站及部落格是支持立場的言論最多；而社群網站及討論區則是反對立場的言論最多。情緒分析方面，四類來源皆以負向情緒的言論為主，正向情緒的言論相對少；透過情緒與立場的交叉分析顯示，機器會產生兩類誤判情形，第一類誤判是被機器判讀是正向情緒，但人工判讀為反對立場的言論，以社群網站的來源居多；第二類誤判是被機器判讀是負向情緒，但人工判讀為支持立場的言論，以新聞網站的來源居多。 依此研究發現，本文建議未來實務者在應用網路輿情分析時，不能僅以整體網路輿情分析的結果輕斷，必要時應將不同網路言論來源個別觀察，特別是當負向情緒的輿論出現時，應優先留意社群網站的動向。此外，針對輿情的高峰期也可對照新聞網站的分析結果，了解是否受到特定新聞報導的牽動而引起網民的討論。值得注意的是，針對社群網站中正向情緒的輿論，實務者也不能過於樂觀，因為部份正向情緒的言論可能是帶有網民「拐彎抹角」的反對。 / In the era of Web 2.0, more and more people express their opinions for public governance on the Internet. Massive public opinions are quickly generated. However, it seems difficult to analyze for government because of the feature of big data. Internet public opinion analysis(IPOA) has become new analytical methods for public managers. The purpose of this study is to use IPOA to mine large amounts of policy opinions and conduct sentiment analysis(SA) comparing with political positions analysis(PPA) in the news sites, forums, social networking sites and blogs. Finally, interpreting the network of public opinion by SA and PPA. Secondary data analysis and content analysis are applied. Secondary data collected by the Research, National Development council, the Executive Yuan. A Case Study of Free Economic Pilot Zones Policy is selected. In terms of PPA, the results reveal more supporting political opinions in the news sites and blogs. And more opposing political opinions in the social networking sites and forums. In terms of SA, four types of sources are negative emotions in large part. By cross-analysis, SA and PPA have difference on results. There will be two types of false judgments by SA with machine. One is judged positive emotion by machine, but opposing political opinions by coders, such as social networking sites. The other is judged negative emotion by machine, but supporting political opinions by coders, such as news sites. From this study, author suggests that practitioners should separately make the necessary observation of various networks rather than only determine on overall results as using IPOA. Especially, giving priority to the social networking sites when the opposing political opinions emerge. Moreover, the peak period for opposing political opinions in the social networking sites can be compared with the events in the news sites. It is noteworthy that practitioners should pay attention to the partial positive comments in social network sites with“irony”remarks.