摘要 |
<p>According to the present invention, chatting corpus data, which comprises user utterance data and system response data, is received as learning data; learning is conducted to generate index information between the user utterance data and the system response data; when the user utterance data is a compound sentence, a mutual information amount in connection with the system response data is calculated with regard to each of simple sentences that constitute the compound sentence; one of the simple sentences is selected on the basis of the mutual information amount; learning data is generated using data regarding the selected simple sentence and the system response data; and re-learning is conducted, thereby providing system response data which is reliable in connection with compound sentences as well. In addition, according to the present invention, when user utterance data is input, system response data corresponding to the user utterance data is detected and output; or, when no system response data is detected and when the user utterance data is a compound sentence, system response data corresponding to each of simple sentences constituting the compound sentence is detected; a mutual information amount between the detected system response data and the simple sentences is calculated; and one of pieces of system response data is selected and output on the basis of the mutual information amount, thereby providing system response data which is reliable in connection with compound sentences as well.</p> |
申请人 |
SOGANG UNIVERSITY RESEARCH FOUNDATION;KNU-IINDUSTRY COOPERATION FOUNDATION |
发明人 |
SEO, JUNG YUN;KOO, MYOUNG-WAN;KANG, SANG WOO;KIM, HARK SOO;CHOI, MAENG SIK;SONG, YEONG KIL;JEON, WON PYO |