社會性標籤(social tagging)系統讓人們可以加註標籤於資料上,環繞著標籤而衍生的研究及應用因而蓬勃發展。由於標籤的多樣性和個體認知的差異,人們經常找不著合適詞彙做為標籤,系統也蒐集不到廣泛的標籤資訊,造成資料分類及檢索等服務難以推展。欲使社會性標籤發揮真正的功用,有效的標籤建議(tag suggestion)已成為眾多系統努力的目標之一,亦是本論文的研究動機。近年來,相關文獻已經提出許多搜尋或篩選標籤的方法,我們進一步研究如何將大量標籤加以組織,快速建立一個階層式標籤分群的樹狀結構,以協助系統建立一個漸進式的標籤建議機制。我們所提出的方法將單一標籤的重要性和兩兩標籤的相似性都納入考量,運用分群及離異值偵測的技巧,可根據給定的分群個數計算各層的標籤分群。在方法的效能評估上,我們對網路取得的標籤資料進行實驗,透過一系列自訂的量測指標如群內相似度及群間差異度,分析方法建立的樹狀結構,超過半數以上結果顯示我們的方法在至少兩項指標上達到優良。
Social tagging systems allow people to attach tags to data. Studies and applications surrounding the tags are thus prevalently developed. Due to the diversity of tags and the difference of individual perception, people often fail to use proper terms as tags and systems also fail to collect a wide range of tagging information. Consequently, the tag-based classification and retrieval services cannot be forwarded. To enhance the usage of tags, effective tag suggestion, also the motivation of this study has become one of the goals pursued by many systems. In recent years, many methods for searching or filtering tags have been proposed in the literature. We further studied how to organize the large number of tags to quickly build a tree structure composed of hierarchical tag clustering, which would help the system build a progressive tags suggestion mechanism. Our approach takes in account both the importance of individual tags and the similarity among tags. Using the techniques of clustering and outlier detection, it can compute the tag clustering at each layer according to the given number of clusters. For performance evaluation of our method, we made experiments on the tagging data obtained from the Web. Through a series of self-defined measurement indicators, such as the intra-cluster similarity and the inter-cluster dissimilarity, we analyzed the tree structure built by our method. More than half of the results showed that our approach achieved the degree of excellence under at least two of the indicators.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。