導入矩陣分群之視覺化圖書推薦系統

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，為使讀者可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滿意度，且雙層式主題地圖比單層式主題地圖更適合呈現推薦結果。

關鍵字

時間衰減；矩陣分群；主題地圖；圖書推薦

並列摘要

Traditional library recommender system can not only employ users' borrowing records to recommend books with similar subjects which they have read, but also use borrowing records of users who are in the same social network to recommend books they never borrow but may be interested in. However, as users' reading interests changes from time to time, treating their borrowing records at different time periods equally seems to lead the recommendation results not to meet the users' current needs. Moreover, as the borrowing records are highly dimensional and sparse, the traditional clustering methods cannot tackle clustering issue effectively. Besides, in order to allow users to examine recommendation results in multiple aspects and offer a clear picture of items ranked by users' perceived reading interests, interactive information visualization need to be implemented. Therefore, this paper exploits time decay weight, matrix clustering using dynamic threshold and topic maps to propose a novel visualized book recommender system. According to the experimental results of users' satisfaction questionnaire, the proposed recommender system can be useful to represent the recommendation results and helpful for users to find their interested books. Furthermore, two-layered topic map is easier to understand than one-layered topic map, and it can effectively satisfy the users' needs.