▽TOP▽活動実績▽２００２年▽

(43)　Improvement of Automatic Generation Method of the Reply Sentence
　　　　by Inductive Learning Using Common Portions on E-mail
　　　　　Proceedings of FIRST INTERNATIONAL CONFERENCE ON INFORMATION TECHNOLOGY &
　　　　　APPLICATIONS (ICITA 2002), pp.231-236, 2002-11

　We performed informational exchange and understanding of an intention using E-mail in daily life. However, the generation of reply sentences to E-mail takes a lot of time and labors. In order to solve this problem, a demand of the system which generates automatically the reply sentences to the received mail is increasing. In this paper, we propose an automatic generation method of the reply sentences by Inductive Learning using common portions on e-mail. This system learns the way how to reply to the received mail. When a similar received mail inputted to the system, the system generates several reply sentences from the results of the learning. This system generates the reply sentences to a similar received mail for reducing a cost of labor and time. This system uses a received mail as an input and output reply sentences. We evaluated this system by two kinds of value. One is recall evaluated by a rate of a received mail which is applied generation rules to a received mail inputted to the system. The another is precision evaluated by a rate of reply sentences which is generated to reply sentences without any errors. For the experiment, we use 100 pairs of received mail and reply sentences. In results of this experiment, the more received mail and reply sentence to it increase, the better a recall is. Finally, recall became about 80 %. This is due to increase of generation rule. However, precision remains about 50 %. And generated reply sentences are not perfect. We consider that this is due to the way of acquiring generation rules and the way of a selecting generation rule.