共分群行動位置相依推薦系統

由於網路上資料爆炸性成長，以及智慧型手機日漸普及，使得推薦系已被統廣為應用，為了降低系統的回應時間，事先將資料預處理是相當重要的議題。若要提供更便利的適地性服務予使用者，系統後端可事先替使用者過濾不必要的資訊，再依照使用者目前位置及使用者的移動方向，讓推薦清單的項目更符合使用者未來可能造訪的位置。本論文提出了共分群行動位置相依推薦系統(CLBRS)，將資料庫中興趣景點透過共分群演算法作適當的分類，讓系統於使用者發出查詢需求時，能夠快速過濾出使用者感興趣之同類景點。對於使用者周遭尚未評分過的景點，經由計算使用者間評分相關度，可替使用者做臆測評分的動作。透過計算興趣景點轉向權重值，可依照使用者短時間內轉動方向，適度調整於各方向的景點權重值，讓產生之推薦清單更貼近使用者移動方向，使推薦項目較符合行動用戶的需求。根據模擬實驗結果，本論文所提出之CLBRS相較於比較策略RGSCU、IFCCF及UCICF，可有效降低推薦誤差率與平均推薦計算時間。當行動用戶以步行速度每小時4公里，查詢範圍半徑0.5公里之移動狀態發出查詢需求時，CLBRS之推薦誤差率優於其他策略約低39 %，推薦計算時間約優於其他策略約少32 %，而推薦覆蓋率只低於其他策略約11%。當行動用戶以行車速度每小時40公里，查詢範圍半徑0.5公里、1公里及1.5公里之移動狀態發出查詢需求時，CLBRS之推薦誤差率平均優於其他策略約低49 %，平均推薦計算時間約優於其他策略約少41 %，而平均推薦覆蓋率只低於其他策略約13%。依實驗結果顯示，CLBRS具有較佳的推薦精準度與較快的回覆速度。

關鍵字

適地性服務；共分群演算法；評分相關度；興趣景點轉向權重值

並列摘要

By the mass information over the Internet and the popularization of smart phones, it becomes an important issue on filtering redundant data via location-based service (LBS). The system could provide point of interest (POI) lists based on user’s current position or the predicted location of users later. We proposed a Co-means Location-Based Recommendation System (CLBRS) to cluster data to speed up searching time, and predicted the score of unrecorded POIs through user’s rating relevance. This study also defined the turning weight of POIs to reflect the moving direction of users. The POI along user’s moving direction would be recommended preferentially. The simulation results showed that our policy CLBRS had less recommendation error and system calculation time than RGSCU, IFCCF and UCICF. When user’s velocity is set to 4 km/hr with 0.5 km query range, CLBRS in recommendation error was less than others about 39 % and in system calculation time was shorter than others about 32 %. Nevertheless, CLBRS in recommendation coverage was only less than others about 11 %. When user’s velocity is set to 40 km/hr with 0.5 km, 1.0 km and 1.5 km query range, CLBRS in recommendation error was less than others about 49 % and in system calculation time was shorter than others about 41 %. However, CLBRS in average recommendation coverage was only less than others about 13 %. The experiment results showed that CLBRS had better recommendation accuracy and shorter response time.

並列關鍵字

LBS ； Co-means Clustering ； Rating Relevance ； Turning Weight of POI

參考文獻

[45] H. Yan, B. Zhang, Y. Zhang, F. Liu, and Z. Lei, "Web Usage Mining Based on WAN Users' Behaviors," in Proceeding of the 6th International Conference on Wireless Communications Networking and Mobile Computing, Chengdu, China, Sep. 2010, pp. 1-5.

[4] A. Desai, Z. Laliwala, and S. Chaudhary, "Context and Location Based Service Grouping and Group Notification," in Proceeding of the Asia-Pacific Services Computing Conference, Singapore, Dec. 2009, pp. 227-232.

[51] D. Pellege and A. W. Moore, "X-means: Extending K-means with Efficient Estimation of the Number of Clusters," in Proceeding of the Seventeenth International Conference on Machine Learning, Stanford, USA, Jun.-Jul. 2000, pp. 727-734.

[46] D. G. Moradi and M. Mahdavi, "A New Collaborative Filtering Algorithm Using K-means Clustering and Neighbors' Voting," in Proceeding of the 11th International Conference on Hybrid Intelligent Systems, Malacca, Malaysia, Dec. 2011, pp.179-184.

[13] X. Chen, F. Zhang, M. Sun, and Y. Luo, "System Architecture of LBS Based on Spatial Information Integration," in Proceeding of the International Geoscience and Remote Sensing Symposium, vol. 4, Sep. 2004, pp. 2409-2411.

國際替代計量

共分群行動位置相依推薦系統

全文下載

主題瀏覽