結合協同主題模式與熱門主題分析之線上新聞文章推薦

Abstract

本研究主要探討新型態的新聞網站文章推薦方法，新型態的網路資訊平台提供特定主題如生活時尚新聞等資訊，本研究主要是探討使用者在新型態的新聞網站上瀏覽新聞喜好分析，以及探討線上推薦的問題，研發新的線上新聞推薦機制。 本研究研發新的線上新聞文章推薦方法，並實際實作線上推薦機制來進行評估與比較，傳統結合矩陣分解及隱含主題模式的探勘方法並未考量文章和隱含主題的熱門程度，此外，目前推薦機制研究之相關文獻大多未將推薦方法實作整合於推薦系統平台進行線上推薦評估，本研究考量使用者瀏覽生活時尚資訊之特性，提出結合矩陣分解、隱含主題模式分析及與熱門主題分析，以及使用者長期、近期與線上喜好分析的線上推薦方法。實驗結果顯示本研究所提的方法優於傳統推薦方法，能改善線上推薦點擊率。

This research investigates news article recommendation methods for new types of online news websites which provide specific subject of information, such as life style and fashion news. The research mainly include the mining of user preferences on browsing online news, developing novel mechanisms for online recommendation. This study proposes a novel online news article recommendation method, and evaluates online recommendation results. Traditional article recommendation approaches, which combine matrix factorization (MF) and Latent Dirichlet allocation (LDA) topic model, do not consider the popularity of latent topics. Moreover, most of the existing studies do not evaluate online recommendation. This research considers the features of browsing life style and fashion news, and propose novel online recommendation methods, which combine MF, LDA, latent topic popularity analysis, and the analysis of long term, short term and online user preferences. The experimental result shows that the proposed method outperform traditional methods, and improve the click-through rate of online recommendations.