Towards automatic evaluation of customer-helpdesk dialogues

We attempt to tackle the problem of evaluating textual, multi-round, task-oriented dialogues between the customer and the helpdesk, such as those that take the form of online chats. As an initial step towards automatic evaluation of helpdesk agent systems, we have constructed a test collection comprising 3,700 real Customer-Helpdesk multi-round dialogues by mining Weibo, a major Chinese microblogging media. Each dialogue has been annotated with multiple subjective quality annotations and nugget annotations, where a nugget is a minimal sequence of posts by the same utterer that helps towards problem solving. In addition, 34% of the dialogues have been manually translated into English. We first propose a nugget-based dialogue quality evaluation measure called Utility for Customer and Helpdesk (UCH), where a nugget is a manually identified utterance within a dialogue that helps the Customer advance towards problem solving. In addition, we propose a simple neural network-based approach to predicting the dialogue quality scores from the entire dialogue, which we call Neural Evaluation Machine (NEM). According to our experiments with DCH-1, UCH correlates better with the appropriateness of utterances than with customer satisfaction. In contrast, as NEM leverages natural language expressions within the dialogue, it correlates relatively well with customer satisfaction.