Analysis of User Feedback Cost for Document Similarity Judgment

This paper investigates the behavior of users judging the similarity of documents from the viewpoint of user feedback cost, in particular judgment time and accuracy. An experiment is conducted, in which 21 test participants were asked to judge the similarity of documents. As the clue for the judgment, 3 types of information: original text, snippet, and term, are mutually provided. The judgment accuracy and judgment time are analyzed using analysis of variance (ANOVA) and multiple comparison tests to examine the difference of snippet, term and text. The result shows that displaying term is the best in terms of time cost, whereas the judgment accuracy when a snippet is provided is improved with experience. The obtained result will contribute to the design of interfaces that can minimize the user’s feedback cost.