Library and Information Science Society Home → English page

三田図書館・情報学会誌論文(論文ID LIS042023)

The Implications of the Non-Keyword Information on the User's Judgment of Relevance in Information Retrieval
No.42, p.23-48

Most of the present information retrieval systems presuppose subject keywords as an access point as well as the criterion for relevance judgment. In other words, IR systems mainly determine relevant documents by keywords. However, users may look at other document characteristics in addition to the subject when judging relevance of the documents. This means that criteria of judging relevance between the system and end-users may be different. In order to overcome the discrepancy, it is necessary to make clear what kinds of non-keyword information are necessary for user’s relevance judgments and how they are identified and used for relevance judgment.

Firstly this paper investigated whether and how each of the 17 characteristics that documents for 3 graduated students and 3 information specialists. usually have was used for the judgment of relevance by IR users. A survey was conducted. The result of the survey showed that users judged relevance using various information from the 17 characteristics. Non-keyword information which contains events, research methods and novelty of the idea, as well as the length of original documents and of abstracts were determined as predominant information for judging relevance. Non-keyword information has the same importance as the information from keywords in retrieving documents.

Secondly, the author discussed the applicability of those kinds of information to enhance retrieval performance. Such kind of information as the length of documents and publication dates will be easily implemented into IR systems since they are parts of bibliographic data. Others such as novelty are more difficult to use, because the value of those kinds of information are dependent on individual user.

Finally, this paper suggests the possibility to introduce user models into a IR system since user models can represent such characteristics of each user as novelty. They will be helpful for the system to deal with non-keyword information.

本文PDF (3,177K)