Research Of Natural Dialogue System For Smart Internet

With the rapid progress of machine learning algorithms,artificial intelligence has been showing more excellent than humans in some areas and returning to public view on a stunning way and promoted the new changes in the form of the Internet,that is,mobile Internet to smart Internet.The advantage of smart Internet is to grasp the information more active,analysis more accurate and determined.This includes the perception and cognition of artificial intelligence,the increasing number of smart terminals and sensing devices perceives the world deeper and deeper,cloud and big data analysis are faster and more accurate.Dialogue system is the core technology in the field of artificial intelligence,it will act as a new harmonious human-computer interaction and an integrated cloud service,makes smart terminal get natural perception and powerful knowledge.This kind of user-friendly interaction can be applied to intelligent customer service,personal assistant and other areas and play a role in connecting multiple smart devices.The research has great theory significance and application value.The status and advances of dialogue system and the application of deep learning model in dialogue generation were comprehensively summarized in this paper.First,the type of intelligent dialogue system was elaborated.The module framework of the system was introduced,including the research contents of the modules and key components of the mainstream technology ideas and its research status.Then,the various solutions to language understanding and dialogue generation were systematically analyzed from the perspective of theoretical models,advances and usability.The neural network structures and the construction principle of sequence-to-sequence applied to natural language generation and text classification was analyzed emphatically.(1)Template matching and domain classifier were used to analyze the intent of the text.Aiming at the shortcomings of the traditional document representation of high dimension and high sparse and missing semantics,this paper proposed a method to improve the text representation using the similarity information of the text distributed representation computed by neural network model.The support vector machine algorithm was used to classify the text.The probability distribution of the classification results were analyzed by the visualization effect.We Proposed a classification strategy: according to the confidence of simple classification model,find a threshold by cognitive experience to extract the simple tasks classifying by simple model.The multi-mode matching is called according to the result of the classification,and the template is effectively reduced match the spatial complexity and improve the efficiency of intentional extraction.(2)The natural language generation part generates a reply for the text that can not be extracted from the template by using the depth neural network language model.Based on the sequence-to-sequence neural network structure,the generated language model is implemented.Attention mechanism is used to effectively use the context information to make the generated response more consistent with the context.The optimization objective function adopts the maximum mutual information function to solve the problem of the security answer questions when using the maximum likelihood optimization function;test phase using beam search,effectively select the top scoring the highest,that is,more interesting and more interesting answer.(3)The paper designs the system framework of the natural dialogue platform for the intelligent Internet,and adds the deep learning technology to realize the task-oriented dialogue system on the basis of the template matching,and completes the specific tasks in the vertical field while providing the open-domain humanized dialogue service,and ultimately for third-party developers to provide the standard of the core technology based services to provide a simple and convenient natural language analysis business,with the help of knowledge sharing design concept,to achieve the knowledge of different developers between the mutual assistance and assistance to achieve cross-partners of the cooperation space,and thus achieve a broader application and commercial imagination.