由文本詞彙分析角度探討國際圖書資訊學研究發展趨勢

Abstract

本研究以自動化方式、大規模資料量為基礎，探討2000年以來圖書資訊學領域研究趨勢，共分為兩部分： 第一部分建構一可線上使用之網路平台作為分析系統，統計輸入文件集之詞彙使用及組合情形、擷取高文件頻率詞彙作為停用詞，並以視覺化方式呈現。 第二部分應用前述工具，以近年之圖書資訊學研究論文作為標的資料，觀察並分析其發展情形與研究趨勢，最後對所得結果與前人研究進行比較。 利用文件頻率建立停用詞清單，可擷取出功能詞、一般學術寫作常用詞及information、system等明顯的領域特徵詞，而其他圖書資訊領域特徵詞在擷取門檻降低時亦慢慢出現。 觀察各詞彙之歷時趨勢，呈增加趨勢者主要為社群網絡與資料處理相關詞彙，呈減少趨勢者則為搜尋引擎、web、（數位）圖書館與資訊尋求行為相關詞彙。 2000年以來圖書資訊學領域之研究趨勢轉移，最初從資訊檢索系統、資訊行為及web相關研究開始，再加入搜尋引擎與引文分析，接著在2005至2014年是引文分析主宰的時代，社群網絡則自2010年開始崛起，到現今（2016年止），以社群網絡、社群媒體為研究之主要潮流。 與Chang et al.（2015）之研究使用關鍵字抽樣分析之結果進行比較，發現二者分析出之主要趨勢大同小異，但以本研究之詞彙分析方式較能找出作為研究工具或研究對象之事物，關鍵字分析則對抽象之主題之辨別較為明確。 若以以資訊檢索之TF-IDF模型為核心思想，僅在少數幾篇、甚至單篇文章中大量出現之詞彙將造成干擾。在針對整體趨勢觀察之應用中，DF過低、單篇文章TF過高之詞彙不應有如此高的權重。

This study aims to investigate the research trends in the field of library and information science (LIS) from 2000 to 2016, leveraging an automatic method for processing a large amount of data. This study includes two parts: Part I: Constructing a web application that counts the terms and its combinations in the input data set, extracts the terms with high document frequency to build up stop word lists, and then show the results by visual representations. Part II: We then use the tool constructed in part I, with recent articles in LIS field as input data, observe and analyze the research trends. Finally, results are compared with former research. The stop word lists are constructed in accordance with the document frequency of the terms, then function words, common words in academic writing, and common words in LIS field, such as “information” and “system”, are extracted. As for the trends of terms in the observed period, the terms related to social network and data processing have ascending trends, whereas the terms related to search engine, web, (digital) library and information seeking behavior have descending trends. The trends of popular research topics from 2000 began with information retrieval system, information behavior, and web research, followed by search engine and citation analysis. Citation analysis dominates the LIS field from 2005 to 2014. Social network sprang up around 2010 and became the mainstream of LIS researches nowadays. Comparing with the results of keyword analysis by Chang et al.(2015), we can see that they are basically similar. However, the method in this study can extract more things that are regard as research tools or objects, and the keyword analysis has the advantage of identifying abstract topics. Since we took TF-IDF model as the core idea of weighting the terms, the terms that only occur in few articles—even only one article—might have unproportionate weight, and become noise in the results. The terms, which have high term frequency but only occur in few articles, should not have such high weights in the application of observing the overall trends.