Effective user intent mining with unsupervised word representation models and topic modelling

TL;DR

This paper presents an unsupervised approach combining word representations and topic modeling to effectively mine user intents from customer service conversations, aiding understanding of unlabelled textual data.

Contribution

It introduces a novel unsupervised method that integrates word embeddings and topic modeling for intent detection in customer service chats, involving domain experts for interpretation.

Findings

High accuracy in intent ranking using cosine similarity

Effective interpretation of topics with domain expert involvement

Applicable to large-scale unlabelled customer interaction data

Abstract

Understanding the intent behind chat between customers and customer service agents has become a crucial problem nowadays due to an exponential increase in the use of the Internet by people from different cultures and educational backgrounds. More importantly, the explosion of e-commerce has led to a significant increase in text conversation between customers and agents. In this paper, we propose an approach to data mining the conversation intents behind the textual data. Using the customer service data set, we train unsupervised text representation models, and then develop an intent mapping model which would rank the predefined intents base on cosine similarity between sentences and intents. Topic-modeling techniques are used to define intents and domain experts are also involved to interpret topic modelling results. With this approach, we can get a good understanding of the user intentions behind the unlabelled customer service textual data.