具動態時段之適地性協同過濾推薦系統

近年由於行動手持裝置的發展與普及，適地性行動商務應用需求也大幅增加。提供行動用戶最新、最正確且符合用戶需求的在地資訊也成為最主要的挑戰。傳統推薦系統只考慮過去用戶的歷史評價做為推薦依據，實際上用戶喜好與興趣景點 (Point of Interest, POI) 熱門度會隨時間不斷改變。為了讓推薦項目能更符合的目前時空條件，本論文提出結合期望相對距離、用戶近期喜好與POI新鮮度之具動態時段之適地性協同過濾推薦系統 (Location-based Collaborative Filtering Recommendation System with Dynamic Time Periods, LCFDTP)，利用POI新鮮度過濾演算法，減少無效推薦機率及不必要之相似度計算時間，並透過近期喜好時間函數推薦給用戶符合用戶近期喜好之POI，改善傳統推薦機制隨時間日益偏差之推薦誤差。並進一步考量用戶移動性，加入期望相對距離讓位於用戶移動路徑上之POI能優先被系統推薦，減少行動用戶迴轉機率。根據模擬實驗結果，本論文所提出之LCFDTP相較於TWCF、TPPCF及DFBT策略，不僅減少推薦誤差及更高的推薦覆蓋率，也減少平均推薦時間。在推薦誤差方面，在用戶速度為50 km/hr，查詢範圍為0.5 km時，LCFDTP之推薦誤差平均優於其他策略約61 %。在推薦覆蓋率方面，平均優於其他策略約9 %。最後，在平均推薦時間方面，平均優於其他策略約62 %。因此本論文所提出之LCFDTP更能符合行動用戶在行動環境下之查詢。

關鍵字

時間動態；協同過濾； LBS ；期望相對距離；評分相似度

並列摘要

Demands for the application of the location-based mobile commerce have highly increased because of the development and popularity of handheld devices in recent years. To provide mobile users with the latest, most accurate localization information meeting user requirements is the major challenge. The traditional recommendation system only uses the historical assessments of users as the recommendation reference. In fact, user’s interest and the popularity of the point of interest (POI) will change frequently over time. It proposed the location-based collaborative filtering recommendations system of dynamic time period (LCFDTP), which combines expected related distance, user’s recent interest and the POI freshness. It adopt the algorithm of POI freshness to reduce invalid recommendation and unnecessary calculation of similarity. Through the function of the recent interest, it could recommend the POIs that the user likes recently and improve the recommend error. Further, it used expected factor to regulate the related distance in order to have priority to recommend the POIs in the user’s moving direction. The simulation result showed that LCFDTP has not only less recommend error and more recommend coverage, but also less average recommend time than TWCF, TPPCF and DFBT. In recommend error, when user’s velocity is 50 km/hr and query range is 0.5 km, LCFDTP is better than the others about 61 %. In recommend coverage, LCFDTP is better than the others about 9 %. Finally, in average recommend time, LCFDTP is better than the others about 62 %. These showed that LCFDTP is more suitable to use in the mobile environment.

並列關鍵字

Temporal Dynamic ； Collaborative Filtering ； LBS ； Expected Related Distance ； Similarity of Evaluation

參考文獻

[88] 蔡秉錡、陳奕廷、趙延祥，「全國路況中心建置成果與維運」，地理資訊系統季刊，第一卷，第1期，2007，第35-41頁。

[91] 臺北市公共運輸處大眾運輸科，臺北捷運各站旅運量，臺北市政府交通局，2012，4月。

[1] J. E. Padgett, C. G. Gunther, and T. Hattori, “Overview of Wireless Personal Communications,” IEEE Communications Magazine, vol. 33, pp. 28-41, Jan. 1995.

[9] J. L. Herlocker, J. A. Konstan, L. G. Terveen and J. T. Riedl, “Evaluating collaborative filtering recommender systems,” ACM Transactions on Information Systems (TOIS), vol. 22, Issue 1, Jan., 2004.

[12] L. Baltrunas, B. Ludwig, S. Peer and F. Ricci, “Context relevance assessment and exploitation in mobile recommender systems,” Personal and Ubiquitous Computing (Published Online), 2011.

國際替代計量

具動態時段之適地性協同過濾推薦系統

全文下載

主題瀏覽