運用Skyline方法於最佳化應用市集App推薦機制之研究

A Study on the Optimization of Application Recommendation Mechanism Using Skyline

張巧旻(Chiao-Min Chang)

指導教授 ： 柯志坤

國立臺中科技大學/資訊與流通學院/資訊管理系碩士班/碩士(2015年)

https://doi.org/10.6826/NUTC.2015.00029

摘要

全球網際網路的使用者從1995年的0.6%至2014年39%，可以顯示出人們漸漸的習慣網際網路所帶來的便利性，而網際網路的發展也促進了行動手機的發展，從1995年全球有1%的行動手持裝置使用者至2014年全球有73%的行動手持使用者，其中有40%的使用者是使用智慧型行動裝置，而智慧型行動裝置則是透過下載應用市集平台上的應用程式(Application，App)來擴充裝置的功能，使得目前應用市集上的各式各樣的應用程式數量越來越多，造成使用者在從如此巨量的應用程式中搜尋出滿足需求的應用程式是非常困難的。雖然目前網際網路上的搜尋技術有提供關鍵字搜尋、類別搜尋、熱門度搜尋跟語意網路搜尋等技術，但關鍵字搜尋是使用者必須清楚知道自己的需求才可以快速找到應用程式；而類別、熱門度搜尋則只針對單一個應用程式的資料來做排名，對使用者來說是不足夠的，因為使用者可能會考慮到價格、檔案大小、評價與應用程式資訊描述等應用程式資料；而語意網路技術則因為門檻值設定的問題，將部分重要關聯刪除，且推薦出較不相關的應用程式給使用者。
因此如何設計一個有效的方法應用至應用市集平台的搜尋系統中，推薦出滿足使用者的應用程式，是一個值得研究的議題，本研究探討使用者在如此巨量的應用程式中，搜尋出滿足需求的應用程式是非常困難的問題。
本研究設計利用Skyline方法來建立Skyline語意網路，由Skyline語意網路來強化總體語意網路的不足，並透過多屬性決策分析，提供運用Skyline方法於最佳化應用市集App推薦機制，透過此機制讓推薦出來的應用程式更貼近使用者需求。本研究利用TF-IDF方法將結構化的應用程式描述進行文字擷取，並透過關連法則來建立總體語意網路。接著透過Skyline方法將含有使用者輸入的關鍵字之應用程式集合中，較差的應用程式淘汰，並以Skyline查詢集合來建立Skyline語意網路，利用Skyline語意網路來強化總體語意網路，並結合多屬性決策分析中的ELECTRE法推薦出最佳化的應用程式。本研究也利用準確率(Precision)、回應率(Recall)與F1指標(F1-Measure)來與過去相關研究進行評估。
本研究運用Skyline方法於最佳化應用市集App推薦機制的研究適用於推薦系統，且未來可以針對應用程式的屬性選擇中採用主成分分析或者是資訊獲利等方法；針對語意網路的建立中可以採用形式概念分析來建立；而在語意網路的推薦過程中可以採用貪婪法、動態規劃演算法或環狀路徑等方法來推薦出不同的應用程式。

關鍵字

Google Play ； Skyline方法 ； 資料探勘 ； 語意網路 ； 多屬性決策分析

並列摘要

The percentage of Internet users around the globe has jumped from 0.6% in 1995 to 39% in 2014, suggesting that people have become accustomed to the convenience brought forth by the Internet, which also promoted the development of mobile phone, as exemplified by the growth of mobile device users around the world from 1% in 1995 to 73% in 2014, among which 40% are smartphone users. Smart mobile devices acquire added features through downloading applications (APP) from application market, spawning a great diversity of applications on nowadays application market, which in turn makes it hard for users to search for the ones that satisfy their needs among a staggeringly large amount of applications. Notwithstanding current Internet search technology features keyword search, category search, popularity search, semantic network. But when using keyword search, users have to clearly understand their needs before finding the corresponding applications swiftly. Yet category search and popularity search merely rank according to the data of a single application. This is not enough for users since they may take information such as price, file size, evaluation, and information description into consideration. The semantic network, on the other hand, may recommend less relevant applications to users due to the deletion of a part of relevant information caused by threshold value settings
Therefore, designing an effective method which can be incorporated into the search system of application market and recommend applications that satisfy the users’ needs has become an issue worth studying. This study discusses how searching for applications satisfying the users’ needs among a large amount of applications turns out to be a herculean task.
This study uses Skyline to establish Skyline Semantic Network, with a view to remedying the deficiencies of the overall semantic network. In addition, through multiple-criteria decision analysis, this study provides Skyline to optimize application recommendation mechanisms so as to make the recommended applications more satisfying to users. Moreover, this study adopts TF-IDF to capture words from structured applications, and, through association rules, establishes an overall semantic network. Afterwards, through Skyline, this study eliminates inferior applications within the application clusters containing keywords typed in by users, establishes Skyline semantic network through Skyline search clusters, reinforces the overall semantic network through Skyline semantic network, and combines the optimized applications recommended by ELECTRE in multi-attribute decision making analysis. Precision rate, recall rate, and F1-Measure are also used to evaluate related past studies.
This study adopts Skyline in optimizing the recommendation mechanism of application market, and is generally applicable to recommendation mechanisms. On top of that, methods such as principal component analysis and information gain can be adopted in the attribute selection of application in future studies. The establishment of semantic network can be completed through formal concept analysis. During the recommendation process of semantic network, methods such as greedy method, dynamic programming and circular path analysis can be adopted to recommend different applications.

並列關鍵字

Google Play ； Skyline ； Data Mining ； Semantic Network ； Multiple-Criteria Decision Analysis

參考文獻

[5] 任昶諭. (2014). 基於品質考量智慧語意網路分析於最佳化應用市集軟體推薦機制之研究. 臺中科技大學資訊管理系碩士班學位論文, 1-163.

[6] 李振主. (2014). 以科技接受模式探討智慧型手機使用者對手機應用程式 (App) 之付費意願¬—以 Google Play 商店為例. 義守大學資訊管理學系學位論文, 1-63.

[7] 林士彥. (2004). 應用多屬性決策評價觀光旅館業聲望. 觀光研究學報, 10(4), 101-123.

[10] 張景皓. (2012). 免費 App 關鍵成功因素之研究. 淡江大學資訊管理學系碩士班學位論文, 1-50.

[11] 戚玉樑, 李志偉, & 陳仲儼. (2004). 網路服務架構下之最佳化流程組合的選擇方法. Electronic Commerce Studies, 2(3), 297-314.