運用天際線方法與密度為基礎分群法用以優化興趣點推薦機制之研究

在資訊科技的長足進步之下，智慧型行動裝置與社群網路平台的使用率也隨之快速成長，有越來越多的使用者透過手持行動裝置或穿戴式電子產品，來分享和上傳各種生活資訊、心情以及評價等至這些網路平台之中，隨著基於位置的社會網絡(Location-Based Social Network)的出現，透過地方景點、場域等所產生興趣點(Points of interest)資訊也正在快速的累積著，如何分析這些大量資訊並對使用者進行有意義的推薦則成為了一項困難的任務。而在現有的研究之中，大多都旨在探討最短距離，或者使用諸如價格、城市等單一屬性作為推薦依據，鮮少有透過多項屬性作為推薦準則所制定出的路徑推薦機制。鑒於上述問題，本研究將利用以密度為基礎之分群法(Cluster Analysis)以及天際線查詢方法(Skyline)，對過於龐大的興趣點資訊進行資料縮減，將其建立成完整的興趣點地圖後，透過多準則決策分析方法(Multiple Criteria Decision Making)來進行比較及排序這些興趣點，並利用準確率(Precision)、平均絕對誤差(Mean Absolute Error)、平均平方差(Mean Square Error)、均方根誤差(Root Mean Square Error)、平均絕對百分比誤差(Mean Absolute Percentage Error)等五項指標來對所提出的方法進行評估，藉以驗證本研究之興趣點推薦系統的有效性。

關鍵字

基於位置的社會網絡；興趣點；以密度為基礎之分群法；天際線方法；多準則決策分析

並列摘要

With the rapid advancement of information technology, the usage of smart mobile devices and social networking platforms has also grown rapidly. More and more users share and upload through handheld mobile devices or wearable electronic products. Various life information, moods, and evaluations are included in these online platforms. With the emergence of the Location-Based Social Network, points of interest are generated through local attractions and fields. It is also rapidly accumulating. How to analyze this large amount of information and make meaningful recommendations to users becomes a difficult task. In the existing research, most of them are aimed at exploring the shortest distance, or using a single attribute such as price and city as the recommendation basis, and there are few path recommendation mechanisms developed through multiple attributes as the recommendation criteria. In view of the above problems, this study will use Cluster Analysis and Skyline to reduce the information of too large points of interest, and then establish them into a complete map of interest points, then compare and sort these points of interest through Multiple Criteria Decision Making, and Five methods, such as Precision, Mean Absolute Error, Mean Square Error, Root Mean Square Error, and Mean Absolute Percentage Error, were used to evaluate the proposed method to verify the effectiveness of the recommendation system of interest points in this study.