PubMed快速搜尋與MeSH輔助搜尋之比較研究

Abstract

關鍵詞搜尋又稱為快速搜尋，在PubMed當中很常被使用的搜尋方式。醫學主題詞表（Medical Subject Heading，簡稱MeSH）是另一種搜尋選擇。使用者在檢索過程中，對於快速搜尋與MeSH輔助搜尋之檢索結果好壞，通常不會直接回饋任何意見給PubMed，因此評估兩種搜尋方式之檢索表現很重要。再者，過去研究很少著重於使用者觀點。而先前研究亦指出主題熟悉度在檢索過程中有其重要性，因此，本研究利用真實使用者與真實檢索主題，比較PubMed當中快速搜尋與MeSH輔助搜尋，同時探討主題熟悉度在過程中之影響。 研究方法為實驗法，研究對象為生物醫學與生命科學相關系所研究生，共招募32位受測者，並請受測者事先前準備一個熟悉與一個不熟悉的主題。實驗採用拉丁方格設計與受試者內設計，控制主題熟悉度（熟悉與不熟悉）與兩種搜尋方式（快速搜尋與MeSH輔助搜尋），將受測者隨機分至4個組別。實驗一開始蒐集受測者背景資料，接著請受測者依序填寫實驗前問卷、搜尋任務問卷及實驗後問卷。 研究結果顯示：（一）快速搜尋與MeSH輔助搜尋在平均精確率、累計增益值（Cumulative Gain，簡稱CG）、折扣增益值（Discounted Cumulative Gain，簡稱DCG）及正規化折扣增益值（Normalized DCG，簡稱NDCG）的表現沒有明顯差異。受測者對快速搜尋與MeSH輔助搜尋的滿意度亦無顯著差異。但受測者使用MeSH輔助搜尋過程受到比較多挫折。（二）快速搜尋與MeSH輔助搜尋對於最初與最終查詢語句相似度沒有顯著影響。（三）主題熟悉度對於最初與最終查詢語句適用性沒有顯著影響。（四）主題熟悉度對於檢索結果相關程度有顯著影響，越不熟悉的主題越可以找到越相關的資料，反之越熟悉的主題找到的資料越不相關。建議未來研究可探討如何幫助使用者檢索不熟悉的主題。

Keyword search, also called quick search, is a common way used to retrieve relevant data in PubMed. Medical Subject Heading (MeSH) is another option in PubMed. Since users usually don’t provide any feedback to PubMed, it might be very difficult to know whether the performance of quick search or advanced search (MeSH) is good enough. Therefore, it’s important to evaluate the performance of both quick search and advanced search. In the past, researchers seem to rarely focus on the viewpoint of users. Also, topic familiarity plays an important role during the process of information retrieval. As a result, we would like to compare the PubMed quick search and advanced search with real users and real search requests as well as exploring the impact of familiarity. A total of 32 participants in the bio-medical areas will be recruited, who, prior to the experiment, would be asked to prepared two search requests of their own, one familiar, the other, unfamiliar. A 2x2 Latin square factorial design was adopted where topic familiarity (high vs. low) and search interfaces (Quick vs. Advanced search) served as two within-subject factors. The background information is collected, following by the pre-search questionnaire, search task, post-search questionnaire, and post-experimental questionnaire to collect all the data of search performance and search behavior. The result shows that there are no obvious differences between the performance of quick search and advanced search in terms of mean average precision, cumulative gain, discounted cumulative gain (DCG) and normalized DCG. There are no significant differences between participants’ satisfaction toward quick search and advanced search either. However, participants feel more frustrated when using advanced search. Quick search and advanced search have no significant impact on the similarity of initial and final query. Topic familiarity has no significant impact on goodness of initial and final query. However, topic familiarity was found to have a significant impact on users’ relevance judgment. Search results were found to be significantly more relevant, as judged by the users, for unfamiliar topics than familiar ones. Future research can be done to investigate how to assist users who are not familiar with their topics.