使用者查詢擴展行為研究：以MeSH為詞彙來源的建議詞彙工具為例

Abstract

由於傳統的資訊檢索評估以回收率及精確率做為系統評估之指標，忽略了使用者在資訊檢索中所扮演的角色，因此近來產生了互動式資訊檢索評估。此外，由於使用者在檢索系統中多只會利用簡單關鍵字進行查詢，因此會得到大量的檢索結果而無法管理。因此本研究設計了一針對PubMed醫學資料庫的詞彙建議系統MAP(Multiple Access to PubMed)，目的在於輔助使用者取得其所需的資訊。 本研究以實驗法及問卷法進行研究，同時設計詞彙建議系統並採用使用者的真實資訊需求做為檢索任務，以生物科學及醫學系所研究生為研究對象，總計44名受試者。並就使用者查詢行為及查詢擴展行為進行分析，同時比較不同介面過濾檢索結果之功能以及說明實驗產生之次序效應。 研究結果顯示：(一)主題熟悉度會影響使用者查詢行為，不熟悉主題任務其查詢相似度會低於熟悉主題任務。(二)不同介面會影響使用者查詢行為，MAP系統能夠提供使用者大量的建議相關詞彙，因此使用者可從中利用之，而在PubMed介面中使用者查詢詞彙的刺激有限。(三)不同主題熟悉度與不同介面對使用者查詢行為產生之交互作用，MAP介面在使用者較不熟悉檢索主題時可以發揮較大的作用(四)不同主題熟悉度未影響使用者查詢擴展行為，使用者在不同主題熟悉度下，其瀏覽及最終選用詞彙類別之比例並未呈現明顯之變化。(五)不同介面會影響使用者查詢擴展行為，原因在於MAP能夠提供全面性的領域內的相關詞彙，而PubMed則僅能提供使用者相關的檢索結果，造成使用者在詞彙選擇上的差異。(七)MAP介面能夠輔助使用者取得在大量檢索結果中排序較低的相關文件。

It has been argued that traditional information retrieval evaluation is ill-equipped to address the need to validate the efficacy of today’s highly interactive systems, which require users’ active participation to be effective. To answer the challenge of interactive information retrieval evaluation, a novel methodology was applied to test the effectiveness of MAP (Multiple Access to PubMed), a metadata-guided search interface for PubMed bibliographic search. The most distinctive aspect of our methodology is to use real users searching for real search requests on real system, instead of using assigned tasks common in traditional IR evaluation. To control the impact of individual search requests on search performance, a repeated measure design was adopted where users' search request served as its own control of variance. Comparisons of information behaviors between MAP and the regular PubMed interface were made. The purpose of the study is to examine whether interfaces and topic familiarity might interfere user’s search and term selecting behaviors. Some major findings are as follows. 1. The participants were found to input more diverse terms and make more submissions were made using MAP. There were also differences in term category selection between the two interfaces. 2. Topic familiarity was also shown to influence users' query expansion behaviors. Differences were found between the categories from which terms were selected between MAP and PubMed. 3. MAP was shown to help users to uncover relevance document ranked much lower in the original dataset. 4. MAP is more effective in search situations where the users were less familiar with the topics.