| The e-tax system effectly solves the problems of time-consuming and low efficiency of traditional offline tax processing mode,but it is also affected by the complexity of tax affairs itself,which makes the e-tax system complex and confusing.To solve the problems of e-tax system,an intelligent voice interaction technology was proposed and applied to the voice interactive AR operation guidance system for e-tax.The intelligent voice interaction technology mainly involves technologies like the construction of knowledge base,semantic analysis,intent recognition and so on.Therefore,aiming at the field of tax affairs,researches were mainly conducted on the training of word vectors,the construction of semantic database,the correction of speech recognition errors and the replacement of unregistered words.The specific research content is as follows:Firstly,the method of semantic database construction based on semantic understanding was studied.The classes of words in the tax field are required to be highly accurate and professional,but they are hard to be obtained from the original corpus accurately.Moreover,it will cost time and effort to manually finish the semantic database construction completely.In order to reduce the workload of manually collecting words and classes of words,make the semantic database update autonomously,and therefore classify classes of words correctly,the classes of words and their related core words were collecting manually,while the word embedding method and synonyms expanding method based on semantic similarity threshold were used to expand the synonyms related to each core word.In the end,the automatic construction and expansion of semantic database are achieved.Secondly,the technology of user intent recognition integrated similarity matching was studied.There are problems of speech recognition errors and unregistered words in the process of voice interaction,which lowers the accuracy of user intent recognition.Therefore,a statistical method of user intent frequency merging similarity matching method was proposed to improve the user intent recognition technology.Experiments showed that the method improved the accuracy and proportion of true positive samples when recognised user speech questions which are loosely structured.The method effectively solved the problems of speech recognition errors and unregistered words.Finally,the tax knowledge base,semantic database and tax intelligent voice interaction software system were designed and constructed.The system includes two modules: tax intelligent voice interaction module,management system of tax knowledge base and semantic database.The tax intelligent voice interaction module has the main functions of voice interaction,network communication,parameter configuration,etc,to implement the intelligent voice interactions for tax.In the aspect of management system of tax knowledge base and semantic database,the representation of tax knowledge and semantic knowledge was designed.Based on the representation,the tax knowledge base and semantic database were constructed.In the end,the development of management system of tax knowledge base and semantic database was completed. |
