依個人喜好度進行學術文獻再搜尋之研究

Abstract

近年來由於網際網路的普及與Web2.0的概念盛行，網路上充斥著大量的資料，如何讓使用者快速找到有用的資料成了重要的課題。個人化推薦系統透過建立標籤或標記(Tagging)的方式讓使用者直接參與資訊傳播與分享的過程，有效地改善資訊超載(Information Overloading)的現象。然而一般推薦系統大部分是透過間接分析使用者的興趣、習慣或是社群關聯性等來加強推薦效果，這樣的隱性學習過程容易發生無法正確過濾使用者需求或興趣而降低文章推薦的準確性。 本研究建置一套反覆式文件再搜索系統，特別加強系統與使用者之間的互動過程，以反覆且累計使用者喜愛度的方式讓使用者親自篩選文章的重要字詞，直接將個人的需求或興趣與系統進行反覆互動，再套用文章權重評分標準，以文章帶出原本隱身於中間或後面容易被忽略的文章，最後配合文件再搜尋的收斂度判斷，本研究預期可以更正確且有效地提供符合使用者需求的文件搜尋清單。

In recent years, as the internet grows and Web2.0 evolves, the network is flooded with huge amount of data, so to help users quickly dig out useful information has become an important issue. By creating tags, the personalized recommendation systems allow users to directly participate in the process of information dissemination and sharing, which effectively eliminates the phenomenon of information overloading. However, most recommendation systemsuse indirect analysis of user's interests, habits or other community association to strengthen the recommendation effects, but such an implicit learning process is prone to provideinappropriatearticles and will reduce the correctness and fitness of the suggested articles. In this study, first, we build an iterative article Re-Searching system with particular emphasis on the interactive process between the users and the system; it allows users to explicitly filter out important user preferencespersonally by an iterative and accumulative approach.Second, we calculate the article grade by applying the weighting and scoring rules, this may bring out the articles hidden in the middle or the back of article lists that are easily overlooked. Finally, users could judge the availability of the searching results by evaluating the convergence degree. To sum up, this study can be expected to provide searching article lists that are more accurate and more effective to becompliantwith user satisfaction.

Keyword: citeulike, iterative, personal, explicit, preference