| Summary: | 碩士 === 國立嘉義大學 === 資訊工程學系研究所 === 101 === Effectiveness of response generated is important to communication between human and computer, even if the system to produce the appropriate action, if not correct and effective expression, it can also cause communication failure. Therefore, the natural language generation into the generation of responses is an important issue. In this study, in order to generate a response effectively, use a statistical natural language generation. To the idea of semantic concept graph achieve the content decision, and uses Partially Observable Markov Decision Process(POMDP) to train the scorer of semantics and syntax score, and choose the best response to the generated sentence. In the experiment, use POMDP to generate best response sentence, has 85% success turn rate and task average turn 6, better than example-based that has 77.5% success turn rate and task average turn 7.6. It declare that is effective in best response generation using POMDP.
|
|---|
