以標籤為基礎之個人化文件推薦

Abstract

資訊科技日漸成熟與網際網路逐漸普及化後，以web2.0為概念發展的網站成為近年趨勢。標籤標記(tagging)已變成web2.0網站熱門技術，讓使用者不僅藉由標籤註記與分類資訊內容，還可以有效且快速管理與組織資訊。另外，衍生標籤建議、查詢語擴展、整合標籤於推薦系統等應用服務。標籤標記透過群體合作方式增加新的詮釋資料。大多數整合標籤於推薦的研究都假定相同標籤文字具有同樣意義，而忽略標籤語意的模糊性。 本研究提出以標籤為基礎之個人化文件推薦。本研究認為可以將一個標籤視為一個興趣，相同標籤在不同使用者間具有不同涵義，因此本研究以使用者標記文章建立專屬的標籤特徵檔來代表使用者之標籤興趣。另外，本研究從標籤持續時間與標籤標記量分析標籤時間效應以區別使用者興趣喜好之程度。實驗結果顯示標籤有助於個人化文件推薦，只考慮使用者標籤興趣對於推薦效果之提升有限，還必須考量標籤時間效應才足夠。

As information technology is maturing and Internet is getting more popular, web2.0 based websites are emerging to promote knowledge sharing. Tagging on web2.0 becomes a hot technique that users not only create tags to annotate and categorize content but also organize and manage information efficiently. Moreover, it launches related research on tag suggestions, query extension, and recommender considering tags. Tagging increases new meta-data through social intelligence. Most recommendation research on tags assumes that the same tag has the same meaning, which ignores the vagueness of the semantics of tags. This study proposes personalized document recommendation based on tags. A user’s tagging is regarded as the user’s interest. The same term (tag) may denote different meanings to users when they apply the same term to tag documents. In this paper, a personalized tag profile representing user’s interest is generated from the user’s tagging of documents. Besides, we analyze the time effect of tags according to the duration of tag and the proportion of documents belonging to the tag. Novel recommendation methods incorporating tag profiles and the time effect of tags are proposed to improve the quality of documents recommendation. Experiment result shows that the proposed approach can effectively improve the quality of document recommendation.