Electric Merchant Customer Service Automatic Q & A System

“Made in China 2025” was proposed to accelerate the development of intelligent manufacturing equipment and products. Service robots and smart homes sre important parts and highly sought for in industry. Service robots will get policy support and design in the wake of industrial robots. After China’s e-commerce market developed with many years and now the scale is large, users’ amount is big. The online users’ demands of service quality are increasing not only in terms of quality of the goods that users experienced. The cost of customer service continues to increase, the turnover rate is high and it is difficulty to recruit professional customer service members. E-commerce customer service robot is a new artificial intelligence form to serve businesses and customers efficiently.The key step of the communicating process between Customer service members and users as well as E-commerce customer service robots and users is to indentify the intentions of the customers. By using the intention identification businesses can service clients effectively, improve customer satisfaction continuously and cultivate loyalty customers to get profits. The article combed through the Question Answering System(E-commerce customer service robot) research, introduced the basement knowledge of question-and-answer system from development, system type, structure. What’s more it introduced the technology in question-and-answer system involving keywords extraction and calculation of words similar degrees and got a deep understanding of the BP neural network algorithm. Firstly, based on the knowledge above the sentence processing of E-commerce customer service robot used the segmentation technology of Chinese Academy of Sciences and it built professional cosmetics dictionaries involving goods, beauty, skin care and other aspects to improve the segmentation accuracy. According to the features of online shopping language, the stoplist was built with the statistical analysis of communication data. Then, the article selected three aspects from semantics, itself and position as features to extract keywords by using BP neural network model. Secondly, the intention network was built with users’ caring about aspects in goods and service. The network service for the system late answering with each customer had a portraits descripted by the intention network. Finally, according to human forgotten law proposed by ebbinghaus and communicating time features between users and customer service robot, the “Single Stage Model” and “Multi-Stage Model” were built to describe the strength of commodity intention. The degree of similarity between extraction keywords and products were calculated by Word2 vec tools and the result of users’ commodity intention strength could be achieved.It improved question-answering system accuracy to some extent. The result could be used for robot to provide personalized recommending and other service.