Identification of Food Quality Descriptors in Customer Chat Conversations using Named Entity Recognition

Chatbots are increasingly being used for providing customer support. One of the fundamental challenges for a bot, or for that matter any human agent, is to understand the context of a customer message. Chat conversations are typically associated with agrammatical structure, spelling mistake/variants, informal and slang words, and code-mixing, i.e., the use of words from more than one language. We focused on a use case related to the conversations between customers and agents regarding issues with the quality of food delivered by an online food delivery company. Accurate identification of words that describe the poor quality of food can immensely benefit prompt resolution of the issue and also provide vital feedback to the company and its partner restaurants. This feature can be used in a chatbot to effectively resolve customers' food quality related concerns. This paper presents a named entity recognition (NER) approach to identify the food quality descriptors in a given message. On an internal benchmark dataset, we achieved an F1 score of 0.93 while outperforming classical baseline approaches in NER.