面對網站上大量且更新快速的資訊,使用者找尋資訊已成為一個負擔。消費者在購買時,會有其決策過程。並且在每一個決策過程階段,都有其所需的資訊與服務類型。在購前資訊蒐集階段,可以提供消費者建議方案,協助其減少「資訊負載」的問題。在方案選擇評估階段,能提供使用者相關的方案深度資訊,或主動協助使用者刪除掉不感興趣的方案,就能讓使用者更容易、更有效率地做出購買決策。此外,目前的推薦機制缺乏了「適性化」的本質;推薦的內容並沒有依據使用者線上的瀏覽行為,而做適性化地修正或刪減。並且,本研究將探討「方案建議」模組所用之推薦方法與「消費者決策過程判斷」模組所用之涉入程度計算方式對推薦效果/績效的影響,希望能提供往後適性化推薦系統之設計參考。 本研究共分為兩部份,第一部份為以消費者購買決策為基礎,實作出一個適性化推薦系統,本系統共分成三大模組,分別為方案建議模組、消費者決策過程判斷模組、資訊深化與方案內容適性模組。在方案建議模組部分,包含有產品屬性基礎與協同過濾兩種推薦方法,和建議方案產生元件,在消費者決策過程判斷模組部分,包括有點閱次數、點閱加權、瀏覽時間三種涉入程度計算方式,以及涉入程度分數資料庫,在資訊深化與方案內容適性模組部分,有鏈結適性元件、建議方案內容適性元件、與瀏覽行為收集元件。第二部份則是實施ㄧ實地實驗,藉以測試前述系統在「方案建議」模組所用之推薦方法與「消費者決策過程判斷」模組所用之涉入程度計算方式對推薦效果/績效的影響。 從本研究的實地實驗結果中,歸納出以下主要結論: 1.在推薦系統中採用不同的「適性化」設計方式,在推薦效果上有不同之成效。 2.推薦系統中加入「消費者決策過程判斷」將有助於適性化之推薦效果:以「涉入程度」做為判斷消費者決策過程之適性化推薦系統,比起沒有進行消費者決策過程判斷之系統,其推薦績效相對較好;唯不同之涉入程度計算方式─點閱次數、點閱加權、瀏覽時間之推薦績效並無差異。 3.推薦系統採用不同的推薦演算方法,其推薦績效有所不同:採用協同過濾推薦方法之推薦系統比以產品屬性為基礎推薦方法之推薦系統,其推薦績效較佳。 4.整體而言,當推薦方法採用協同過濾推薦方法,搭配消費者決策過程判斷機制中之「點閱加權」或「瀏覽時間」計算方式,在所有適性化設計方式中,是系統績效最好的。
It is a burden to seek mass and update-rapidly information in the web. Consumers have the decision process, as they buy something. They need the different kind of information and service in the different steps of decision process. They need suggestions to help them to relieve information overload in the step of collecting information. They need the deeper information about buying suggestion or help them to remove suggestions consumers are not interested in, and then they can make decision easily and efficiently in the step of choosing and evaluating. Furthermore, the recommendation mechanism lack the nature of adaption now, the recommendations are not corrected and removed adaptively by the web users' on-line browsing behavior. Finally, this study investigate the influence of the suggestion module's the recommendation method and the decision process judgment module's the way of calculating involvement on the recommendation effect and system's performance. This study includes two parts. Firstly, this study built an adaptive recommender system based on consumer decision process. This system includes three modules, which are suggestion module, decision process judgment module and information deepened and adapted module. The suggestion module includes the two recommendation methods, which are attribute-based and collaborative filtering, and suggestion-producing component. The decision process judgment module includes the three ways of calculating involvement, which are the frequency of click, the weighted frequency of click and browsing time, and involvement score database. The information deepened and adapted module includes navigation adapted component, presentation adapted component, and behavior-collecting component. Secondly, this study made a field experiment to evaluate the effects of the suggestion module's the recommendation method and the decision process judgment module's the way of calculating involvement on the recommendation effect and system's performance. After evaluation, the main conclusions are following: (1)The different adaptive design ways take different recommendation effects. (2)The decision process judgment mechanism can improve the adaptive recommendation effect: the adaptive recommender systems with the decision process judgment mechanism based on involvement take more effects than ones without the decision process judgment mechanism; nevertheless, the recommendation effect on the different ways of calculating involvement, which are the frequency of click, the weighted frequency of click and browsing time, is not significant. (3)The different recommendation methods take different recommendation effect: the recommender systems with collaborative filtering take more effects than ones with attribute-based. (4)Overall, The adaptive design ways of collaborative filtering with the ways of calculating involvement, which are the weighted frequency of click or browsing time, take the best effects in the all ones.