Multi-Task CRF Model for Predicting Issue Resolution Status in Social Media based Customer Care

In this paper, we present a multitask learning method for predicting the resolution status of the issues expressed in social media conversations among customer-care agents and social media users, along with the nature of dialogues of those conversations. Our method extends beyond social media conversation analysis, and is naturally applicable to general multiple sequence labeling tasks where each example sequence has multiple label sequences. Our method learns multiple models, one model for each task, i.e., issue status prediction task and dialogue act prediction task. Each model computes the joint probability of both label sequences (dialogue act and issues status) given the example sequence, i.e., conversation among customers and agents. Such multiple models are learned simultaneously by facilitating the learning transfer among models through explicit parameter sharing. We experiment the proposed method on real social media conversations dataset collected from Twitter as well as on a publicly available NLP dataset, and show that our method outperforms the state-of-the-art method. In addition, we illustrate how the issue status and dialogue act prediction tasks can be an integral part of socially aware customer care engagement system.