Multi-Turn Dialogue Agent as Sales' Assistant in Telemarketing

Wanting Gao, Xinyi Gao, Yin Tang

Abstract: The traditional telemarketing call centers are being hunched by significant human resources costs related to high workers' workload, as well as even higher turnover-rate, which are also proven to be impossible to be solved by Artificial Intelligence (AI) chatbot alone. In view of this dilemma, applying AI techniques to assist human salespeople in call centers by instantly analyzing the phone call between salesperson and customer, thus recommending a response for the salesperson, becomes a challenging yet applicable means. In this paper, we propose a multiturn dialogue agent based on Deep Attention Matching Network (DAM) with SimBERT, used as a data augmentation tool to efficiently improve the response selection accuracy. Experiments on a real-world telemarketing scenario dataset have confirmed its validity.