Research Of Conversation Model For Assistance Of Custom Service

With the development of the Internet economy,electronic business platforms develop fast both on the scale and volume.This trend of developing requires higher efficiency and better quality for online customer service.Therefore,how to assist custom service staffs improve service efficiency and quality with computer technology is a problem worthy of study.On this basis,we focused on two kinds of customer service support technology:query service on knowledge base and custom service response recommendation service,researched and explored the related dialogue models.We designed a query service on knowledge base,which provides quick access of professional knowledge during working and help custom service staff achieve high-quality service.The service accepts natural language question as input,then extracts the keywords and attributes by using the AIML template matching technique,at last returns the corresponding knowledge.The matching method suffers from a problem of keyword matching failure.We analyzed and altered the semantic-based synonym matching algorithm,proposed a multi-round iterative synonym matching algorithm,which improves the performance of synonyms detection.To improve the efficiency of the customer service,we use deep conversation model for custom service response recommendation.We tried to solve this problem in two ways:retrieve-based deep conversation model and generation-based conversation model.On the retrieve-based model,we designed a conversation model with context modeling,which has better performance on response recommendation than non-contextual conversation model.Later we use intention information improved this model,and obtained some performance gains.On the generation-based model,we implemented a Seq2Seq model,use it as a baseline model.Then we designed and implemented our contextual generation-based conversation model.Through the experimental analysis,we found that our model performs better than baseline model in generation quality.