When using an open-ended response collected by social surveys as basic data in statistical processing, after-coding must be conducted to classify it into one of the pre-defined codes (classes). After-coding is a heavy burden for a coder and can possibly be misclassified when the open-ended response is ambiguous or insufficient. Hence, we need to collect a response containing sufficient information for classification to avoid such a situation. However, requesting this task to a respondent and a survey taker is not easy. In this paper, we propose a new system in which a survey taker brings a computer with knowledge that can classify an open-ended response into a valid code of all codes. The computer then asks additional question to the respondent if it perceives that a response does not contain sufficient information for classification, and subsequently extracts the remaining information effectively. This decision is determined when a confidence level of the computer to a result, which is estimated by the scores accompanied by the results, is lower. After collecting information, the computer reclassifies a new open-ended response into a valid code. The proposed system has the additional advantage of being able to supply basic data immediately. We are constructing a new system for occupational coding which is a representative after-coding. The system has not been completed yet, but shows efficacy by a small experiment. In future work, we will completely implement the system and evaluate it by respondents, survey takers, and coders. Moreover, we will expand the system for generalization.
View full abstract