| With the rapid development of artificial intelligence,service robot has already entered all area of daily life,playing a substantial role.Meanwhile the auto speech recognition and natural language processing are increasingly mature,which make it possible to apply spoken dialogue system to service robot.Focusing on the usage scenario of service robot,we studied the spoken dialogue system for service robot.A preprocessing module,a question and answer module based on the frequently asked questions set and a dialogue management module have been deeply studied.In the preprocessing module,various methods of word segmentation have been studied,including word segmentation based on word dictionary,comprehension and statistic.Then the features of those methods have been compared.A statistic and dictionary combined method is applied to segment the auto speech recognition result and a experiment has been carried out.Then stop words are removed based a stop words list.Finally the disadventage of the traditional method of key words expansion based on a semantic resource has been studied,which shows that the semantic resource covers very few words to apply it to spoken language system.The method that those have larger TF-IDF values in the words list are expanded by Word2 Vec is proposed.In the question and answer module,a data structure is designed to make the candidates questions extraction more effective and the percentage of extraction has been studied.Then the model of similarity based on TF-IDF has been studied and the method that the similarity between the target and the candidate question is calculated based on the Word2 Vec has been proposed.Finally,two models of similarity are combined and a experiment is carried out to determine the weight,which improves the accuracy of question matching.In the dialogue management module based on slots feature,the TF-IDF method of topic extraction is improved by applying a moving window to the dialogue history.In addition,the cooling simulation is applied to monitor the topic heat.At the end of this paper,a web app is made to demonstrate the whole spoken dialogue system. |
