以概念空間搜尋個人資訊的桌面協同標記系統

Abstract

隨著網際網路的出現，資訊量急速成長並造成了資訊超載的問題。資訊超載指的是人們擁有太多資訊而造成決策的困難。為了解決此問題，協同標記系統建立一種稱為通俗分類的分類架構來整理網路上的資源。通俗分類累積個人自由標記各種資訊和物件的結果，利用群眾智慧建立分類架構。與傳統由專家定義的分類架構相比，通俗分類更適用於組織網路上巨大的資訊量。但是，協同標記系統的特徵和他們通俗分類的特性，並不能直接適用於個人環境。 我們設計了一個讓協同工作者可以標記文件的桌面協同標記系統。本系統根據個人標記文件的行為，利用概念空間建立通俗分類。概念空間提供推薦新文件標記、關聯式主題搜尋、同義詞控制功能。除此之外，我們使用文件的關鍵字而非全文，建立已被標記文件與未被標記文件之間的關係，藉此維護使用者個人文件的隱私。 實驗結果顯示使用者接受系統所推薦給新文件標記的比率，於使用者標記五到六篇文件後就成長了百分之十。除此之外，若新文件與之前被標記過的文件屬於相同主題，系統推薦的標記將更容易被使用者接受。比較關聯式主題搜尋與傳統關鍵字搜尋的結果後，我們發現以常用標記搜尋時，關聯式主題搜尋的搜尋結果比關鍵字搜尋來得好。關聯式主題搜尋的平均準確率、召回率、F-measure比關鍵字搜尋分別提高了12.12%、23.08%、26.92%。 桌面協同標記系統為協同工作者提供了多面向描述資源的功能，並且給予使用者描述它們的建議。另外，關聯式主題搜尋也提供了比關鍵字搜尋更有效的搜尋資源方式。

With the advent of Internet, the amount of information grows dramatically and causes information overload which refers to the situation of having too much information to make a decision. To solve this problem, collaborative tagging systems form a categorization called folksonomy to organize web resources. A folksonomy aggregates results of personal free tagging of information and objects to form a categorization structure utilizing intelligence of crowd. Compared with traditional taxonomy established by expert catalogers, folksonomy is more appropriate for organizing huge amount of information on the Web. However, characteristics of collaborative tagging systems and their folksonomy make them infeasible for organizing resources in personal environment. We design a desktop collaborative tagging (DCT) system which allows collaborative workers to tag their documents. Folksonomy in DCT is constructed by aggregation of personal tagging results and represented by concept space. Concept space offers synonym control, tag recommendation, and relevant search. Besides, we build relations between tagged documents and untagged ones by extracting document’s features instead of using full text for protecting personal document privacy. Experimental results show that adoption rate of recommended tags for new documents grows 10% after users tagged five or six documents. Besides, when new documents are on similar topic of previously tagged ones, DCT can recommend tags with higher adoption rate. It is observed that relevant search in DCT is superior than keyword search when we use frequently used tags as queries. On average, precision, recall, F-measure of DCT are 12.12%, 23.08%, 26.92% greater than those of keyword search. DCT enables a multi-faceted categorization of resources for collaborative workers. It also recommends tags for categorizing resources to make categorization easier. In addition, we provides relevant search, which is more effective than traditional keyword search, for searching personal resources.