Relevance: The whole history 「相關」歷史回顧


Mizzaro, S. (1997). Relevance: The whole history. Journal of the American Society for Information Science, 48(9), 810–832.

1. Introduction

1-1. Why to write this article?
  1. Relevance is one of the central concepts for documentation, information science and information retrieval. 「相關」很重要
  2. Relevance’s history is very useful for understanding what relevance is. 歷史便於理解「相關」
  3. There is no recent paper that describes in a complete way the history of relevance. 最近缺歷史回顧文獻
  4. This work can be situated at a higher level than the above mentioned surveys 這份回顧要比前人寫得更好
1-2. How to write it?
  1. 範圍限制:documentation, information science, and information retrieval
  2. As objective as possible 盡量客觀
  3. Not only to present the history of relevance, but also to give a framework for understanding the history and the concept.

2. A Framework for Various Kinds of Relevance

2-1. 分類框架

相關 = 第一組要素 + 第二組要素 in 三種組成成分

  1. 第一組要素
    • Document 文件: 使用者找到的實體資料
    • Surrogate 中介資料: 呈現文件的資料,如作者、書目資料、摘要等
    • Information 資訊: 使用者閱讀文件之後接收的資訊
  2. 第二組要素
    • Problem 面臨問題: 使用者所面臨的問題,需要資訊來解決
    • Information need 資訊需求: 使用者內在的需求,可能無法對外表述
    • Request 請求協助: 使用者用自然語言表達資訊需求
    • Query 系統查詢: 使用者用系統語言查詢資料
  3. 三種組成成分
    • Topic 主題: 使用者關注的主題領域,例如特別是在資訊科學或檢索
    • Task 任務: 註明使用者的動作
    • Context 情境: 除了主題跟任務之外其他事務,像是地點、結果評估


2-2. Relevance judgment 相關評判
  1. The kind of relevance judged;
  2. The kind of judge (user and non-user);
  3. What the judge can use (surrogate, document, or information) for expressing his relevance judgment
  4. What the judge can use (query, request, information need, or problem) for expressing his relevance judgment.
  5. The time at which the judgment is expressed.

3. 相關歷史分類說明

  1. 時代區分:大約20年一個間隔
    • Before 1958, 1959-1976, 1977-present (1997)
  2. 研究類別區分
    • Fundations 基礎研究: be defined from different standpoints, using different mathematical instruments and conceptual approaches.
    • Kinds 類型研究
    • Surrogates 中介資料研究: The type of surrogate used can affect relevance judgments
    • Criteria 評鑑研究: 從使用者角度來進行相關評判
    • Dynamics 變動研究: 相關會受到時間影響
    • Expression 表達研究: 什麼方法呈現相關評判結果才是最符合使用者需求
    • Subjectiveness 主觀研究: 不同相關評判間是否一致;不同使用者間的相關評判是否一致

4. 相關的歷史

4-1. Before 1958
  1. 相關資訊的問題剛被發現,但尚未成為聚焦討論的議題
  2. 相關研究: Lotka (1926), Bradford (1934), Zipf (1949), Urquhart (1959), Price (1965)
  3. 相關的正式基礎: Pritchard (1969)的書”bibliometrics”
  4. IR先驅者: Mooers (1950), Perry (1951), Taube (1955) and Gull (1956)
4-2. 1959-1976
  1. 回顧文章: Saracevic (1970~1976), Schamber et al (1990)
  2. Foundations: 奠基未來研究基礎
    • Probabilistic retrieval 機率檢索: Maron and Kuhns (1960)
    • Mathematical logic 數學邏輯: Cooper (1971) and Wilson (1973)
    • The user’s sotck of knowledge 先備知識: Rees (1966) and Wilson (1968)
  3. Surrogates: Quality & Surrogate’s length: 越長品質越好?
  4. Expression: 不同的相關評判適用不同的表達方式
  5. 重要學者: Cuadra & Katter; Rees & Schultz
4-3. 1977-Present (1997)
  1. Foundations
    • User-oriented, cognitive approaches (Schamber et al., 1990; Harter, 1992) 使用者導向,認知取向
    • Defined a logic for IR (Rijsbergen, 1986~1989) 提出更複雜的模型
    • “paradox of relevance” -> “subjective, not measurable” 對立演變成主觀
    • Consider the relevance of a set of documents instead of a single document appear (Gordon & Lenk, 1991)
  2. Kinds
    • Many studies mistake system-relevance for topic-relevance, do not consider all the existing kinds of relevance.
    • Measure the until then retained unmeasurable relevances.
  3. Surrogates
    • 研究里程碑: The “length hypothesis” (Marcus et al., 1978) & Janes (1991)
    • surrogate-based relevance judgments tend to become similar to full-document judgments as the surrogate
  4. Criteria: user defined criteria & document characteristics
  5. Dynamics
    • The existence of a presentation order effect 次序效應
    • the dynamic nature of query, request, information need, and problem justifies at least in part the dynamic nature of relevance 相關變動的最後部分是請求協助、資訊需求與面臨問題
    • cognitive considerations based on learning, mental models, and criteria can explain the variations in relevance judgments 基於學習、心智模型、內心標準的認知思考可作為相關評判的變數
    • the time point at which relevance is measured 可測量的時間點
    • some mathematical models are proposed 數學模型
    • Iterative and interactive IRS: 高互動的檢索系統
  6. Expression
    • magnitude estimation (numeric estimation, line length, and force hand grip) is an effective and reliable method for expressing relevance judgments 數字呈現方式很有效率也很可靠
    • it is preferable to both category rating scales and dichotomous judgments. 分類度量與分歧判斷
  7. Subjectiveness: the conditions (features of the judges, but also criteria and dynamics) that lead to inconsistency.

Discussion 相關研究成長趨勢

  1. 從1960s年代到最近10年間研究持續增加
  2. 分類數量最多:foundations, kinds
  3. 分類foundations, criteria, dynamics, expression穩定成長


Relevance is a necessary part of understanding human information behavior. The field should be encouraged by commonalities across perspectives, not discouraged by disagreements. Relevance presents a frustrating, provocative, rich, and—undeniably—relevant area of inquiry. (Schamber , 1994)


  • 非常有架構的review文章,清楚好閱讀!