非營利新聞網站瀏覽量之分類研究台灣某新聞網站之個案分析

Predicting News Traffic for Non-Profit Organizations by Classification Models: Case Study of a Taiwanese News Website

隨著大數據的普及和資訊科技的重大發展，資料科學在商業應用中扮演著日益重要的角色。數位平台已取代傳統新聞媒體成為民眾主要獲取新聞的途徑，為了更理解使用者的需求和行為，研究者探討了不同方法與應用，包括統計方法和資料探勘等。這些方法可深入瞭解網站使用者的特徵、偏好和其他關鍵指標，並提供有意義的改進建議。本研究旨在協助非營利組織的新聞網站，透過深入探討和分析使用者行為，以提供未來經營方向參考；應用機器學習方法，對非營利組織經營之新聞網站進行個案分析，包括目標受眾、內容策略、和成效評估等。本研究使用2022年1月1日至12月31日的網站瀏覽數據，總計1,630,811次瀏覽，獲得3,047個有效樣本，並擷取包括瀏覽量、年齡、性別等11個特徵，比較Naïve Bayes (NB)、Sequential Minimal Optimization (SMO)和Bootstrap aggregating (Bagging)三種演算法預測使用者瀏覽率；結果顯示SMO被視為構建預測模型的首選，其次為NB模式。本研究不僅為非營利組織提供了網站策略建議，亦為未來相關研究提供了參考和啟示，期望後續研究能夠深化和擴展對新聞網站瀏覽率的預測分析。

Due to the popularity of big data and the powerful development of information technology, data science has been playing an increasingly crucial role in business applications. Digital platforms have replaced traditional news media as the primary source of news for the public. In order to better understand user needs and behaviors, numerous studies have explored different methods and applications, including statistical methods and data mining. These methods can provide in-depth insights into the characteristics, preferences, and key indicators of website users, and offer meaningful improvement suggestions. This study aims to assist non-profit organization news websites in formulating future operational directions by analyzing user behaviors. Machine learning methods, including Naive Bayes (NB), Sequential Minimal Optimization (SMO), Bootstrap aggregating (Bagging), are applied to conduct case analysis of the non-profit organization-operated news website, covering aspects such as target audience, content strategy, and performance evaluation. This research utilizes website browsing data from January 1 to December 31, 2022, totaling 1,630,811 visits and integrating 11 features, such as page views, age, gender, etc., resulting in 3,047 effective data samples. Three algorithms (Naive Bayes, SMO, and Bagging) are compared in predicting user engagement rates, where SMO is identified as the preferred choice for constructing predictive models, and NB is the second preferred. This study not only provides website strategy recommendations for non-profit organizations but also serves as a reference and inspiration for future related research. Future studies are expected to expand and explore the analysis of predicting news website engagement rates.