Learning End-to-End Goal-Oriented Dialog with Maximal User Task Success and Minimal Human Agent Use

Abstract

Neural end-to-end goal-oriented dialog systems showed promise to reduce the workload of human agents for customer service, as well as reduce wait time for users. However, their inability to handle new user behavior at deployment has limited their usage in real world. In this work, we propose an end-to-end trainable method for neural goal-oriented dialog systems which handles new user behaviors at deployment by transferring the dialog to a human agent intelligently. The proposed method has three goals: 1) maximize user's task success by transferring to human agents, 2) minimize the load on the human agents by transferring to them only when it is essential and 3) learn online from the human agent's responses to reduce human agents load further. We evaluate our proposed method on a modified-bAbI dialog task that simulates the scenario of new user behaviors occurring at test time. Experimental results show that our proposed method is effective in achieving the desired goals.

Learning Low-Resource End-To-End Goal-Oriented Dialog for Fast and Reliable System Deployment

“…Learning to switch. Since there are no actual labels for S to indicate whether it is correct to ask human or not, some previous work (Woodward and Finn, 2016;Rajendran et al, 2019) proposes to use the REINFORCE algorithm (Williams, 1992) for weakly-supervised training, but their reward settings fail to penalize the case when the model asks human while it can give right prediction, which may lead to redundant requests. To consider this effect, we propose a new reward definition here.…”

“…Rajendran et al (2018) learn dialogs with multiple possible answers. Our work is inspired by the work of (Rajendran et al, 2019; propose to solve unseen user behaviors through human-machine teamwork. The research of Chen et al, 2017;Lu et al, 2019) also show the advantages of incorporating the role of human to teach online.…”

“…The total number of dialogs for each task is 3000 (1500/500/1000 for train/dev/test set respectively). More statistics are given in Table 4 A.2 Reward Settings (Woodward and Finn, 2016;Rajendran et al, 2019) defined rewards for S as follows:…”

“…To boost the online performance of dialog systems, there have been some recent work (Rajendran et al, 2019;Lu et al, 2019) on designing end-to-end models in a human-machine joint-teaming manner. For instance, the dialog system in (Rajendran et al, 2019) can identify an ongoing dialog during testing when the system might fail and transfer it to a human agent. But all these methods are trained with sufficient data, which hinders the possibility of rapidly prototyping the models in new domains with restricted resources.…”

A Role-Selected Sharing Network for Joint Machine-Human Chatting Handoff and Service Satisfaction Analysis

“…However, due to the complexity of human conversation, auto-chatbot can hardly meet all users' needs, while its potential failure perceives skepticism. AI-enabled customer service, for instance, may trigger unexpected business losses because of chatbot failures (Radziwill and Benton, 2017;Rajendran et al, 2019). Moreover, for chatbot adoption in sensitive areas, such as healthcare (Chung and Park, 2019) and criminal justice (Wang et al, 2020a), any subtle statistical miscalculation may trigger serious health and legal Figure 1: A snippet of a moderately satisfied customer service dialogue.…”