Research On Automatic Evaluation On Dialogue System Reply Quality

With the rapid development of artificial intelligence,the form of interaction is also changing,man-machine conversation began to appear in all aspects of our lives.The study of the conversation task not only produced a lot of achievements in the academic circle,but also took root and sprouted in our life gradually.However,even with the rapid progress of dialogue technology,how to evaluate the response quality of the dialogue system is still an unsolved problem.In this paper,a new method of replying quality evaluation of dialogue system is proposed,and the present status and problems of replying quality evaluation of dialogue system are discussed,and some new solutions are put forward for some problems.In this paper,a trial experiment was conducted to evaluate the quality of the task-oriented dialogue system and the results were obtained.The data sets are constructed from various sources,so that the model can be migrated and reconstructed to some extent.From the more detailed point of view,this paper explores the method of response quality evaluation of task-type dialogue system.In the aspect of replying to the quality evaluation of open domain Dialogue system,this paper first makes an experiment on the existing objective evaluation index,and designs the data set which can be tested with pertinence to the existing indexes.In addition,the experiment and summary of the characteristics of the open domain Dialogue system are carried out by the machine learning method,besides,this research also realizes the real-time evaluation of the open domain Dialogue system by designing the model of depth learning,and proves that this method is feasible by experiment.According to the data set,the corresponding features are extracted as the characteristics of the classification task to join the experiment.