導入矩陣分群之視覺化圖書推薦系統 Visualized Book Recommender System Using Matrix Clustering

Abstract

傳統圖書推薦系統依據讀者過去的借閱紀錄，推薦相關書籍給讀者，也可以藉由讀者所屬社群的資訊，推薦讀者從沒有借閱過的書籍。然而，讀者的閱讀興趣會隨著時間改變，借閱時間越近的圖書越能反應讀者當前興趣，每筆閱讀紀錄的重要性不可等同視之。圖書借閱紀錄高維度和稀疏的特性使得資料探勘的分群方法無法有效對應。再者，為了使讀者可以從推薦結果中有效地發現所需資訊，必須導入視覺化呈現技術。因此，本研究導入時間衰減因素，提出動態閥值矩陣分群，並導入主題地圖，以提高判斷圖書推薦適性之準確率。實驗結果證實視覺化圖書推薦系統比傳統圖書推薦系統具有更高的滿意度，且雙層式主題地圖呈現比單層式主題地圖呈現更適合呈現推薦結果。Traditional library recommender system can not only employ each user’s loan history to recommends books which she(he) is interested, but also use the load history of other users who are in the same social network with the user to recommend books which she(he) never loans but may be interested in. However, as the users’ information interests are being changed continuously, the same treatment for the user library usage at different time will lead to the recommended results departure from the users’ current information needs. Moreover, as the data of library usage are highly dimensional and sparse, the traditional clustering methods can not tackle clustering issue effectively. Besides, interactive information visualization can allow users to more easily see multiple aspects of recommended results and offer a clear of items ranked by perceived interests. In order to deal with the three issues, this paper exploits time decay weight, matrix clustering using dynamic thresholds and topic maps to propose a novel visualized book recommender system. Additionally, according to experimental results of users’ satisfaction using a questionnaire, the proposed recommender system can be useful to represent the recommended results and helpful for the users to find their interests. Furthermore, two-layered topic map is more easily understood than one-layered topic map, and can effectively satisfy the users’ information needs.

Keywords

• Book recommender
• Topic map
• Matrix clustering
• Time decay
• 圖書推薦
• 主題地圖
• 矩陣分群
• 時間衰減