Research On User Tag Generation And Product Recommendation Technology For WeChat Customer Service System

With the development of Internet,customer-focused service purpose was important for lots of enterprises.Customer service system enables to establish good relationship between customer and firm and enhance the customer satisfaction and adhesion to the enterprise.The traditional customer service based on telephone and SMS has the disadvantages of high human cost,single interactive mode and high operation cost.Subsequently,the enterprises cost was reduced and the interaction was enhanced effectively because of the communication software and other customer service system,but enterprises and users eagerly look forward to intelligent,personalized,universal customer service system appearing with the problems of information explosion,meetting user's complex requirements and other issues.We Chat official account is widely used in product promotion,marketing activities and announcements with the convenience and coverage of We Chat.Customer service based on We Chat official account was preference by many companies and organizations,so the service question and answer log has accumulated a great deal of information that can be used to mining valuable data which is significant for enterprises to understand user's requirements,promote products and enhance customer satisfaction.In the paper,an intelligent customer service was designed and developed combining both advantages of We Chat official account and customer service system.what's more,the method of generating user profile label and recommending product by label information under the framework of service system is proposed.The user's profile label was generated by extraction keywords and the method of recommending with the help of the generated label was proposed.The method is elaborated as follows.It is doubtful whether the traditional method of keyword extraction focused on web pages,scientific literature and text documents can be adapt to the dialogues with the characteristics of short text,loose construction.Compared to the traditional keyword extraction for dialogues that are mostly based on the term-frequently and co-occurrence relation,ignoring the semantic and topic information,an automatic user tag generation method for dialogues which are based on the semantic and topic wasproposed.The semantic weight and the frequency weight of words were taken into account to generate user's label.Experimental results show that the proposed method is superior to the method based on the term-frequently and co-occurrence relation.Compared to the traditional method of collaborative filtering which exists cold-start,sparse,and lacking information problem,an recommending method dealing with cold-start problem with the tag information which concretely based on the sparse Autoencoder neural network was proposed.By the way of taking tag-vector as the vector of user and resource,then the hidden feature of user vector through the training model of Autoencoder was abstracted,finally the Top-N items based on the similarity of user and items were recommended.Experimental results show that the label can improve the recommended effect.Finally,the intelligent customer service system based on We Chat official account was designed and implemented.At the same time,the main research content was integrated into the customer service system and the function performance of the method was showed.