自動推薦於資源整合最佳化之研究-以健康休閒產業育成資源為例

The Study of Applying Automatic Recommendation to Optimize Resource Integration - A Case Study of the Incubator Resource in Health and Leisure Industry

楊銘鴻(Ming-Hung Yang)

指導教授 ： 方鄒昭聰 ； 陳瑞陽

國立臺北大學/商學院/資訊管理研究所/碩士(2013年)

摘要

近年來國內健康休閒活動風氣盛行，我國民眾從事休閒活動的人數亦逐漸增長。因此，國內健康休閒產業新創業者數量也陸續增加。透過政府的各項培育計畫與育成輔導機制實施使這些新興微型企業業者在創業初期也能夠取得適當資源協助營運。但現有之育成資源分配機制並無法完全滿足育成廠商之資源需求，對育成廠商而言也缺少一個可主動推薦且具即時性的撮合機制及工具平台供其使用，以解決即時性的資源需求或提供其潛在的資源喜好推薦。
因此，本研究考量使用者之偏好資訊以及其隱性評價行為之特徵，結合由物聯網機制所回傳之即時資訊屬性，透過歐基里德距離相似衡量、Top-K選擇以及模糊歸屬函數等方式，為使用者提供一符合其行為習慣、偏好的推薦結果，並納入即時資訊考量以解決使用者的即時資源需求。本研究並以查準率、查全率以及F1指標進行推薦結果之效能衡量，並在使用者借用次數變化、不重複借用項目數量變化以及相似群體大小變化的情況下進行效能衡量指標的趨勢觀察。透過實驗結果可知本研究所提出之推薦模型在使用者借用記錄達一定規模時可以使推薦結果具備準確性及完整性的特徵，且在相似群體比例小於2.5%時可以獲得六成左右的準確性。另一方面，經實驗觀察得知本推薦系統的推薦結果也不會因使用者的使用行為集中或分散在某些特定資源上而受到影響。
透過本研究所提出之推薦模型在實務上可使健康休閒產業育成資源做更有效率的分配，降低資源閒置的狀況發生。在研究貢獻上，透過此利用使用者行為偏好並結合即時資訊考量推薦模型的提出，提供後續推薦研究領域一個新的比較與參考模型。

關鍵字

健康休閒育成產業 ； 推薦系統 ； 協同過濾 ； 模糊理論

並列摘要

In recent years, The health and leisure activities have become very popular, the public engaged in leisure activities has increased day by day and the number of new entrepreneurs in health and leisure industry have also increased continuously. Because of the incubator plans, incubator and tutoring service provided by government, these micro entrepreneurs could operate normally the with appropriate support resources. However, the incubator resources allocation method nowadays could not satisfy the needs for incubator entrepreneurs and there is no actively recommended and real-time matching method or tools for incubator entrepreneurs to use for solving real-time needs or providing resources recommendation with its potential preferences.
Therefore, this research consider the preferences of users, the features of implicit evaluation and the real-time information transfer from Internet of Things (IoT) , it uses the methods of Euclidean distance, top-k query and fuzzy membership function to calculate a recommended results which conforms the preferences and behaviors of users. It can also include the real-time information to solve the real-time resources needs for users. This research will use precision, recall and F1 measure to evaluate the performance and observe the trend changed of recommendation in different parameters which like used times, the numbers of non-duplicate resource and the size of similar group. Results of the experiments reveal the proposed recommendation model can recommend precisely and completely when used times reached to a size , it also can get more than 0.6 precision when the similarity group rate less than 2.5%. We also observe that the recommended results were not influenced by used record centralized or decentralized on some specified items.
Through this recommendation model, we can not only allocate resources effectively but prevent resources idleness. The contribution of this study is propose a novel recommendation model, which combine with the user implicit preference and real-time information, this novel model can provide a recommendation system model for future studies as a reference or comparison.

並列關鍵字

Health and Leisure Incubator Industry ； Recommendation systems ； Collaborative filtering ； Fuzzy theory

參考文獻

5. 胡榮勝，李達生. (2010). RFID系統及EPC標準架構. 臺北市: 國立臺灣大學出版中心.

2. 吳緯閔. (2008). 網際網路服務推薦系統. 國立中央大學資訊工程研究所, 桃園縣.

3. 周家宇. (2012). 基於餘弦和模糊相似度方法之漸進式企業電子郵件分類. 國立中央大學資訊工程研究所, 桃園縣.

13. 黃良鈞. (2009). 以模糊理論為基礎之個人化英文文章推薦系統. 國立成功大學工程科學系碩博士班, 台南市.

14. 廖品姸. (2010). 以顯性評價為主之相似性推薦. 朝陽科技大學資訊管理系碩士班, 台中市.