以視覺化標籤雲輔助圖書標記之使用者研究

Abstract

本研究旨在嘗試以視覺化的角度探討標籤字體大小對於使用者標記行為的影響，並在上述情況下，探討使用者選用的標籤品質是否較佳。 本研究建立兩種標籤雲介面，實驗組為有字體大小變化，控制組為無字體大小變化，採用受試者內設計和拉丁方格法設計實驗。實驗組的標籤先以TF-IDF演算法計算出標籤的權重，再依權重對應到標籤字體大小，隨機排列於標籤雲內；而控制組的標籤沒有經過TF-IDF的計算，以固定的字體隨機排列於標籤雲內。 參與實驗的受試者在系統安排好的介面順序之下，皆會使用兩種介面，從標籤雲內自由選用適合的標籤去標記指定書籍。受試者完成標記任務之後需接受研究者的簡短訪談。另安排兩位專業圖書館員標記相同的書籍，所得結果作為使用者選用的標籤品質評估基準。 本研究分析實驗結果與歸納訪談內容後，得到下述結論。受試者在實驗組與控制組的標記行為有差異。從標籤被選用次數之分佈狀況來看，在實驗組的介面裡，使用者傾向選用字體大的標籤，使得標籤被選用次數分佈較不平均；而在控制組的介面裡，標籤被選用次數分佈較為平均。 受試者在實驗組與控制組中選用的標籤品質有差異。以館員選用的標籤評估受試者在兩種介面裡選用的標籤品質，發現在實驗組的介面裡，受試者選用的標籤與標籤名次接近館員選用的標籤和標籤名次。亦即在有字體大小變化的介面裡，受試者選用的標籤品質較高；而在控制組的介面裡，受試者選用的標籤品質不高。由此判斷採用實驗組介面(有字體大小變化)標記書籍可以提升標籤品質。

The purpose of the study is to examine whether the users’ selection of tags is affected by the font size of the tags in a visual interface, and to examine whether TF-IDF weight based tag size visualization is more conducive to the quality of the tags they choose. A book-tagging experiment was conducted with Latin square design where the participants were asked to use alternately two tag cloud interface to tag sixteen books, one without visualization, the other vary the size to tags based on their TF-IDF weight Thirty-six participants who had basic familiarity of these books took part in the experiment. After tagging the books, they were also asked to rank the importance of the tags they had assigned for each book. Two professional librarians were asked to tag and rank the same set of books, the result of which served as the benchmark for comparing the quality of the tags selected by participants. The results showed a difference in the selected tags between visualization and non-visualization interfaces in terms of tag concentration. Tags selected using the visualization showed significant more uneven distribution as measured by Gini index and entropy. A correlation of tag size and tag selection was found where visualization was present, which indicated the impact of varying tag font size on the participants’ selection. A difference in the quality of the tags selected was also found. Using tags selected by librarians as quality criteria to evaluate the quality of users’ selected tags in two different groups, we found a significantly higher similarity between the tags selected by the librarians and the participants using the visualization interface. Furthermore, the results also showed the effect of visualization was over and above the effect of users’ own knowledge.