以兩階段動態調整機制結合使用者喜好分析之線上活動推薦方法

Online Activity Recommendation Approach based on User Preference Analysis and Two-phase Dynamic Adjustment Strategy

黃郁淳(Yu-Chun Huang)

指導教授 ： 劉敦仁

國立交通大學/管理學院/資訊管理研究所/碩士(2017年)

摘要

近年來，多位推薦系統的研究者嘗試從龐雜的網路資訊中，透過相關度的計算，分析使用者的喜好與興趣落點，以提供有效的推薦項目。然而隨著電子商務規模的不斷擴大，使用者需要花費大量時間才能找到想閱讀的資訊，或是有興趣的活動與課程。為此，許多個人化推薦系統的研究者致力於有效且即時地找出使用者感興趣的項目。
本研究主要在探討網路資訊平台之個人化推薦系統的線上推薦機制，期望能夠在矩陣分解(NMF)以及隱含主題模式(LDA)探勘的基礎上，以閱讀新聞和點擊活動的歷史紀錄，找出新聞和活動之間的關聯性，對使用者作跨領域的喜好分析。我們也將研究數據實作至線上系統，針對目標使用者線上即時閱覽紀錄，更新喜好分數，並透過推薦清單動態調整機制的改良，提出兩階段動態調整機制，讓被推薦次數較少，或是較熱門的活動，獲得較高的曝光機會，使線上活動推薦在有限的推薦版面上，能夠更準確地獲得使用者的青睞，進而提高推薦活動點擊率。

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

並列摘要

Owing to the rapid growth and complexity of network information, a variety of research on the recommendation system aimed at analyzing user preferences and interests through different kinds of correlation calculation in recent years. With the e-commerce growing, users have to spend a great amount of time in finding what they want. Therefore, how to dig out user interests effectively and instantly had become the essential target of the designing personalized recommendation system.
We carry on a research of the online recommendation mechanism of the web-based information platform. Based on Non-negative Matrix Factorization (NMF) and Latent Dirichlet Allocation (LDA) topic model, we are able to find the association between news and online activities, and further, derive cross-domain user preferences. Then, we modify the mechanisms of dynamic adjustment strategy and proposed two kinds of Two-phase Dynamic Adjustment Strategy to an online system. In our approach, fewer recommended or popular activities can have greater opportunity to be clicked under the limited recommendation layouts.

並列關鍵字

Recommender system ； Online Recommendation ； Data Mining ； Latent Topic Model ； Matrix Factorization ； Dynamic Adjustment of Recommendation List

參考文獻

[1] S. Aciar, D. Zhang, S. Simoff, J. Debenham, Recommender System Based on Consumer Product Reviews, Proceedings of the 2006 IEEE/WIC/ACM International Conference on Web Intelligence, 2006, pp. 719-723.

[2] M. Balabanovi, Y. Shoham, Fab: content-based, collaborative recommendation, Commun. ACM, 40 (3), 1997, pp. 66-72.

[3] D.M. Blei, A.Y. Ng, M.I. Jordan, Latent dirichlet allocation, Journal of machine Learning research, 3 (January), 2003, pp. 993-1022.

[4] R. Burke, Hybrid Web Recommender Systems, in: P. Brusilovsky, A. Kobsa, W. Nejdl (Eds.) The Adaptive Web: Methods and Strategies of Web Personalization, Springer Berlin Heidelberg, Berlin, Heidelberg, 2007, pp. 377-408.

[5] I. Cantador, A. Bellog, D. Vallet, Content-based recommendation in social tagging systems, Proceedings of the fourth ACM conference on Recommender systems, Barcelona, Spain, 2010, pp. 237-240.