Recently, the development of businesses, made possible the extent of their customer support services in social media, the place where customers have direct access to companies and can easily request their support. Customer support is one of the most challenging and critical areas of a business, because the quality of this service can hugely impact the customers loyalty or the opposite, their attrition. The challenging part of Customer support service comes across with addressing and solving in a timely manner a huge number of requests. Here comes the opportunity to make use of the most recent advances in neural machine translation, in order to automate the customer support service. Therefore, the value of the Chatbot systems comes from their ability to automate conversations, ensure consistent customer support experience and significantly reduce the case volume for customer service representatives. For this purpose, we leverage a conversational chatbot system designed to automatically answer questions to customers in Twitter platform. In this paper we discuss the sequence-to-sequence (seq2seq) model with attention, based on a dataset with over one million Twitter interactions between customers and companies' representatives from 20 major brands and we measure the model's performance using the BLEU score. Based on the results, we then analyze and identify the factors that impact the model's performance.

Chatbot; Customer support; social media; deep learning, machine learning