Inquiry Classification in a Speech-Oriented Guidance System Using Discriminative Learning

概要

論文の詳細を見る
The issue of providing appropriate responses for users' inquiries in a speechoriented guidance system can be addressed with an example-based response generation method, which compares a user's utterance with example questions stored in a question and answer database (QADB), and the answer corresponding to the most similar example question is selected. An advantage of this method is that the system can be easily expanded to cover other classes of inquiries. However, questions from different classes could present high similarity, which could hinder the selection of an appropriate answer. The selection of an appropriate answer is basically a multi-class classification problem, and to deal with it, we study the use of the bag of words (BOW) kernel to represent the features of users' inquiries and using support vector machine (SVM) for the classification, implementing a one-vs-all multi-class classification approach. Preliminary experimental results using transcribed data show that the proposed method improves the classification performance of users' inquiries, compared to a baseline method.
一般社団法人情報処理学会の論文
2009-07-10