標題: 以推薦清單動態調整機制為基礎之線上活動推薦方法
Online Activity Recommendation Approach based on Dynamic Adjustment of Recommendation List
作者: 李佳徽
劉敦仁
Lee, Jia-Huei
Liu, Duen-Ren
資訊管理研究所
關鍵字: 推薦系統;線上推薦;資料探勘;隱含主題模式分析;矩陣分解;推薦清單動態調整機制;Recommender system;Online Recommendation;Data Mining;Latent Topic Model;Matrix Factorization;Dynamic Adjustment of Recommendation List
公開日期: 2017
摘要: 本研究探討新型態網路(媒體)資訊與商務交易平台的線上推薦機制,並以使用者喜好探勘分析為研究架構主軸,對於cold-start活動及資料稀疏問題,整合不同資訊來源以探勘使用者喜好。本研究考量使用者瀏覽新聞與參加活動的喜好,進行跨領域喜好分析,設計結合矩陣分解(MF)、隱含主題模式(LDA)為基礎,分析其關聯性,來預測使用者對活動喜好。 另外,線上推薦時,推薦版面有其限制,無法同時推播太多推薦項目,本研究研發線上推薦版面有限之推薦清單動態調整機制,讓使用者線上瀏覽文章時,系統可依據使用者瀏覽網頁的改變而動態調整推薦清單,以提高使用者參與活動的意願。而如何考量推薦版面限制,動態調整更換推薦清單,是線上推薦研發之重要議題,相關文獻未探討此議題。我們針對線上推薦版面有限所設計之推薦清單動態調整機制,並實作至線上系統,目前推薦機制有關的方法也大多未將方法實作至線上推薦系統做線上評估。實驗結果顯示,我們結合實務運作所設計之推薦清單動態調整機制及創新的線上推薦方法,即時預測使用者興趣並推薦活動給使用者,展現了活動推薦之成效。
This research investigates online recommendation method for new types of online news websites, mainly including the mining of user preferences on the integrated online news and e-commerce websites. By considering user preference analysis from different sources, we deal with the cold-start activities and the problem of sparse data. We conduct cross-domain analysis on user browsing news and attending activities, and design a novel method based on Matrix Factorization (MF) and Latent Dirichlet Allocation (LDA) to predict user preferences for activities. Furthermore, we develop novel mechanisms for dynamic adjustment of recommendation list to tackle the issue of limited recommendation layouts and increase users’ attending willing. Existing studies have not addressed this issue. We implement our mechanisms for dynamic adjustment of recommendation list on the online news website and evaluate online recommendation results. The experiment results demonstrate that our method can predict user interest to recommend activities for each user and achieve effective recommendation.
URI: http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070353419
http://hdl.handle.net/11536/140467
Appears in Collections:Thesis