- 學位論文
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
Research on optimize an application market’s application recommendation mechanism by quality consideration and intelligent semantic network analysis
摘要
網際網路日益成熟，各式各樣應用程式(Application，App)不斷被推出於應用市集平台之中，使用者透過平台所提供應用程式搜尋功能，進行應用程式搜尋時，平台中儲存了巨量應用程式資訊，導致使用者無法從結果之中找出滿足使用者需求的應用程式。雖然平台有提供以類別的方式進行過濾，使用者透過類別的方式進行應用程式過濾，會因為應用程式本身被分到的類別所影響，導致有些滿足需求應用程式無法被搜尋出來。而且使用者在選擇時，不單會針對功能進行考量，使用者會針對平台所提供的描述作為選擇的依據。如何設計有效方法在平台之中，讓使用者取得滿足需求應用程式，是一個非常值得研究議題。有鑑於此，本研究以應用市集平台做為研究議題。探討使用者在應用市集平台中進行應用程式搜尋時，搜尋出來應用程式數量過多的問題。 本研究設計利用資料探勘技術建立語意網路，並透過多屬性決策分析，提供基於品質考量智慧語意網路分析推薦機制，期望透過此機制讓推薦結果更貼近使用者對應用程式之需求。基於品質考量智慧語意網路分析推薦機制，運用TF-IDF統計方法進行非結構化文章處理，並運用Top-K方法將TF-IDF統計出來之結果中，重要程度較高前K個詞彙，進行關聯式法則分析，找出類別之間隱含關聯，透過詞彙之間關聯建立語意網路；多屬性決策分析，利用使用者所設定應用程式屬性權重加以計算，確保推薦出的應用程式，能滿足使用者需求。並且藉由召回率(Recall)、準確率(Precise)及F1指標進行系統評估，透過整合性科技接受模式 (Unified Theory of Acceptance & Use of Technology，UTAUT)，以問卷調查的方式，進一步探討本研究所提出之研究架構，滿意度與使用意願。 本研究設計以資料探勘為基礎建立語意網路，並結合多屬性決策分析進行智慧化推薦，相信有一定程度的貢獻。未來研究方向可針對使用者過去自身使用經驗修正多屬性決策分析權重矩陣。也可考慮加入各個屬性之間正向與反向影響，作為多屬性決策分析考慮指標之一。此外也可在推薦方法上可以考量基因演算法、支援向量機等方法來強化推薦的結果。
並列摘要
As internet becomes more mature, a wide range of Applications (App) are constantly introduced in the app store platform. When users conduct app search via the search function provided by the platform, the huge app information stored in the platform may cause users not able to find the app meeting their needs among the search results. Although the platform provides search-via-category function, the category in which the app is defined may influence the search result and that certain apps which satisfy users’ needs may not be sorted out. Moreover, when users are making choices, they would not only take the functions into consideration but also the descriptions provided by the platform for reference of choices. Thus, how to design an effective method to be applied in the platform, so that users can obtain apps that meet their needs, is a very worthwhile search topic. Therefore, this study uses app store platform as a research topic to explore the issue of excessive sorted app when users conduct app search in an app store. This study was designed to establish semantic network via data mining technique, and through multi-attribute decision analysis to provide a quality-based intelligent semantic network analysis and recommendation mechanism, hoping to allow the search results further meet users’ needs of apps. Quality-based intelligent semantic network analysis and recommendation mechanism adopts TF-IDF statistical method for unstructured data processing and Top-K method to process the first K-terms among the results via association rule analysis to find identify the implicit association between categories; and establish semantic network through the association of terms. Multi-attribute decision analysis uses the app attribute weight set by users to calculate and ensure the recommended apps can satisfy users’ needs. Furthermore, system evaluation is conducted via Recall, Precise, and F1 indicator, and through UTAUT(Unified Theory of Acceptance & Use of Technology) and questionnaire survey, the research structure, satisfaction, and users’ willingness to use is further explored. This study was designed to establish semantic network via data mining technique and combines multi-attribute decision analysis to process intelligent recommendation; we believe there is a certain degree of contribution of this study. Future research direction may focus on users’ past usage experience to adjust multi-attribute decision analysis weight matrix. Also, the influence of positive and negative correlation between each attribute can be put into consideration as one of the indicator for multi-attribute decision analysis. Moreover, the use of genetic algorithm, support vector machine and others can be considered as alternative recommendation method to strengthen the result of recommendation.