Research On Memory Dialogue System Technology

With the continuous popularization of smart terminals such as smart phones and smart speakers in life,people's dependence on these devices has gradually increased,and smart terminals have also played an increasingly important role.At the same time,people are also facing the problem of data explosion.It is estimated that by 2025,humans will produce 463 billion bytes of data every day.The goal-oriented dialogue system that combines retrieval and generation models mainly consists of three parts: problem retrieval module,semantic matching module and reply generation module.We have built a data set in the field of health counseling to train the models that will be used in these modules.At the same time,we demonstrate the effectiveness and superiority of the algorithms in these three modules through a series of experiments.By combining the retrieval model with the generation model,the overall performance of the dialogue system has been greatly improved.The rapid growth of data has caused the existing ways of retrieving information to be cumbersome and inefficient.A person uses smart terminals and the Internet every day to realize various tasks such as ticket reservations and takeaway reservations,and uses memos to record various types of affairs,but the information storage and query methods of each application and software are different,and the different interactive interfaces are Dazzling,many data are listed together to make people confused.It is very difficult to find the records and even the desired results in a short time.This topic studies a simple,convenient,and efficient method that can memorize and query structured and unstructured text in the form of natural language.It can be applied to smart voice assistants of smart terminals to improve user experience.The memory dialogue system mainly includes three parts: semantic understanding module,precise query module and query optimization module.We constructed a data set of memory dialogue tasks to train the models used in these modules.At the same time,we designed a series of experiments to prove the effectiveness and superiority of the algorithms applied by each module in the system,which can well solve the memory dialogue task.