Text Intention Understanding In Human Computer Interaction

Speech interaction is the most natural, convenient and direct way in human computer interaction. And natural language intention understanding technology has great importance in speech interaction. Certain research work on user intention understanding is done in the scene of home service robot and intelligent car speech navigation. Firstly, an improved method combining maximum entropy and keyword matching(ME+MATCH) is proposed to solve the core module-intention understanding module in the system. This improved method tries to combine semantic relevance information and statistic information. Secondly, intentions included in these two scenes are defined as intention tree. Corpora on home service robot and intelligent car speech navigation are constructed by collecting corpus in these two scenes. And the entire corpus are segmented and tagged by corresponding intention tags. Finally, in the experiment, a multi-fold cross validation is performed on home service robot corpora and intelligent car speech navigation corpora. To measure the effect of intention understanding method, macro-average and micro-average are introduced. Comparison with k-nearest neighbor(KNN), support vector machine(SVM) and maximumentropy(ME) method, it is shown that ME+MATCH method is better than KNN and ME on both two corpora. And on intelligent car speech navigation corpora, ME+MATCH is better than SVM.